[5H
n
:-fi
!]
UDC:624,042.7:620.1
Journal
ofStructural
andCenstruction
Engineenng
Eptfles#ftMxxXinYvat."se
{TransactionsofAIJ>No.
353,Julv,
1985
rg
353'ny・waAt
60fi7A
PART
1
:
RELATIONSHIP
BETWEEN
INTEGRATION
TIME
INTERVAL
AND
RESPONSE
STABILITY
IN
PSEUDO
DYNAMIC
TESTING
(Stability
and
accuracy
behavior
of
pseudo
dynamic
response)
by
MASAYOSHI
NAKASHIMA',
Member
of
A.LJ.
1.
Introduction
It
was1974
whenTakan
ashi etal.
Mof
the
Institute
of
Industrial
Science,
the
University
of
Tokyo,
first
reperted ona simulation
technique
of earthquake responseof
building
structuresby
means
ef
a combined experimentand
numerical ana]ysis,In
this
technique,
the
dynamic
responseof
a
structure
is
computed
by
solving
the
equations
of motionof
a
discrete
system
representingthe
structureby
diTect
integration
with respectto
the
time
domaln.
Because
of
the
versatile capability ofthis
technique
to
directly
simulate
the
earthqllake
responsebehavior
of
structural
systems,this
technique
has
attracted
many
research
bodies
of
variousfiegds
in
both
Japan
and
overseas.t2'S]
In
their
studies,this
simulation method wasdesignated
in
various ways such asthe
computer actuator on-line system,[i)the
on-linehybrid
experiment,E2) orthe
pseudo
dynamic
testing
method.(S'S]Throughout
this
paper,
the
method willbe
referredto
asthe
pseuclo
dynarnic
(PSD)
test
method.Since
the
outset
of
the
PSD
test
method,the
central
difference
methed(CDM)
has
been
usedexclusively
for
the
computation of equations of motion,
CDM
is
one of explicitintegration
methods,Since
noiterative
procedure
is
requiredeven
for
the
inelastic
response computation,this
integration
methodhas
been
found
suitablefor
PSD
test.
This
method,however,
is
conditionally stabLe unlike rnostimplicit
integration
methods.Therefore,
the
time
interval
used
for
the
computation
of
equations
of motion shouldbe
smalterthan
acertain value, calledthe
criticalinterval,
in
orderto
maintainthe
nttmerical stability ofthe
solutien.This
stability condition,however,
is
a condition whichguarantees
that
the
solutiondoes
notdiverge,
but
not a condition whichpromises
the
accurate solution.In
the
numerical analysis,the
solutionbecomes
more accurate asthe
integration
time
interval
approaches zeio,but
neverthe
same asthe
true
analytic solution aslong
asthe
interval
is
finite.
It
is
now clearthat
one should carefully selectthe
integration
time
interval
in
referenceto
both
the
stability and accuracy conditions ofCDM
when employingthe
PSD
'
'
test,
As
an alternative ofCDM,
the
explicitNewmark
method(fi=O)
is
often employedin
PSD
Test,
Two
comments areto
be
given
with respectto
the
relative effectivenessin
PSD
applicationbetween
the
two
methods.First,
{t
shouldbe
notedthat
the
explicitNewmark
methodis
identical
withCDM
in
the
computation algorithm[E).Second,
in
the
realimplementatien
ofPSD
test,
errors areinevitably
involved
because
ofthe
finite
accuracies ofthe
displacement
andforce
sensors as well asthe
actuatorcontrol
capacity.
According
to
the
study
madeby
Shing
andMahin[7),
the
explicitNewmark
methodis
better
in
the
error controlthan
CDM,
The
study,however,
indicated
that
the
two
methods areidentical
even
in
terms
ofthe
error controlif
the
computeddisplacement
is
usedin
place
ofthe
measureddisplacement
for
solvingthe
equations of motion, and alse commentedthat
better
fesponse resutts canbe
obtainedwith
the
use ofthe
computeddisplacement.
In
this
respect, use ofCDM
combined withthe
computeddisplacement
canbe
said no worsethan
the
explicitNewmark
method.There
is
a classical studydone
by
Leech
et al.+S),in
whichthe
stability conditionof
CDM
wasderived
mathematically, and a suggestion was made on
the
tirne
interval
insuring
the
accurate solution.As
a matterof
fact,
many
later
works onPSD
test
referredto
the
fesults
ofthis
Leech
study.In
this
study,however,
the
stabilitycondition
wasderived
onlyfor
undamped systems(no
velecityterm
included
in
the
equations of motion).'
