NII-Electronic Library Service
LntDc:6!.2o42.7:62o.
. ,3i.6{eT",
.a,i..Otfi.S.tg".CftUA'fl)""Nd.E3es"grtrJ".c.'to.",ylilggis"6eeri"gS*3etgza\e#me:k6eegfvai.fi:
ON
ENERGY
CONSERVING
LAWS
FOR
THE
NEWMARK'S
B
METHOD
by
EIZABURO
TACHIBANA",
Member
of
A.I.J.
1.
Introduction
There
are manytime
integration
algorithmsin
structuraldynamics,
For
example
we cangive
the
Euler
method,the
Runge-Kutta
method'}(so-called
explicit
method), andthe
Newmark's
B
methodZ),the
Wilson's
e
method3} andthe
Hilber's
a
method`}Cso-called
implicit
rnethod).And
when we usethem
practically,
the
numerical stabilityand
th6
accuracy come upimportant
problems,
Nickell5),
Bathe6),
Hilber')
discussed
the
stabilityby
usingthe
concept
of
spectral radius of matrix.These
are considered asextentions
of way ofCourantS)
and
yonNeumanng).
And
alsoFujiiiO),
Oden]i),
Belytehenkoie)
discussed
the
stabilityfrom
the
point
of view of "energy rnethod" which wasdeveloped
by
Richtmyer
andMortoni'}.
But
asto
the
accuracy,there
are
manypoints
which mustbe
clarified,In
general,
nextthree
methods are usedto
error estimations.1)
Comparison
of approximate solutions withthe
exactsolutions
by
taking
a single-degree-of-freedomsystem
in
free
vibration.2)
Comparison
of several solutionsby
changing
the
value ofdiscrete
time
interval
orthe
constants
fi,
e,
or a.3)
Comparison
ofthe
energyof
the
system
withthe
energy workedby
externaL
forces.
The
first
methodis
usefulif
we wantto
know
the
rough estimate of errors.And
it
looks
onlythis
method with which we can appteachtheoretically.
With
this
method some censiderabledevelopments
aredone
by
abovepapers2L5)"T)
andby
Browni4),
And
by
introducing
mode superposition,these
discussions
canbe
appliedfor
amulti-degree-of-freedom
system.But
they
do
notinform
us ofthe
aecuracyfor
each nonliner,forced
vibrationproblems.
The
second methodis
verypractical.
And
it
seemsthe
most reliablemethod
in
present
techniques.
But
if
we maketime
interval
shortin
oder
to
decrease
the
di$cretized
error,the
round-offerror
mustgrow
upbecause
alarge
numberof
time
steps are needed.So,
wecannot
decide
which solutionis
more accurate.The
third
methocl,known
as "energybalance
check",is
very easyto
use.But
when weget
the
good
balance
between
input
energy andoutput
energy,how
should we considerthis
good
balance?
Can
we saythat
the
discretized
euors and
the
round-offerrors
are small or only one ofthem
is
small, or nothing we cantell?
In
orderto
clear
this
problems,
we needto
know
the
energy reiations which mustbe
hQlded
in
those
numerical methods.About
these
energyproblems,
Greenspan's
works]5)'i6) asto
the
Newtonian
n-bodyproblems
are very suggestivefor
us,however
his
woikshave
not
been
taken
in
structural
dynamic
fields,
He
showedthat
the
energy conservinglaw
is
still satisfiedfor
the
classicEuler's
methodby
modifyingthe
equationof
motion.In
previous
references,the
finite
difference
is
introduced
only
to
telations
between
displacernents,
velocities andaccelerations.
And
the
equation
of rnotionis
not changedits
form.
But
in
Greenspan's
way,the
equation of motionis
also modifiedin
orderto
eliminate
the
vaguenessin
the
senceof
energy
conservation.In
this
paper,
from
the
samepoint
ef view ofGreenspan,
weprove
two
new energybalance
equationsfor
the
Newmark's
P
method.One
is
for
linear
structuremodels
and
the
otheris
for
geometrical
nonlineartruss
models.
However
onlyfor
the
latter
case we needto
modifythe
fo[m
ofthe
equation of rnotion.'
Research
Associate
ofOsaka
University
Maouscript receivedMay
S,
1985
-50-Architectural Institute of Japan
NII-Electronic Library Service
ArchitecturalInstitute of Japan
2:
,Energy
Balance
Equation
for
Linear
Structural
System
Let
At
be
a
length
6f
discrete
time
interval,n
be
apositive
mteger andDn
be
'a
displacement
(oJ
generalized
cobrdinate>
vector of structural system attime
ofnXAt.
Similarly
let
Vh,
A.
andE.
be
a velocity vector, an acceleration vectorand
an
external
force
vector, respectively,Then
the
equation of motion'for alinear
structura! system canbe
written asfollows.
