1
研 究 論 文】
UDG :
624
.
042.
7 620.
1Journal
efStructural
andConstruction
Engineering
日本建 築 学会 構 造 系 論 文報 告集 (TTansactions
ofAIJ
)No
.
358
.
December
,
1985
第358
号・
昭 和 60 年 12 月PART
2
:
RELATIONSHIP
BETWEEN
.
INTEGRATION
TIME
INTERVAL
AND
ACCURACY
OF
DISPLACEMENT
,
VELOCITY
,
AND
ACCELERATION
RESPONSES
IN
PSEUDO
DYNAMIC
TESTING
(
Stability
and
accuracy
behavior
of
pseudo
dynamic
response
)
by
MASAYOSHI
NAKASHIMA
*,
Member
Qf
A
.
1
.
J
.
1,
lntroduction
Through
the
discussion
in
the
companion
paper
”,
重t
wasfound
that
the
stability
limit
of
the
central
defference
method
(
CDM
)
usedin
pseudo
dynamic
(
PSD
)
test
is
ωム 置=2
regardiess ofthe
value ofthe
viscousdamping
coefficient,
andthe
erroris
character 亘zed as a combined natura 且period
distortion
and numericaldamping,
This
statement on
the
erlor conditionis
specifiedfQr
the
displacement
response of systems subjectedto
free
v三bration
.
This
paper
is
an extension ofRef.
1
,
andthe
objectiveis
to
studythe
reiationshipbetween
the
selectedintegration
巨me
interval
andthe
obtaineddisplacement
, velocity , and acce 正
eration
respon 寧esof
PSD
tested
systems sustaillinggeneral
externalfQrces
.
The
paper
alsQincludes
some numeri ρal resultsthat
.
support
the
findings
in
this
Study
.
Atthe
last
pa
τt
of
this
paper
,
comments
are
give
【l
for
the
selebtion
of
the
time
interval
in
PSD
test
.
2、
Accuracy
Under
Forced
Vibration
First
,
let
us supPose an undamped system subjectedto
a normqlizedtime
varyingforce
r
(
の
:tC
(
t
)
+ω 2・
コじ
= r
(
t
)
・
……・
一 ……・
…・
……・
………・
・
…・
………・
…・
…・
・
……・
…………
.
一・
…………・
…一 ・
一
(
1
)
If
the
equationis
solvedby
CDM
, 出e recursiveform
is
asgiven
in
Eq
β
QfRef.
1as
.
:.
n
Xn
=
An
・
Xo
十Σ
An
一
‘・
L
・
r[
4t
(
‘−
1
)
]
…・
…………・
・
∵…・
一………・
…・
…・
・
・
・
……・
…一 一 ……・
……・
(
3
,
Ref
.
1
>
‘=1 ・・… ・一
[
(
2
−
tO:At
:)
−
1
1
0
]
・ ・一
[
剰
’
And ,
symbols aredefined
正nRef.
1
.
The
fi
πstterr
ゆ oIthe 匸ight
hand
side o{Eq
.
30f
Ref
.
1represents
the
free
vibration response,
whereasthe
secondter
皿 showsthe
response causedby
the
externaHorce,
That
is
,
t
}1efirst
and secondterms
respectively correspondto
出e complementary andparticular
solutiops ofthe
different
童al equation ofEq
.
1
.
Let
uslook
further
into
one ofthose
forced
vibrationterms
:.
4n
−
1・
五・
r(
0
)
.
This
term
indicates
the
response atthe
n−
th
time
step causedby
the
externalforce
atthe
initial
step.
The
ampiifica しion
matrixis
decQmpQsed
as shownin
Eq .
50f
Ref.
1to
:.
An
−
1=
φ
’
Jn−
1’
ip
−
1’
「
…’
………’
…’
……’
…’
……”…鹽
…’
’
”……’
……’
…’
鹽
……’
………「
…「
’
…’
……’
…
(
5
,
Ref
.
1
)
In
CDM
,
these
terms
canbe
writteh(
symbols aredefined
in
Ref.
1
>
as :・・一・
一
[
[
c
°s
漏
△‘ち
‘’
s
血th
(
n
− 11At
]
[
。 。s
。
。.