*
Production
Department
Bu{lding
Research
lnstitute
Ministry
ofConstruction
ManuscFipt
receivedAugust
23,
19S4-29-Furthermore,
their
suggestion onthe
accurate solutien,i.
e.that
the
time
interval
be
smallerthan
one sixth(1!6)
ofthe
critical
time
interval,
was rather vague;
they
neitherindicated
how
much error couldbe
expecteclif
this
interval
wasused
nor
stated
how
much
the
solution wouldbe
aggrayatedif,
say,
an
interval
oftwice
ofthis
interval
was used.Consider.ing
the
incompleteness
of
information
onthe
stability
and
accuracy ofCDM
applied
to
PSD
test,
the
objective
of
this
study
is
to
investigate
in
detail
the
stability
and
accuracy
characteristicsof
CDM
in
terms
ofthe
integration
time
interval
andpropose
guidelines
for
the
selection ofthe
time
interval
in
PSD
test.
This
paper(Part
1)
focuses
its
investlgation
onthe
fundamental
stability conditlonof
CDM
andthe
accuracy characteristics ofdisplacement
response whenCDM
is
employed
for
system$ subjectedto
free
vibration.In
the
companion
paper
(Part
2),
the
accllracy
characteristicsof
displacernent,
velocity,ancl
acceleration
responses
of
systernssubje
¢ted
to
general
external
forces
are
examined
in
detail,
andpractical
comments onthe
selection ofthe
time
interval
in
one's
PSD
test
are
provided,
This
part
alsoincludes
numerical experimentationin
whichthe
yalidity ofthe
rpajoT
findings
ofthis
studyis
demonstrated,
Throughout
this
study,investigation
is
given
onlyfor
linear
singledegree
offreedorn
(SDOF)
systerns since response of multidegrees
offreedom
(MDOF)
systems canbe
treated
as an assembly ofSDOF
systems
if
the
concept of modal analysisis
employed.When
aMDOF
systemhas
an aTbitraryform
of viscousdarnping
matrix,the
modaldecoupling
may notbe
achieved with respectto
the
realdomain.
The
investigation
here,
howeyer,
is
made assumingthat
the
viscousdamping
matrixhas
aform
suchthat
the
system wouldbe
decoupled
with respectto
the
realdomain
as commonly madein
both
numerical anatysis andprevious
PSD
tests.
2.
Stabiljty
Condition
ofNumerical
Methods
Fundamental
stability condition of numericalintegration
methodsis
reviewedhere.
The
equation of motion of alinear
elasticSDOF
systemis
given
as1
m'X(t)+c・X(t)+h・x(t)=r(t}---・-・---・--・---・-・-・・---・---(1)
where
m,c,h.x(t),
and r(t),
arethe
mass, viscousdamping,
stiffness,displacement,
and externalforce
respectively.In
the
nume[icat computation oftransient
response ofthe
system,the
displacements
arediscretjzed
with respectto
the
time
domain,
andthe
differential
equation
is
approximatedto
adifference
equation.Stability
conditionof
diffe[ence
equations
has
been
studiedin
variouspapers.Cg']3)
The
essence of stabiEity conditionis
that
the
errorintroduced
into
the
approximate solutionby
aparticular
difference
rnethod Temains uniforrnlybouncled
asn-co
with n asthe
solutiontime
step.This
conditioncan
also
be
intepreted
such
that
the
solution
does
notgrow
wlthoutbound
for
arbitraryinitial
conditions.t'2)
For
clear understanding ofthis
condition,it
it
helpful
to
writethe
difference
equation
in
[ecursiveform
as:
Xh+i=A・Xh+L・r(nAt)・---・・-・---・-・----・-・---・・-・・-・---・-・・--・・・----・---・(2}
in
whichAt
is
the
integration
time
interval.