''
''
[M]An+[C]V;i+[K]Dn=En''''''''''''''''''''''''''''''''''"''H''''''''''''"'''''''''''''H''"'''''''H''''''''''''';'''''(1)
where
[M]
is
a mass matrix,[C]
is
aviscousdamping
matrix and[K]
is
a stiffness,matrix.And'both
[M]
and[K]
are assumed as symmetric matricesi.The
Newmark's
finite
difference
expressions canbe
written asVi+im
VL=l>L/
(Ak+i+AD
''''''''''''''r'''''H''''''''''''''''''''"''-'''''''''''''''''''''''''''''''''''''''''''"''''''H(
2
}
'
D,.,-b,=!liLt
{v),.,+
v,}-At2{114'-i?)
(A,,,-A,}
.:...,...,...,,...,,,.,,,.,,,..,,.i,H...H...1....(
3
)
'
where
fi
is
a constant, andhere
another con$tant7
is
assumedto
112
already.For
these
equationsthe
nexttheorem
canbe
proved,'
'
'
Theorem
(
I
)
If
we applythe
N'ewmark's
finite
difference
expressions(
2
),
<
3
)
to
the
linear
structuralsystem of which
the
equation of motionis
given
asEq.
(
1
),
then
the
next energybalance
equationis・,
satisfied.
'
,
'.,
n-1 n-1(7;t-To)+,
£
.-,dic+(P-;,-Vll)+Ce.-eo)=i,pit'''-'-''"''''''''-'''''・・--・・・-・・・・・・-・・・・・・--・i'''・・・・・・・・・・・(4)
,where
.
.
,
7;,=}
V;
[M]
V;,
d,=e(Vi.,-D,)'[C](Vl.,+K}
uc,=SDS[K]D.
'
,
e.!
lft:i
(p-t)'AA[M]An
'
'
P.=g(D,.,-D,)t(E,.,+ED
,Proof
:
Taking
n ofEq.(
O
ask+1
andk,
andby
adding
the
two
equations
wehave
[M]<A..,+A.}+[C]{V;,.,+mp+[K]{D..,+D,)=E,.,+E.・・・・・・・・・・・・-・・・・・・・・・・・・・・・・・・・・-・・・---・---(5)
Multiplyipg
by
{D,.,-D,)'
from
left
,to
this
equation weobtain
(D,.,-D.)t[M](A..,+AD+(D..,TD,)t[C](U,.,+Vl)+{D..,-D.)t[K](D...,+DD!(D,.,-DDt(E,.,+E.)
'
H-'''''''''''-'''''-'''-1''-'''-'''-'''--'''''''''''''''''''''''"''''1''''''H''''I・・t・-・・(6)
From
Eq.(2)
andEq.C3)
the
fi'rst
term
ofEq.(6)
becomes
'
(D,.,-DDt[M](A,.,+AD=(L,.,+V;,)t[M](VL.,-VD-At2(t-B)(A..,-A,)t[M](A..,+A,)・・・・・・・・・・・・(7)
Since
[M]
is
a
syinmetric
matrixV:.,[M]-=Vi[M]VL.,,
AL,,[M]A.=A:[M]A,.,
;
So,
Eq.t7
).
is
rewritteri as'
,
'
(Dk.,-Dit)t[M](At.i+ziD=2(Th.i-TD+2(elt+i-en・・・・・・・・・・・・・・・--・・-・・・・・・・・・・・・・・・・・・・・-・・・--・・・-・・・・・・・--・・・-・・・(8)
'
Similarly
the
third
term
6f
Eq.(6)
becomes
'
'
(D,.,-DDt[K]
(DE.,+D,)=2<
V-;,.,-
W;,)・・--・・・
・・・-・----・---・
---・----・---・・・-・・・・-・・・・・・・(
9
)
Then.
Eq.(6)
is
[eviritten asfollows.
,
,
'
{Tk.L-Tk)+;(Dh.i-Dk}t[C](Vi.L+Vl)+{M'L+i-IVk)+(ei+i-et)=S(Dit+L-Dh)t(Et.i+Eh)・・・・・・・・・・・・・・・(10)
By
summingup
th{s
equation
from
k=O
to
h=n-1,
wehave
NII-Electronic Library Service
n-1 n-1(71-n)+Zdk+(PP;t-wt)+(en-eo)=Xph・-・・・・・'・・・・・'''-・・・-i・・・・-・-・・・・・・・・・-・・・-・・・・・・-・・・・・・・・・・・-・・-・・・・・・(10
k-O k!O
(q.
e.d)
In
Theorem
(
I
),
e.-
e,
is
con$idered asthe
errorterm.
Belytschenko
alsointroduce
the
last
term
ofEq,
(
7
}
in
his
Eq.
(23)
of reference12>.
But
the
key
equation ofthis
proof
is
Eq.
(10).
With
this
equation we can eliminateterms
of'Ti・・・T;tTi,wr"'PiCi-h
er''en-i・
By
restricting
our
problems
within moresimple
case
Theorern
(
I
>
is
rewrittenas
follows.