1
、
。,∴
.
。、。。,
。−
1
、
。,、
]
・
一[
、
。。S
。 。 、∴
.
。剛
(
。_∴
.
。、。硼
]
…・
……・
…・
…一 ・
一 …………・
一・
一 ……・
…
…
・
』[
(
sin そ万∠it
−
i
・
COS 五〕△t
)
i
2
・
sin
石 △t
2
・
sin
石 △ 諺(
sin
石1
△t
十i。
COS
石}△の
一
ご
2・
.
si
皿 五iA
言2
・
sin 石1
△t
]
零Production
Departfnent
,
BuHding
Research
Instit
ロte,
Man
且scrip け ecelvedAugust
23,
1984NII-Electronic Library Service
Then:
A"-'-L-r(O)=:ip.Jn-i.ip'i・L・r(e)=sin
-wnAt
sin
th4
t
sinth(n-1}At・AtZ・r(o)・・・t・・・・・・・・・....,....---・(3)
sin
thA
t
From
this
expression, one canfind
that
the
initial
externalforce
r(O)
produces
attime
t
the
displacement
response of:
Ad.:::
{ltlk!I}liYin.nWrr.nAAtt
At:・
r(o)
=Si:-Win-A.t
iAAtt,'ir4(O)
・・・・・・・・・-・・・・・・・・・・・・・・・・-・・・・-・・・・・-・・・・・・...,...,.,...,...(
4
)
This
expressioncan
be
consideredas
adiscrete
form
of
the
Duharnel
Integral,
The
correspondingtrue
Duhamel
Integral
js
written as:dx(t)=Sina)t・dT'r(T),"-,""H".,"""v-",".,."H"h-".-"".,H"h"""v.""",."""."".""""""H.(s)
wIf
the
externalforce
r(t)
is
assumedto
be
constantduring
this
time
interval,
the
true
response attimetis
given
as:
sin a,t・At・r(O)
Ax(t)=
--・--・・-・・・-・・・--・・・-・・・・・・・・・・・・・・・・-・-・-・・・-・・・・--・・・-・・・-・・・・・・・・・・・・・・--・・-・・・-・-・・・-・・・・・(6)
tu
This
assumption
produces
an errorin
the
solution,because
the
externalforce
mostlikely
varies evenduring
this
time
interval.
Reminded
that,
in
anydifference
equations,the
externalforce
is
picked
outdiscretely
only at everytime
interval
andapproximated
to
be
constant
in
CDM
during
this
interval
<see
Eq.
4),
weshould
carefully
setect
the
time
interval
so
that
the
discretization
error ofthe
externalforce
would notimpair
the
solution.This
is
another criterien we should notoverlook
when selectingthe
time
interval.
The
effect
of
the
discTetization
error
onthe
numerical
response
is
associated withthe
nature ofthe
externalforce
and was studied,for
example,in
Refs,
2
and-3.
As
the
discussion
in
this
studyfocuses
onthe
errors characteristicto
the
CDM
algorithm,the
effect ofthe
discretization
error
is
notconsidered
;
it
is
assumed
tacitly
that
the
external
force
changeduring
the
integration
time
interval
is
minimal.
With
this
reservation,the
true
Duhamel
lntegral
is
asgiven
in
Eq.
6.
Comparing
Eq.
4
with
Eq.
6,
one
canfind
that
the
naturaicircular
frequency
is
changed
from
wte
M,
ancl
the
ratioof
this
changeis
identical
to
the
period
distortion
ratio
in
the
numerical
free
vibration
response(Fig.
1
of
Ref,
1).
This
change
in
the
frequency
means
that
the
response obtainedby
CDM
is
notthe
response efthe
analyzed systembut
the
response of a systemhaving
the
natural
frequency
of"tu.
This
erroris
nowdesignated
the
period
distortion.