By
repreated use ofEg,2,
we can obtain: nX.=An.X,+ZAnui・bT[At(i-1)]-・-・・---・-・--・-・・・--・・-・-・・-・・・-・---・・-・・-・-・・・・--・-(3)
ilr-
1In
the
above equation,X..i
andX.
are vectors storingthe
solutionquantities
such asthe
displacement
and velocityat
time
t==(n+1)At
and
t=:nAt,
Matrix
A
is
calledthe
amplification
matrix, andL
the
load
opeiatoi.Since
the
stabilityis
evaluatedby
examiningthe
behavior
ofthe
numerical solutionfor
anyinitial
conditions, considerationis
given
regardingthat
no externalforce
is
specified.Then,
Eq.3
is
reducedto:
Xn=A"'Xo"''-''''''"'"'''"''"''''''''''''""m-hm''''-'''"-"''"''"-''-''-'"''m-'''-'''-'m-'--'(4)
This
equationindicates
that
the
boundness
ofthe
solutionis
afunction
ofthe
amplification matrixA,
Matrix
A"
canbe
decomposed
to:
An=
¢
・Jn-e']--・-・----・・--t・・・--・-・---・・--・--・---・----・--・-・-・・・--・---・---・(s)
Where
di
is
the
matrix of eigenvectors ofA,
andJ=diag{A,)
withX,
as
the
i-th
eigen
vaiue ofA,
This
matrixis
known
as
the
JDrdan
form
ofA.
Introducing
the
spectral radills ofA
as:p(A)=maxlAA--H--"H"---・・---・---・---・・-・-・・・--・・--・・・--・--・-・・--(6)
one
can
find
J"
andA",
(ancl
accordinglythe
solution ofEq,4)
is
botinded
for
n-oo
if
p(A)gl.
This
is
the
condition
of
stability.
3.
Stability
of
CDM
Stability
ofCDM
for
the
equation of motion of aIinear
undamped system was studiedby
Leech
et al.`S'Their
-30-derivation,
however,did
notfollow
the
procedure
described
in
the
last
section.Here,
stability ofCDM
is
examinedby
use ofthe
recursiveform
ofEq,4.
This
approach enables usto
evaluate
not onlythe
stabilitybut
alsothe
numericalaccuracy
ofthe
solution.
First,
considerationis
given
to
the
undamped system.If
no externalforce
is
applied(free
vibration),the
undampedequation
of motion ef aSDOF
system
is
given
as:
to(t)+a,!・x(t)=O・-・・-・----・-・-・-・-・・・--・・---・-・---・・-・-・・・----・-・・・-・-・-・・---(7)
where w'
is
h!m.
This
equationis
now convertedto
the
difference
form
by
CDM.
Considering
:
dn+i-2dn+dn-i
dn=
At:
"H"H"'H'HHH-"-HH-'"'""H-"""H"H"""H"H'"''-''''H"H"'r"""'H"'--"(8)
in
whichd.
andd.
arethe
displacement
and acceleration attime
t=nAt
specifiedby
CDM,
andinserting
this
expressionin
Eq.7,
wecan
obtain:
dn+i+{w2At'-2)'dn+dn.i=:O-'H"''H"H"H"HHm"'H'""-"-'''HH'''"'"'H'-'"--'"'"'H"H'""'"(9)
This
equationis
now rearrangedto
recursiveform
as:Iddn.'il=[(2Ltoi2At')
L31idd.".,l-
-・・・
---・
-
-・
・
・
-・
-
・
・
---
--・
・
・ao}
The
eigen values ofthe
amplification matrixis
1
"+(w2AtZ-2}・A+1=O・・・-・・---・・---・・--・・-・-・---・--・---・・---・・---・--(11)
and'
x,,,..(2-to2'At2)
±i.p
m
・ -・ ・ ・ ・ --・- ・
・ -・- ・'----""''- ' -"(12)
Then,
we can ebtainfor
the
condition of stability:p(A)Sl:wAtS2・・・・・・・・・・・---・---・・・・・・・・・・・・-・・・・・・・・・・・・・・・・・・・・・''・・・・・・・・・・・・・・・・・・・・・・・・・・・・・・・・・・-・・・・・・・・・・・・・・・-・・・・・・・・・・・・(13)
This
conditionis
identical
with what waspioven
by
Leech
et al.[S)Identity
ofEq.11
withEq,9
exceptXs
instead
of
d.'s
leads
usto
the
following
expressionfoT
a)At'ssmalLer
than
2:
d.=ci'A?+c:'A:"HHHHH'"''"''-"''"''H'-'-'''""""-"--''-'"'"-"-"-""''-''"''"'"HH""''--(14)
Where
c,
andc,
arethe
integration
coefficientsto
be
determined
from
initial
conditions,Suppose
that
A,,,=A
±i・B
d.=
exp(-h-w
tn}'(c,'ces
th
tn+
c:'sin-tu
tn)
'''''''''''''''''''''''''''''''''''''''''"''''''''''''''''''''''''''-'''''''''(15)
Where
:
e=tarp'i(BIA)
h=-ln(A'+B2)!2e
o=efAt
This
is
the
analytic expression representingthe
numerical solution.In
the
above equation,h
is
defined
asthe
coefficient of nurnericaldamping
introduced
into
the
difference
approximation, and to1th asthe
numericalperiod
distoition.