1)
For
the
case ofB=114, and allinitial
displacements
and velecities and alldamping
coefficients are zero,Do
==VL=O,[C]=[O],
then
we obtainn-1
Z,+-l,=Zp,---・・--・t・--・-・-・-・・・---・-・---・----・-・-t・--・-・・・・・.・--・-・・---・",.-.".-,"(]z)
t!o
Here
we need onLy results atthe
last
time
stepin
orderto
calculatethe
left
terms,
2)
Even
though
P=#114
in
the
case1),
by
addingthe
conditionA,!O,
we obtain'
n-1
7;,+;e;,+en=
£ pk・-・--・-・・・-・・・J・・-・・・・・・-・・--・--・・・-・・・-・-・・・・・・・・・--・--・・-・・-・・・・・・・・・-・・・-・・-・・-・・--・・・・・・・・・・・・{13)
ic=o
So,
in
the
case ofBtll4the
errorterm
ofe.
is
relatedto
the
energybalance
equation.However
it
is
noticeablethat
e.
has
not cumulative naturein
its
form.
3)
For
the
case ofB=l14,
[C]=[O]
andE.iO
(free
vibration), we can obtainthe
following
classical energyconservative
law.
Th+;ICi=To+;K''-'--"--"'-'''-'m'''"--m'--"''-"-"'-m'-'-'""'----m-'-"'-""'""'"(14)
It
is
remarkablethat
this
equationholds
regardlessthe
time
interval
At.
Newmark2)
andNickel15)
showedthe
fact
that
in
free
viblation of a single-degree-of-freedom system,decays
of amplitudesdo
not occurfor
the
case ofB=1/4
and
[C]=[O],
This
characteristic nature canbe
interpreted
generally
by
Eq,
(14),
In
the
single-degree-of-freedomsystem
the
maximumamplitude
wouldbe
measurediat
the
time
whenthe
velocityis
zero.So
in
that
time
following
equation must
be
satisfied.14L=TVb・・---・---・--・---・・-・---・--・----・---・----・-・-・---・-・----・--・-・-・--・・-(15)
Since
the
value ofwr
and;IC,
aredeteTmined
onlyby
each amplitude, sothe
maximum amplitudekeeps
the
samevalue.
If
any change of maximurn amplitude appear,those
mustbe
derived
from
the
round-off errorbut
notfrom
the
Newmark's
algorithm.4)
If
the
calculationis
done
for
relativelylong
duration,
the
motion ofthe
system willgradually
come
to
a stop.In
that
case, we willbe
ableto
assumeD,,
Vh,
A,,
D.,
Vh
andA.
be
zero.Then
we obtainn-1 n-1
Zd,==Zp,-・---・・・-・--・--・・-・----・----・---・---・-・-・-・---・--・-・・---・-(16)
ktiO k=O
As
can
be
seenin
these
process,
we can saytwo
different
sidesof
the
Newmark"s
method.First,
the
Newmark's
methodhas
atrend
that
the
theoretical
errorsdo
notgrow
upin
the
sence oftotal
energy.And
by
Theorem
{
I
)
we can check whetherthe
,calculation
is
done
validly accordingto
the
Newmark's
argoTithm.Second,
evenif
Theorem
(
I
)
is
satisfied exactty, we should not concludethe
theoretical
errors ofD.,
VC,
anclA.
are small,Because
Theorem
{I)
is
only a necessary condition.In
orderto
interpret
this,
let
consider
a single-digree-of-freedom system.(Shown
asFig.
1)
The
mass m, andthe
stiffnessts,
aregiven
as(O.
312
rr)2and1.
0,
respectively.
So
the
naturalperiod
equalsO,
3
(sec).
External
excitationsfor
this
example
are shownas
Table
1
(It
keeps
reproducibility ofthis
paper,
).
The
linear
interpolation
is
adoptedfor
estimating valuesbetween
peek
to
peek.
Direct
integration
formula
ofthe
Newmark's
methodis
used whichis
proposed
by
ChaniS),
and calcuiations areexecuted
in
a singleprecision
(36
bit>,
Fig.
2
shows
the
displacement
of
two
cases(At=O,
Ol
andAt==O.
02)
forB=1/4.
In
both
casesT.+
1-;,
agreesn-1
well with
E]
p,
ats(sec),
but
out-looks ofthe
response arequite
different
to
each other,k#1
This
is
only
one
casein
which
a
good
agreement
has
not
been
obtained,
but
it
shows
enough
that
the
energy
balance
is
not sufficient conditionto
assurethe
accuracy.Next,
Table2
showsthe
energybalance
for
various cases.-52-Architectural Institute of Japan
NII-Electronic Library Service
ArchitecturalInstitute ofJapan 2
Cq)2.0
1.0
o.'-LO
-2.0
Fig.1
9
mOne-mass
systemTnewil.54!83e
g
I=LS41822
n...
T"':n::I:s3;3,:Z>
/
X
n xTXN>
>(NLt
c
XN,,
)
SECo.0.04O.10Q.16O.22O.26O.29O.33O.37O.43O.47O.5BO.62O.6Se.72O.79O.87O.941.001.07Table1
'
I.5 l41v.6,4.7
Fig.2
Compa[isonof
1
4.8.t
5.