Comparion
between
Eqs.4
and6
also showsthat
the
numerical arnp],itudeis
notthe
sarne asthe
true
amplitude, andthe
ratio ofthe
nurnericalto
true
amplitudes
is
:
Aau=
toAtlsindiAt=11
1'wiAt!14
''''''''''''''''・・-・-・・・・・・・-'・''''・-・・・・-・・・・・・・-・・・・・・・・・・・・・・・--・・・-・・・-・・・-・・・-・・<
7
)
Figure
1
(h=
O)
showsA..
vs. wAt
relationshipfor
ctizSt's
between
O
and:uMtReSlcToVAaRee:ispan/actt'LAmdie:t
2,
indicating
that
the
discrepancy
between
the
true
and numericalthe
free
vibration response,there
is
no suchdiscrepancy
in
amplitude oyerthe
entire stabllity range,Characteristics
ofthe
response error attime
t
introduced
by
the
general
ao
t5
to
ao
Fig.1
Displaeement
Introduced
in
-36-to
Amplitude
CDM
20wAtDistortion
sinthAt・
1-h!ca2AtZ
whereas
the
correspondingtrue
Duhamel
Integral
external
force
from
time
O
to
(t-At)
is
identical
withthose
generated
by
the
external
force
from
,o
to
4t,
because
this
total
erroris
the
sum of・theindividual
eTrorsgenerated
at eachtime
interval
as shownin
Eq.3
ofReL
1,
and,furthermore,
those
individual
errors
are characterizedby
the
sameperiod
di'stortion
(Fig.
1
ofRef.
1}
and amplitudelncrease
(Fig.
1),
Discussion
onthe
accuracy condition offorced
vibration responseis
now extendedto
general
damped
systems.Following
the
procedure
expressed
in
Eqs,2
and3,
one can obtain:exp(-hwnAt)-sinbnAt・Att・r(O)
Ad
H""""M""-"--(8}
Eq.
6)
is
:
exp(-htot)・sin
toot・At・r(O)"H"H-H..HLH""H"."HHHHHH"".."H-""",,"--・---・・-<9)
Ax(t)=
abp
By
comparingEq,9
withEq.8,
it
is
found
that
the
discrepa.ncy
between
the
true
ancl numerical responsesis
characterizedas
a
combination
of1)
period
distortion
withthe
ratio shownin
Fig.4
of
Ref.1,
2)
damping
ratiochange
withthe
rati6 shownin
Fig.6
ofRef.
1,
,and
3)
amplitudeincrease
of:Add=
a,DAt,1{sin
diAt・
1-h2to!At')--・・-・・・・・・・・・-・・・・-・・・・-・・・・・・・・-・・・・・・・・・・・・・-・・・・・・・・・・・・-・・・-・・・・・・・--・・・・・・・・・・<lo)
Figure
1
showsA.,
vs.- toAt relationshipfor
varioush's,
indicating
that
the
numerical amplitudeincreases
drastically
as wAtapproachesits
'stability
limit.
'
'
3.
Accuracy
of
Numericat
Velocity
and
Acceleratien
Resp6nses
-Here,
discussion
is
extended
to
the
velocity and acceleration responses withthe
application ofCDM.
First,
investigation
is
given
into
the
velocity and aeceleration responsesof
an undamped system subjectedto
free
vibration.Referring
to
Eq.I5
of
Re{.1,
din=2kt'(-dn-!+dn+O=21it'[-lci'cosiln-1)At+ct'sinEJ(n-1)Atl+lci・costh(n+1)At
'
+
c!・sin
E](n+1)A
tl]=lci
-(-sin
thnA
t}+
c2・cos
thnA
tl-sin
a,At・・-・・-・・・・・・・-・・・・・・・・・・・・・・・・・・・・・・・・・(11)
On
the
otherhand,
the
true
velocityis:
.