One
can understancl withoutdifficulty
those
terminologies,
remindedthat
the
numerical solutien(Eq,15)
approximates
the
undampedfree
vibration of aSDOF
system<Eq.
7).
Sybstituting
the
eigen va}ues ofEq.12
into
Eq.15
leacls
'to
:
h=O---・・-・・・・・-・---・・-・・-・--・・--・・・-・-・-・・-・・・・・-・---・・・-・・---・-・・---・--・-・・・・・・-・-・・・06>
This
equation rneansthat
the
numerical approximationdoes
not contain anyinherent
numericaldamping.
In
Fig,1,
wl-tuis
plottecl
against
toAt
ifi
the
stability
range.It
is
explainedfTom
this
figure
that
the
period
computedby
CDM
is
shorterthan
the
true
period,
whilethe
difference
between
the
two
decreases
withthe
decrease
ofthe
time
interval
andfinally
becomes
zeroin
the
1Lmit
aJAt=O.
This
period
contractionis
characterized asthe
approximationintroducecl
in
CDM.
The
period
contracLionbehavior
ofthe
explicitNewmark
method"dentical
withCDM)
was alsoinvestigated
in
Ref.14.
Figures
2(a)
to
(c)
showthe
time
history
ofEq.7
computedby
CDM
for
various values of toAt.The
employed
initial
conditions are:
x(O)==1,O&th(O)=O.O・・・-・-・・---・--・-・---・-・・---・-・・・-・・---・--・--・--・-・-(17)
In
those
figures,
also
plotted
are
dAs(in
Eq.15)
having
the
corresponding
a)At.
It
is
seenthat
the
numerical responselinearly
traces
the
points
on
the
d.curve
(Eq,15)
at every specifiedtime
interval
At,
The
figures
alsoindicate
that
no-31-to7ir
(tsetw)
O,
otstFig.1
Period
Distortion
in
CDM
(if
h=O)
entTuoE
te)tstotnl.s
LO
asooe.
e-1.o・
-CDma
.
dnCb)utatat,2
1,o
ci・.,
・・・・・
,・,
-Le
1,oo
-J,o
Fig.2
Numerica]
F[ee
Vibration
Response
(CDM
vs.d.
in
Eq,15)
In
a region:toAt)2Vi=ilT,both
eigen continuouslydecreases
as aJAt
increases
andA2=
changein
A's
is
illustrated
in
Fig,3.
Ieads
to
instability,
The
stability criteriafor
Condition
li
tuAt>2
Unstable
Condition
2
:
blAt<2'twStable
(two
Condition
3:2・Vi-
iTswAtg2
Stable
{two
o'=.t'
tt''t.t-..ttttthe-T;-tttt.-t]
.tL..t
1O
t.O
2
r-'ALlr'r'IJIDnutteReofstt-I:
1,1c
2=,.k:S.:;vtlii,filll 1Lt/ ..tt'F-//T tr.---・-ll-h.asaososO.7ca
/''1 ' 1.t wtou't.H--'...t.///1'
Ll e-O.S1!
/Tr
1・1---411 os'a2 ' L'..Tt.tT;L..1'r.4-i.t.tL・'
Fig.3
Real
Xs
in
Eq,20
(h!t=O)
values
-1.0
at aiZ!t
When
coAt
is
morethan
2
the
1.o
O.5
amplitude
decay
occurs
in
the
response(h=O),
andthe
numericalperiod
becomes
shorter as toAtincreases.
4.
Stability
of
Darriped
System
In
the
foregoing
section,CDM
wasdiscussed
for
its
stabilitywhen
applied
to
the
undampedSDOF
system.