I
I t-AtFO.Ol
NyX----AtTO.02 n-1
in=eo
"c
disp.
and energy
Table2
Energy
GALlo.se
o.ge 15.59-O.11
18.52
O.10 5.8B-1.21
19.60-23.23 7.Sl 41.66 9.2113.52-16.93-46.80-28.22-49.25-77,33-51.34
Balance
for
Linear
External
SEC
GAL
1.09'-42.04
1.17
87.96
1.32-・166.23
1.38
--81.14
1.41
-81.14
1.44.92.70
1.48-86.78
1.51
--le5.84
1.54-.125.44
1.63 112.011.70
230.87
1.SO
139.94
1.86 174.151,92
-255.78
2,Ol-313.06
2.21 289.302,27
258.13
2.32-292.44
2.39 5.322.45
280.77
Model
Excitation
{Fig.])
(EL-CENTRO
SEC
GAL
2.52-45,96
2.57
148.is5
2.65 203.562.71
106.532.77
-31,36
2.B9
101.33
2.98-78.72
3.07
50.963.13
-151.61
3.21' 6.393.25
-2el.94
3.39
ISB.85
3.42-91.e3
3,53 167.393.60
-35.25
3.67 59.073,74
-72.13
3.83
・30,.57
3.90-179.69,
4.01 22.25N.S)m
SEC
GAL
4.06-42.63
4.11
21.18
4,22-193.31
4,31-172.68
4.42 143.0g4.47
-4.67
4.62 252.14 4.67-200.41
4.76
sg.58
4.83.267.B8
4.97 174.37 5.04 29.50 5.li213.93
5.20 26.225.23
122.72
5.30 126.425.33
106.725.34
'
-23.43
5.45 16e.90 5.5I.100.06
'
'
At
O,Ol
e.o2
BInput
Energy
En
IX4
116 1181112 O.4323840 e.3190128 O.27e41e2 e.2271879 114116
1181112 1.S41822O.7330948D.4446092
e.24I7114Output
Energy
Tn+Wn en Tn+Wn+en O.4323923 O,O.3195500
-O.OO05267
O.2710431-O.OO06I84
O.2278263-O,OO06304
O.4323923
O.3190233 O.27e4247O.2271959
HIrEIIig6--'-677---'IIEII555--"
O.7400260-O,O06927q
O.4SO0696
-O.O054570
O.2449881-O.O032729O.7330986
O.44q6126O.2417152
reund-off errorEn-(Tn+Wn+en).O,OOO083
-o,oooles
-O.OOO065
-o.ooooso
-O.OOOO08
-O.OOO038
-O.OOO034
-O.OOO038
'
'
From
this
table,
we can seethat
input
energies
keep
good
balance
witti output energyfer
every cases.The
numbers
in
the
last
column shouldbe
O
by
Theorem
(
I
}.
So
these
errors rpustbe
arlsedfrorn
round-offeffect.
It
iF
reasonablethat
the
eJrorin
At=O.Ol
is
larger
than
At=O.02.
3.
Energy
Balance
Equation
for
Nonlinear
Truss
Structure
In
case oflarge
displacement
problems
the
equilibrium equationmust
be
considered
the
movement of structural elements.So
the
stiffnessmatrix
[K]
is
notindependent
to
the
displacement
vectorD..,
,
'
In
this
section, a energybalance
equationfor
these
geometrical
nonlineartruss
structuresis
introduced.
Let
M,.
be
aposltion
vectorof
noclal
p6int
r(which
is
considered atthe
joint
of members) attime
ofAt
×n.
Similarty
Iet
U,.be
avelocity vector,A.,.be
'an
ahcerelation vectorand
E.,.
be
an exterfia.1foicevector
of nodalpomt
r,.
,
Then
the
equation of motionfor
the
truss
structure canbe
written asfollows,
MrAT,ic=i..
£
d"n
AtEi
!Lkit,oLt'"
\'it
".X}'k-Er,t'''''''''''''''''''H'''''''''':''''"''-'''-''''''':'''',-'''''1"''H'''H(17)
'
where
m.
is
alumped
mass onthe
nodalpoint
r,
adj<
r)is
aset
of
member's
number whichis
jointed
to
the
nodal
point
r,
At
andE,
aresection
area andYoung's
modulus of memberi,
respectively,
andLc,.
is
alength
of
memberi
at
time
ofAtXn.
L,,k
can'be expressed withpoFition
vectors as.
.
・
.L2,=(M,,-X),Dt(M,,-X),D・・・・-・-・・・・-・:--・・・・・・・-・・-・・・・・・・-・・・-・-・・・.・・・・・・・・・・・-・・・・・・・・・・・-・・・・・・・:・・・・-・・・-・・・-・・・・(18)
'
-53-NII-Electronic Library Service
where
indexes
r
andj
mean nodalpoint
ofthe
memberi.
In
previous
sectionEq,
(10)
makes usenable
to
eliminate
the
term
of
Ti・・・Zt.i,
VZ・-・l-;t-i
and
ei-・・en-i.But
wecannot
reduceEq.
(17>
to
recurrenceformula
like
asEq.
(10).
So
in
this
section weintroduce
amodifiedform
ef
Eq,
(17)
as
foilows.
i
Er,ic+i+Er,h
Aak+t+Ar,k
-
£
Fr,h=
"-""""""""-,"-"-""M""-"---・-・---・---・(19}
Mr
2
tcadJcn2
where(LLh+i+LLD12-Lte
(M,k+t+M,D12-(Jif},k+i+X},D12
F;,=A,E,
Lm
(Li,.+i+LLD12
In
Eq.