'
'
X(t)=lct・(-sin
a,t}+c2・cos catFw--・----・-・---・--・・--・--・---・---・----・(12)The
numerical velocity relativeto
the
true
velocityis
therefore
specifiedby
the
period
distortion
withthe
ratiogiven
in
Fig.1
ofRef.1and
amplitudedecrease
defined
by:
'
'
Avc=sinaAtlw=
1-wiAt!14-・・・・・・・・・・・・・・・・・・・・・・・・・・-・・・・・-・・・・・・・・・・・・・-・・・・・・・・・・L・-・・・・・・・・・1・・・・・・・・・・・-・・・-・・・(13)
Analogously,
the
numerical
acceleration responseis
:
dn=
Alt,
(dn-i-2'dn+dn+t)=
Alt,'(ci'cos
thnAt+c2'siri
thnA
t)'(cos
thAt:1}
,
-1
=At2
'{Ci'COS
dinA
t+
c2-sin
-a,nA
t)'(a,!A
ti)=-
a,i'(ci・costhnAt+
cz・.sin
-tunA
t)・・・・---/-・・・・・・・・・・04)
Here,
there
is
no, amplitucledistortion
in
the
numerical acceleration response.The
discrepancy
between
the
true
ancl
numerical acceleration responsesis
characterized ofilyby
the
period
distortion
ddfined
by
Fig.1
ofRef.1.,
Now,
let
us considerthe
velocityand
acceleration
responsesof
a
system
sustaining
forced
vibration.Referring
to
Eg.
8,
・
'
.
Adi2%t(-'Aan-i+Adin.i}=exp{-ETtonAt)・(cos-tunAt・SinAllAt・cosh{7idiAt)
-sin
bnA
t-cos
mA
t-Sin
h2htWA
t)
1'
sin
mAf.t!'i
rlOh-)r.,At,
''''''''''':'r:''-'''''I'''''''''''''"'''''(i5)
whereas,the
true
velocity(with
the
conditionthat
the
external
force
remains constantduring eachtime
interval)
is
:
Ath(t)
±exp{-ha)t)・la,D・c6sa,Dt-liw・sina,DtFAtiII(O)・・・・・・・・・-・・--・・・・・・・・・・・・・・・・・・・・・i・・・・・・・・・/・・・・・・・rL-・・a6)
On
the
pther
hand,
Adi.
canbe
r6written as:'
'
Aa.=
exp(-7iTwAt)・R,・eoscdinAt+nO・
sin
-.AeiZ'irl02,.,At,
'''-・-''''"'''J'''"''''H'''''''''''"'''''''''"(i7)
'
in
which:
R,=[(
SinAllAt
・cosh<7iEAt)]:+,(cos
diAt.sinh2htwAt)
}!]ii2
t
T
tt
and:cosp,=SinAl:At・cosh(71T.4t)IRi
・
・
Analogously
Ath(t)
ctinbe
written as:・
`
Ath{t)=exp(-ha,t}・R,・cos(a,Dt+B2)・At'r(O}・・・-・・・・・-・・・・・'・・-・・・・・・・・・・・・・-・・・・・・・--・・・・・・・・・・・・・・・・-・-・-i・・-・:・Os)
tun
'
in
which:
R,=l(ha))'+{a,p)21ii',
andcos
fiz==
wDIRzAs
a
result,the
discrepancy
between
the
true
and numericalyelocity
responsesis
given
as acombination ofO
period
distortion
withthe
ratio shown'inFig.
4
ofRef.
1,
2)
damping
ratiochange
with'the
ratioNII-Electronic Library Service
20
t
AMPLITLME
RA
T70carp)
oNumerieatlExaetVetocity
h=O,5h=O.4h=O,5h=O.2h--O.ih=O.O
as
ao
AmeLtTtME
RA
"OCAap)
t
ao
to
2o
wAtFig.2
Velocity
Amplitude
Distortion
Introduced
in
CDM
phase
lag
by
(t?,-P,),
and
4)
amplitudeincrease
of:t
wAt
Avp=(Ri'wo'At)l(R2'sinMAt・
1-htw2Ati)
Fig3
Acceleratien
Amplitude
Distortion
・・・・-・'・-'''-H'-H'-'-'''-'''''''''(19)
Intreduced
in
CDM
FIgure
2
shows
A.'.