The
next
stepis
to
extend
the
discussion
to
general
damped
systems.gously
to
the
undampedsystem,
stabilityis
examinedin
terms
ofthe
free
vibration condition,Then,
the
equation of motion ofa
damped
SDOF
systemis:
X(t)+2hurte(t)+to!・x(t)==O・・・・・・・・・・・・・・・・・・t・-・・・・・(18)
where
h
is
the
viscousdamping
coefficient.Following
the
procedures
given
in
Eqs.7
to
10,
we obtainfor
the
eigen values ofthe
amplification matrix:
wrAtt-2
1-htuAt
=O--・・---・--(19)
x!+
・A+
1+hwAt
1+htuAt
and1
Xi・Z=2(1+htoAt)・i(2-weAt:)
±i・tuAt
・
4(1-h')-w'At:l・・・・・・・・・・・・・・・・・・・・・・・・・・・・・・・・(20)
Unlike
the
undamped system,here
the
conditionthat
the
eigen values are a
pair
of complex conjugatesis
notidentical
to
the
condition
ofthe
spectrum radiusbeing
less
than
unity,Note
that,
in
the
undamped system,the
eigen values are apair
of complex conjugatesas
long
asthe
integration
time
interval
is
smallerthan
the
criticalinterval.
For
the
valuesto
be
complex
in
the
damped
system:aAAt<2'Vi-=iii'""-"""-'''・--"'・・--・"--・・-(21)
are
real.
After
seme algebra ontoEq.20,
w'efind
that
Ai
=2, while
A,
remainsless
than
unityin
toAt=2.
This
,
then
the
absolute value ofX,
is
larger
than
unity,
which
damped
system can nowbe
summarized as:
cornplexReals)conjugates)
o
to
(a)
if"s
O.5F[g.4
t,o
71T
O.5
WAt2.0
0
S.O
Cb)
tf
M)O,5
Period
Distortion
in
CDM
WAS2.0Condition
2
is
nowinvestigated
in
fu
rtherdetail.
The
numerical solution canbe
writtenin
aform
ofEq.15.
In
Fig.4,
the
ratioof
numericalto
exact
naturalperiods
(
Tl
T)
ofthe
damped
systemis
plotted
against wAt
for
varioush]s.
For
h's
smallerthan
aboutO.7
(Fig,4(a)),
the
numerical naturalperiod
is
always smallerthan
the
true
period.
The
n,umerical naturalperiod
increases
constantly astuAt
decreases,
reachingthe
true
period
in
the
limit
of wAt=O.
Foi
h's・greater
than
aboutO,7
(Fig.4{b)),
onthe
otherhand,
the
numerical naturalperiod
is
infinite
Ondicating
nooscillation>
in
the
limit
:
tuAt=2ff
anddecreases
precipitously
to
alevel
ofthe
true
period
for
the
$maller
values of a)At.This
criticalpoint
ofh
is
given
as:h=VO15'#o.7o7・-・-・・-・・-r・---・・--・・---・・・・・-・・-・-・・・--・・----・・・-・・-・-・・-・・・-・・・・・---・-・(22)
which
is
derived
from
the
conditionthat
A,,,=O(equal
root) atthe
limit
:
wAt=2Vi=7ii'.
This
criticalpoint
(Eg.22)
can
be
explained
asfollows
:
When
the
eigen values ofEg.20
provide
an equal root,the
analytic
expression
ofthe
solutiond.
is
given
as
followstLS)
:
d.=ci・Ar+c2・n・Ar・--・-・-・-・・-・・-・---・-・・-・・---・・--・---・---・---・-(23)
For
h2<O.5.
the
equal Tootis
negative.Since
d.
is
computed asthe
power
of
n,d.changes
its
sign as eachtime
step
is
advanced,
In
this
situation,
the
numerical naturalperiod
is
twice
the
time
increment,
If
ht>O.5,
onthe
otherhand,
the
equal
rootis
pesitive,
indicating
that
d.
decreases
to
zero withoutchafiging
its
sign.
This
rneansthat
the
numerial natural
period
is
infinite.
In
Fig,5,
the
numericaldisplacement
responseis
shown
for
h=O.1.
The
employedinitial
conditions aTethose
given
in
Eq.17.
It
is
seenin
this
figure
that
the
nttmericalperiod
is
sma]lerthan
the
true
period,
gradually
approachingthe
true
one as toAtdecreases.
The
reiationshipbetween
the
numerical andtrue
damping
ratiosis
illustrated
in
Fig.6.
Considering
that
the
true
solutionof
the
darnped
systern
is:
t,o
MPLITUOE
(o)"Jhttt,9
/t/1
1,
,ii
o・
'・・.