(lg)
all variables are replacedby
their
mean values atthe
time
ofAtxle
andAbx(h+1).
The
Newmark's
finite
difference
expressionfor
these
nodalpointwise
vectors canbe
written asfollows.
At
V;,k+i-
",k=
2
(Ar,k+t+Ar,D''-'''-'''-'''--''--'''-'''-'''''''-'''--'''''''''''-'''''''''''''''''''''''"'''''''・・・t・・・-・・-(2O)
'M,k+t'X;,h=A2t!(V?,ic+i+U,k)-At'(t-B)(Ar,lt+t-Ar,D''''''''''''''''''''''''''''''H''''''-H''-'''-''''''-''"(21)
where also another constant
7
is
assumedto
112.
Then
the
nexttheorem
canbe
prov,ed.
Theorem
(
ll
)
If
we applythe
Newmark's
finite
difference
expressions
nonlinear
truss
structure
of
which
the
equation
of
motion
is
given
as
Eq,
(19)
equation
is
satisfied.4
(
Tnn-
Tr,o>+
4
(
VV},n-
VVLo}+
Zl]
(e
r,n-eT,o)=;
I
:
£
ll
p
where
£
:
Summation
for
all nodalpoints
r
X
:
Summation
for
all memberst
Tr,n=t
V;,nMrUr,n
wr,n-2
1
(A`E`
)
(L".-Lt,o)'
1pr,k=2
er,n`=At!(BLt,o
-1)A;,nfTlrAr,n
(20),
(21}
to
the
geometrical
,
then
the
next energybalance
r,rcl'---H'-H''H''-'''''''"''-'''"''''''''''''(22)
4
(M,k+i-Xr,nt(Er,it+i+Er,D
Proof
:
Multiplying
(X,,,.,-M,Dt
from
left
to
Eq.
(19)
wehave
(X},rc+i-Xr,D`Mr(Ar,t+i+Ann12-
£
(M,rc+i-Xr,DtF;,k=(M,ic+i-Xr,k)`(Er,t+i+Er,k)12'"'''''''''"''"''(23)
tEad""
By
summingboth
side ofthis
equationfor
all nodalpoints,
then
wehave
\{M,it+i'M,DtM"Ar,it+i+Ar,Dl2N\,.i,.(X},k+i-M,k)tF;,k=4(Xnk+i-M,Dt(Er,t+i+Er,D12'"''"''(24)
With
the
expressions of(20),
<21),
the
first
term
ofleft
side ofEq,(24)
becomes
asfollows,
1
ii4(Xr,h+i'AQ,DtMr(Ar,h+i+A"D=4(Tr,k+i-Tr,D+4(er,k+i-er,D'''-''''''''''-'''-'''''''--'''-''''''''''''''(25)
As
to
summation, next equationis
obviously satisfied.4[iEa
£
d,m
fL'1=\'
(rE
n£
odui't}"]'''H''H''H''H''H''H''''''''''''''''''''''''--'''''''''''-''H'''--'''''''-'''''''''''''''''(26)
where
nod(
i>
means anodalpoint
set whichbelongs
to
memberi
and.fL.is
ascalarterm
usingindexes
i
and r.This
transformation
canbe
understoodby
thinking
asfollows.
In
the
left
side, .tl,. are summedby
paying
attentionfirst
to
each nodal
point,
whilethe
right side,those
are summedby
paying
attentionfirst
to
each member.So,
by
setting
nod(i)=lr.jLthe
second
terTn
of
left
sideof
Eq.(24>
¢anbe
wTitten asArchitectural Institute of Japan
NII-Electronic Library Service
ArchitecturalInstitute of Japan
>](,t.Z,.(Xr,it+i-XT,D"F;,s]=2I]l(X?,k+i-M,ntF;,ic+(X},k.t-X),DtFSM・・・・・--・・・-・・・・・:・-l・・-・・・:・・・・・・・・・・'・・・・・(27)
'
From
the
definition
ofF;,,,
the
internal
part
of rightside
canbe
rewritten as'
(M,,.,-M,ntpt,,+(x},,.,-x,,,}tks,i=AiEilL,`ilXI.kici,,2,)L"O)(x;,..,x,,.-X;,.xi,,
±
?rs,k.,Ac,,k.i-x;,kx),k
'
-2x;,,.,x),,.,+2jE';,.xi,.)-=AtE2`2I,i'fii',l..ftiikilij!`'O){(x.,,.,-x},..,)t(xL,.-,-x},h.,)
-(x.,.-x},iot(M・,.-x},Dl.
'
'
.
1
andby
Eq.<18)
',
.