vs. aAAt
relationshipfor
varioush's,
This
figure
indicates
that
there
is
little
amplitudedistortion
appearingin
the
numerical yelocity response exceptin
the
vicinity ofthe
stabilitylimit,
particularly
whenthe
damping
is
small,Following
the
proce.dure
to
obtainthe
numerical velocity, one can computefor
the
nttmericalacceleration
:
'
Aa.==exp(-hmnAt)・[2・ICOSbAt'CAOSt,h(hWAt-i)l・sin-onAt
u2-SinAVAt・SinhiihttoAt>・cos-wnAt]'sinthAl!.t'ir-<Oh),.,,At,
''-''"''''''"''H''-'''-"''-''H'''(20)
The
correspondingtrue
accelerationis
:
At・r(O>
AX(
t)
==exp(-
hwt)'Kh!
to'- a}B)'sin wpt-2hw'
wD'cos wotl'----・----・---・(21)
top
Both
Eqs.
20
and21
canbe
transformed
to
expressions similarto
Eqs.
17
andlg,
andthe
discrepancy
between
the
true
and numerical acceleration responses
is
found
to
be
characterized asa
combination
of1)
period
distortion
(Fig,
4
ofRef.1),
2)
darnping
ratio change(Fig.6
ofRef,1),
3)
phase
lag,
and4)
amplitudeincrease.
The
ratio efthe
numericai
to
true
amplitudes,A.,,
is
shownin
Fig.3,
whichindicates
that
the
numericai ampritudeis
signifi ¢antlylarger
than
the
true
amplitudefor
large
toAt]s.4.
Numerical
Exprimentation
To
demonstrate
the
validityof
the
analytic
findings
obtained
in
this
paper
and
Ref,
1,
numerical
experimentation
was carried out.
The
model usedin
this
analysis was atwo
DOF
spring-mass system(Table1).
The
vibrational characteristics ofthe
system arelisted
alsoin
Table
1.
The
variables usedin
the
analysis wereO
the
integration
tirne
interval,
2)
the
vibration mode, and3)
the
viscousdamping
ratio.The
system was analyzedfor
its
free
vibrational condition withthe
initial
deflection
of eitherthe
first
or second mode.Two
types
of viscousdamping
wereintroduced
in
the
analysis;
one was nodamping,
the
otherRayleigh
damping
withs
and90
percent
ofdamping
ratiofor
the
first
and secend modes respectively.
The
large
damping
ratio,O.
9,
for
the
second mode was chosenintentionally
in
orderto
demonstrate
that
the
stabilitylimit
remains unchanged no matterhow
large
the
viscousdamping
is.
A
total
of seventeenanalyses
weremade.
Designation
of
individual
analysesand
employed
vaiiablesare
listed
in
Table2.
This
table
also shows whether or notthe
responsediverged.
Some
response curves obtainedfrom
the
Table1
2DOF
Medel
andits
Vibrational
Characteri$tics
NumericatlExactAeceterotion
5
h=O.Oh=O.Ih=e.5h--O.4h=O.5
':
ifh.O.Aop.1-(eJAt74'
O.O
ZO
2.0
Vibration-lade ffIFStoryMasstkg]SteryStiffness {kNlm} AodeCircularFrequency(Hl]lstStorylndSterylstZnd2.2M2,ZM159.2156.5
lst2nd16.4Z134!.70GSO.5147121.01.0-O.622S61
-38-Table2Variables
andStabilityConditlon ofNumericalResponse
Deslgnat+en-ede' OapinglstMode2ndlt)deTimeInterveltsec.1tilht
tStabilityDisp.SAcce.