.
・
lti/t
t
's'v
'=:'
5XDafit
asaT
a6
/-dn
'
:J:-i.o
L
i・o
cb)
wAt-t,i-Lot
L
t,e{c)wnt.
o,st.
o---t.
s
-t.o
i'
Fig.s
NurnericaL
Free
Vibration
Response
for
h=:O・1
Fig.6
Nurnerical
Damping
in
CDM
(hio)
(Analytical
SoLution,
CDM,
andd.
in
Eq.15>
x(t)=exp(-hcvt)-(c,・cos
a,Dt+ct・sin toDt)・・---・---・・t--・・---・・・---・--・・・・-・・・---・---{24)
in
which too=a)Ji=7ITand comparing
this
withthe
numerical solution(Eq.15),
the
numericaldamping
ratiois
adjustedto
:
7ie=7i'!'-'"H"'-''hHH''-"'"-"''"H-''"H'''"'H''"H-'''H-H-"HH''1''-'-H"HHHHH'"''-'""H--'(25)
to
As
evidenced
from
this
figure,
the
numerica}damping
ratiois
very closeto
the
true
damping
ratioin
the
entire stability range of caAt
if
h
is
small.'Ash
increases,
the
errorin
the
damping
ratioincreases,
Further,
for
eachh,
the
discrepancy
between
the
numerical andtrue
damping
ratiosbecomes
smaller withthe
decrease
of wAt.5.
Concluding
Remarks
The
important
findings
obtainedin
this
part
ofthe
study canbe
summarized asfollows.,
-33-1)
The
stabilitylimit
elCDM
when-appliedto
the
damped
systemis
identical
withthat
for
the
undamped
system.
In
damped
systems, tobtSt=2-Vi-=Ei
is
acriticalpoint.
For
w's smaLIerthan ajb,the
numerical solution canbe
expressed as
Eq.15
anditherefore
ensures oscillatory solution.2)
The
error causedby
employingCDM
in
PSD
test
is
characterized
as
the
period
distortion
(Figs.l
and4),
For
undamped systems, no arnplitude
decay
(numeTical
damping)
i$
introduced
in
the
computation.In
damped
systems,
on
the
other
hand,
the
numerical
viscousdamping
is
larger
than
the
true
viscousdamping.
The
discrepancy
between
the
numericaland
true
dampings
is
as shownin
Fig.6.
Reference$
1.
Takanashi,
K,,
etal.:
Seismic
Failure
Analysis
ofStructures
by
Computer
Putsatof
On-Line
System,
Joucnal
ofInstitute
ofIndustrial
Science,
theUniversity
ofTekyo,
VoL26,
No.11,
1974,
pp.13-25
(in
Japanese).
2.
Iemura,
H.,
Kato,
T.
andKawatani,
Y.
:
Hybrid
Experiments
onEarthquake
Respo]se
ofOriginal
&
Repaired
RC
Members,
Proceedings
ofthe
6th
Japan
Eafthquake
Engineering
Sympesium,
December
!982,
pp.825-832
3.
0kamoto,
S.,
etal,:
A
Progress
Report
onthe
Full
Scale
Seismic
Experiment
of aSeven
Story
Reinforced
Concrete
Building,
Research
Pape[,
Building
Research
Instltute.
Ministry
ofConstruction,
Ne.94,
March
1982,
92
p.
4.
Mahin,
S,
A.
andWilliams,
M.
E.
:
Cemputer
Controlled
Seismic
Performance
Testing,
Proceedings
of the2
ndASCE-EMD
Speeialty
Cenference
onDynamic
Response
ofStructures,
Atlanta,
GA,
january
1981.
5.
McClamroch,N.
H.,
SeTakos,
J.
andHanson,
R,
D.
:
Design
andAnalysis
ofthe
Pseudo-Dynamic
Test
Method,
UMEE
Repo[t
SIR
3,
Department
ofCiviL
Engineering,
The
University
ofMichigan,
September
1981.
6.
,Nakashima,
M,
:
Stabitity
andAccvracy
ofIntegration
Techniques
in
Pseudo
Dynamic
Testing,
Research
Paper,
Building
Research
Institute,
Ministry
ofConstruction,
No,105,
March
1984,
58
p.
7.
Shing,
B,
P,
andMahin,
S,
A,
:
ExperimentaL
ETfor
Propagation
in
Pseudodynamic
Testing,
Earthquake
Engineering