I
=AtE,illli,1・fii],IZ.fltiiti,:,LSO)(I.:,,.,-L7,,)'
,
・
,
t
'
'
A,E,
=2L,,,i{L`'k"-LLit)2-(Li,h-Lt,o)!l
=
wr,k+1-
wr,k
So,
Eq.(27)
becomes
'
,
>][,emZ,.(M,k+i-M,DtF:Lh)=>l.](WLh"'wr,D'''''''''''''''''''H''''''''''''''''''''''''''''-''''''''・・・・'・'・''''''・・,・・(28)
'
From
Eq.(25>
and(28),
Eq.(24)
is
reducedto
'
Z(Tr,k+i-Tr,D+Z](iVZlt+i-WZD+
£
(er,h+i-er)=
£ pr,kH'''''''''''-''''H''':''''''L''H''''''''''-'''''''''-''(29)
r t r r
By
summing of,this equationfrom
ic;O
to
h;n-1,
we obtain,
4(Tr,n-Tr,e}+>l](Vll,n-i-la)+4(er,n7er,o}='0(:illi,pr,kl'"''''''"'''''''''''"'''''''''''''''''''''''''''''''"I''(30)
(q.
e,d)
'
It
appearedthat,
for
the
geometrical
nonlineartruss
structure,
wecan
saythe
similar
mentions
described
in
the
previous
section.(In
Teorem
(
ll
)
the
damping
coefficientsare
notconsidered
in
but
wecan
easilydevelop
Theorem
'
{ll)
in
the
same manner asTheorem
(I).),
The
maindifference
in
the
deriving
process
Theorem(I
),
(
ll)is
concerned
to
Eq.
<10)
and
Eq.
(29).
Eq.
(10)
canbe
deduced
froTh
Eq,
(
1
)
by
consid'eringtime
stepk
andk+1,
whileEq,
(29)
cannot
be
deduced
from
Eq,
(17>
in
the
cause of nonlinearity.So
in
Thorern
(
fi
)
we ngedto
changeEq.
(17)
to
Eq.
(19>.
However
we can saythat
it
is
better
to
useEq.
(lg)
than
to
useEq.(17L
if.we
needthe
theoretical
clearn6ss
in
the
sense
of
total
e'neigy.
,'
Next
let
consideithe
Timoshenko's
modeli9} as ageometrical
nonlinear
free
vibrationproblem
<Shown
as
Fig.
3).
In
this
rnodelthe
cross-sectional areaAt,
the
Young's
rr}odulusEi
apdthe
initial
'Eolizontal
displacement
are set as1,
andthe
massm,
is
set asO.
OOI.
For
the
nonlinearity, we cannot usedirect
formula
for
the
Newmark's
fi
method, so we use aniterative
scheme.Calculations
for
eachtirne
stepis
done
unti・111ar"i-a?,,L$O.OOI.
Where
a?,.is
the
h-th
assumedholizonta.1
accelerationin
the
h-th
discrete
time
step.To
make clearthe
characteristic natureof
the
Newmark's
B
method, we'
'
'
/t
t.
/
,
Table3
Ratio
ofEnergy
Error
for
Timoshenko's
Models
,(Fig.3)
i
Energ}r
Errer
R=(
Th+Wn-To.-Wo
)1(
To+Wo
)X,100
4
T
t
I
l
(1)
r2
F,ig.3
×
,E,-i
ix
9gxO
,Ti
.itiyayLg"l:e"
%>n.e
Tiinoshenke's
modeirimeRatioofEne
tsec.Newnark'sMet
At.o,olAt
o.'o.
o 1.00-O,OQOIo 1.00-O.OOOIo2.00-o.eoo4o
3.00-o.eoon-o 4.00-o.eoo7-o 5.00-o.eoo7-oMax.R-O.OO09-o
atCt.4.55](tY=O.5
S.O.2
,o2
Atgo.o4.
o..OOOI
O.OOOI
.OOOI.
O.OOOI O.Ooo1'O.OO04
.oool
e.ooo6,.OO02
O.OO06,ooe2
o.oooe.OO04
O.OOOS
,16)
{t.3.88)
vhere Tn,blh
.
rneansiTr,n
i・l. Vrtn-55-NII-Electronic Library Service
Table4Comparison
oftheEnergy
Error
in
caseofEq,(17)
andEq.(19)
R
z100
Time
sec.o.oo1,OO2.003.004.005.00Max.
R at o9Kso30
+BO60'I30
140
oEi'3pt+H=20Ticr
Ratie
of Energy Error R=C Tn+Wn-Te-Wo)!C
To+Wo)XIOO
Newnar'k's
Using
B.1!4
o.ooo
o.
2.157
-1.
11.331 11. 10.14e 11.-8.621
17.
IO.O04 16. 13.815 17.(t=:4.08)(t=4.
fin---NethedT=O.5 NewnarktsMeAt=o.04
ngEq.(17) Vs±ng IX6S[118B=1!12B=1!4B=116oooO.OOOo.oooo.oooo.ooo
78t11.12322.186o.eoe2.462
66316.4885.668e.ooo-32.2e5 3446.05816.S09O.OOI4.8S7O0213.63713.B7eO.OOI5.2S9
5e514.3687.905O.OOI1.434 O0319.55722.297O.OOI-37.077oo]ct=1.24){t=O.56)Ct=3.8B)Ct=:O.92)
'---"fi----Runge-Kutta'
Method
L-Azg-t-gfi
--f
s-rr Methed T=O.5 At;O.04 Eq.(19)
G=1!8
B.lf12
o.ooo o.ooo 7.6419.elB
7.439 10.519・-60.618
..4.160
-O.936
-8.027
O.370 1.796-69.370-126.378
92){t=3.44}(tts1.16}
txx
tf'Art"
XYJ.-f-"-"
v'tt
et--
-x
,--Fv -""
-"r
ll
'nt
t'
'd<`AV
1lii,Ab
sfk,
x'"Newmark'sMethod
'1]At=O.04
:Y=O.5
Using3=114E
.(17>
jEl[
Newrnark
'At=O.