NvmericellTrueVelecityNumericallTrue RUN-1lsto o O.oo1O.D427Ves l.aooa 1,OOoo
RUU-2lste e O.O05O.214Yes
1,OOOO
O.9S92RUN-3lsto o O.04 IJIYes 1.ooeo O.9"5
RUN-4lsto
o
e.o4s1.92Tes 1.ogooO.9293
RUN-5lsto o e.os 2.13ko
-r
-i
RUN-6lsto o O.04S81.999Ves
1.00co
O.9Z32
RUN-7lsto o O.0469Z.co3No
"
T-RUN-S2ndo o O.oo5O.214Tes 1.0ooO O.9943
RUN-9ZndD O O.04 IJIYes 1.0000 O.52oo
RUH-10Zndo o o.es 2.13"o
--
+T
RUN-112ndo o O.04681.999Tes 1.0coO O.03es
RUN-122nae o O.046SZoo1Ne
tt
"-RUN-ISlstO.05O.9O.oo5O.214Tes--
-t
RUN-l4lsto.eso.go.ca IJIYes--
--RuN-15lstO.05O.9D.05 !.13'No-t
--RUN-ISlsto.ose.go,e46e1.999res--
-+
RUN-11lstO.OSO.9O,0469.2.oo3to--
--' CrStiealt・irmerwmm
interval=
Z14Z.706SE
pm"
lfO+sp.Hext
g,22ntn
e.e46s3 sec.ZFPisp,-i-
i
]fi.eml・ls.oottime]
ta)RUN-6
fiptEIa
lf O/sp. Mo:
- 1i3
nyn
c
t1stHode. Hobelping, lnterval
=
O.Og6g)
ErlE(g
IFOisD. He"
-
9.2tmup
rfM/sp.N-i
.
7a.)m1IS.OO
IDO.OO
(b)
RUpt-1{lst-ede,koDtmpi"g Iptervil±
QIS!IE[M
IFD,sp, Heia
ZO.Omm
F
2FDrisp.M".
15.0imF
--
---c[)
O,O-S9] ?FOisp.Mej. /5.0
mm
F
---]oo.ooRUN-16 tlst M)dt,, Oerplng, lntcrv"1
;e.04ee]
td)RVN-17.{lst Hode. Peop1"s, IntrrvtlsO.046gl
Fig.4
Nume[icaL
Free
Vibration
Response
by
Central
Difference
Method
anhlyses are shown
in
Fig.4.
As
the
modelis
atwo
DOF
system,the
criticaltime
interval
(stability'limit)
is
determined
by
Eq.
13
of
Ref.
1
withthe
second
modenatural
circular
frequency.
Then
it
is
coinputed as2142.
7065=
O,
04683
sec.It
is・apparent
trom
Table
2
that
the
response ofthe
systemdiverged
once
the
integratiori
time
interval
exceeded
the
criticalinterval.
This
statementis
correct regardless ofthe
viscousdamping
value orthe
mode of vibration.Table
2'
alsoindicates
that
the
amplitude of undamped n'umericaldisplacement
and acceleration responses remains unchanged regardless ofthe
integration
time
interval,
Whereas
the
amplitude of undaniped velocity responsesdecreases
withthe
increase
ofthe
time
interval
in
accordance
with.
Eq.13.
5.
Comments
on
Time
lnterval
Selection
in
PSD
Test
Ofie
who conducts aPSD
test
may wishto
referto
Figs.4
and
6
of
Ref.
1
and
Figs,
1,
2,
and
3
of
this
paper
tQ
estimate
the
accuracy(error)
ofhis
PSD
test
results.Although
Leech
et'al,
")suggestedthe
useof
ll6
of
the
criticaltime
interval
(tuAt=ll3)
orless
ofthe
time
interval
to
ensure an accurate solution,there
is
no absolute reasonto
sup'port
this
value,If,
ih
alightly
damped
system(h
less
than
o.
O,
a4
percent
period
distortion
and
10
percent
amplitude
increase
is
still considered acc'eptablefor
orie"sPSD
test,
he
onLy needsto
settiis
time
interval
at
aiz!t==o.8.
Such
conditionis
attractiveparti
¢ularly when wedeal
withPSD
test
ofMDOF
systems.A
simple
exampleis
given
below.
A
sixDOF
spring-mass system withthe
vibrati6nalcharacteristics
listed
in
Table
3
was analyzedfor
its
elastic earthquake response with no viscousdamping
assumed.The
N-S
component
of
the
acceleration record obtained atthe
Tohoku
University
during
the
1978
Miyagiken-Oki
earthquake was used asthe
input
ground
motion, withthe
maximum acceleration adjustedto
258.
51
gal.
For
the
direct
intergration,
CDM
was employed.The
integration
lime
interval
was selected asO.
Ol
sec. ,Which
inclicates
that
to,Atis
e.
I
and aA,AtO.
8,
NII-Electronic Library Service
c.
Table36DOF