Y=O.5
Uslng
sTitl5Iil5al
ethod-04B=1/4E
.{19)ft1J-tw
.
ivlAAAitAt
II
fi
lj
,A
vlvIAjfi
l'x
t["x
IVvly
lj
tyvvvvy
1.0
2.
Fig.4
also apply
the
Runge-Kutta
method.Table3
shows
the
energy
errors
for
the
cases
In
these
cases
fi
is
set
as
114,
As
can
be
seen'from
Table
3,
energyeTrors
in
Newrnark's
B
methodthose
are
almost
Next
Table4
sriowsthe
comparison of energy equationof
motion.From
Table
4,
energy errors are not zeroin
notincreased
monotonouslylike
asIn
Fig.4
these
facts
areinterpreted
more
4.
Conctusions
first
oneis
for
the
linear
structural models.An
the
energy errorterm
is
shown as-56-ii,
lf,AAi',1
Ail
iii,i"ilAAAlj-X
il
VVYYVS
VUviUVV,
O
3.0
4.0
5.0(sec)
Comparison
ofthe
energy error(At=O.04)
of
At=
O.Ol,
At=O.
02,
At=O.
04.
the
Runge-Kutta
method areincreased
monotonously, whilein
the
negligible.
errors
for
the
case of usingEq.(17)
and usingEq.
(19)
asthe
case of
Bt=:114
if
Eq.
(17)
is
adopted,however
it
is
noticiablethose
are
Runge-Kutta
method.apparently,
Two
theorems
concerningto
energybalance
areproved
whichshould
be
satisfied
in
the
Newmark's
procedure.
The
d
the
second oneis
for
the
geometrical
nonlinearstruss
models.
And
Architectural Institute of Japan
NII-Electronic Library Service
ArchitecturalInstitute of Japan
J
'have
the
cumulative nature.・
'
With
these
theorem
we cangrasp
the
new aspect ofthe
Newmarkis
methodfrom
the
point
of view' of energyconservatlon.
'
Acknoytledgements
・
・
,
,.
,
The
author
wishto
thank
the
late
Prof.
K.
Washio
for
his
valuable suggestions and encouragements.Also,
the
author wish
to
thank
Assoc.
Prof.
Y,
Inoue
ofOsaka
University
for
his
supports onthe
present
investigation.
ReterenceS
,
O
L.
Collats:The
Numerical
Treatment
qf
Differential
Equations,
3rd
ed.,Springer-Ve[lag,
(1960)
pp.61-71
2)
N.
M.
Newrnark
:
A
Method
ofcemputatiopfg[
structuraldynamics,
Proc.
Am.
Soc.
Civ.
Engs.
,85,
.EM3,
(1959}
p.p,
67-94
si)
K.J.
Bathe,
E.L.
Wilson
:Numerical
Methods
in
Finite
Element
Analysis,
Prentice-Hall,
(1976}
pp.319-322
4)
H.
M.
Hilber,
T,
J.R.
Hughes.
R.
L.
Tayler
:
'Improved
N
umericatDissipationforTimeIntegrationAlgerithmsinStru.ctural
Dynamics,
Earthq.
Eng]g.
StTuc.
Dyn.,
Vot.5,
(1977)
pp.283-292
・
''
5)
R.
E.
Nickerl
:
On
the
Stability
ofApproximation
OPerato[s
Ln
P[oblems
ofStructuraL
Dynarnics,
Int.
J.
Solids
Structures,
VoL7,
(1971)
pp.301-319,
'.
・
'
6)
K.
J.
Bathe
andE.
L.
Wilson
:
Stability
andAccuracy
Analysis
ofDirect
Integ[ation
Methods,,Earthq.
Engng.
Struc.
Dyn.
,
VoLl,
(1971)
pp.283-292
.
7)
H.M.
Hilber:
Collocation,
Dissipation
and`Overshoot'
for
Time
Integration
Schemes
in
StTuctural
DyFa,rnics,
Earthq.
.Engng.
Struc.
Dyn.,
VoL6,
(]9Z8)
pp.99-117
,
'
'8}
cRh'eCAOnU:aall'n ,K']oFolie3d2riC(hiSg'2sH)'pLpe.W3;-:74Uber
die
Partiellen
Differenzengleichungen
der
Mathametischen
physic,
Mathematis.
g}
J.
vortNeurnann.
R.D,
Richtmyer
:A
Method
for
tlteNllmerical
Calculation
ofHydrodynamic
Shocks,
J,.Appl.
Plty.
,VoL21,
(1950)
pp.232-237
,
-
・
'
lo)
H.
Fujii
:
Finite
'Eiement
Schemes
;
Stability
andConyergence,
Advances
in
Computatinal
Mettiods
ig
Structural
Mec'hanics
?:g7:.}eSlgp!
281
-diil・
iF・
Oden,
R-
W・
Ciough
andY・
Yamamote),
The
Univ,
ofAtabama
i,n
'uuntsrilte
press,
Huntsvine,
10
J.
T.
Oden,
R.
B.
Fost
:CorivergEnce,
Accuracy
andStabllity
ofFinite
Elrnent
AppTokimations
of aCIa'ss
ofNon:linear
Hyperbolic
Equations,
Int.
J.
Num..
Meth.
Engng.
VoL.6
(1973)
pp.357-365
,
'
12)
T.
Belytschenko,
D.F.
Sehoeberle
:
On
theUnconditional
Stability
of anImplicit
Algorithm
,for
Nonlinear
Structural
Dynamlcs,
J.
Applied
Mech.,
"975)
pp.865-8fi9
,
13)
R.
D.
Richtmyer,
K,
W.
MoTton
:
Difference
Methods
fo[
Intial
Value
Problems,
Intersclence'PuGtishers,
New
York,
(1967)
,
L
t
t
pp.132-137
-
.
14)
J.M:
Browh
:
The
Discretization
Error
ofNewmark's'Method
for
Numerical'
lntegratio]
ifi
StTuctural
Dynamics,
Earth.
Engng.
Struc,'
Dyn,;Vol.13,
(19855
pp.43'51
''
・
-
'
'
'
15}
D,
'Grtienspan
:
Numerical
Studies
ofthe
3-Body
Problem,
SIAM
J.
Appl.
Math,
VoL.20,
No.1,
(1971)
pp.67-78
16)
D.
GTeenspan
:
An
Algebraic,
Energy
Conserving
Forrnulation
ofCtassical
Molecular
andtyew,tonian
n-BodyInteraction,
Bulletln
efthe
American
MetAematlcaL
Soc..
VoL.79,
No.2,
(1973),
pp.423-427
17)i
E.
Kurihayashi
and et'al:
DigitaL
Earthquake
Reco[ds,(
I
),
ThePublic
Werks
Research
Institute,
Mmis{ry
efConstruction.Report-876,
(1973)
(In
Japanese)
'
'
'
l8)
S.P.'
Chan,
H:L.
Cox,
W.A.
Benfield
:
Transient
Analysls
ofFerced
Vibrations
6f
Cornplex
Struclural-Mechanical
Systerns,
J,
Royal
Aerenaut.
Soc.
Vol.66.
(1962)
pp.457-460
・
l9)
S.
Timoshenko
:
Vibratiett
prob]ems
in
engineering,D.
Van
Nostrand
Co.
New
York,
0954).
pp.143-144
'
tt
/
ltlt
/.
/t
'
t
/
'
'
-
57
NII-ElectronicNII-Electronic Library Service
1
論文
】
UDC :624.
042.
7 :620.
1 :531.
6 日本建 築 学 会 構 造 系 論 文 報告 集 第359
号・
昭和61
年1 月Newmark
の
β
法
に
お け る
エ
ネ
ル
ギ
ー
保存則
に
つ い
て
(
梗概 )
† 正 会員
橘
英
三
郎
*§
1.
緒言
Runge−Kutta
法
1) ,Newmark
のβ 法
z},
Wilson
の θ法
S)等
に代 表
さ れ る数 値 積 分 法
は地 震 外 力
等
を想 定
し た離 散 系
の動 的
な解 析
に しばしば 用い られ る。
そ し て そ の
場 合
,
解
が発 散
し な い た めの数 値 安 定 性
(
stability)と精 度 (
accuracy)が 現 実 的
に重 要
な問 題 と
な る。
こ の
内
,数 値 安 定 性
の問 題
は基
本
的
に n ス テ ッ プか らn
+1
ステップ
の問
の問 題
に帰 着
さ れ,安 定
の た めの条 件
がNickell5
),
BatheS
),
藤 井
7〕らに よ り 理 論 的に明 ら か に されつ つ ある。
し か し
,
も う
一
方
の,
解
の精 度
にっ い て は解 析
の全 過
程
が問
題 と なる た め か安
定 性
の議 論
ほど明
ら か で は な い。
そ れ らの多
く は 正解
の求
ま る 簡単
なモデル を対 象
と し た議
論
か, も し く は数
値実
験 的
に論
じ た も の と なっ て い るQとこ ろ で こうし た
問 題
に対
して の別
の視 点
をGreen−
span の行
っ た研 究
15)・
16}の中
に見
い出
すことが
で き る。
Greenspan
は引 力
や分 子 間 力
の作 用
す るn
一
体 問題
に お い て相 互 作
用 の形 式 を 変 更
する こ と に よ りEuler
法
に よ る差 分 解 が 完 全
にエネ
ルギ
ー
保
存
則 を満
た すことを
証明
してい る。
従 来
とも
す れば,
運 動
方
程 式
や構
成
方 程 式
上の問 題
と差 分 解
や安 定 性
などの問 題
と は別
々 の分 野
で論
じ られて いる が,Greenspan
は そ の両者
の整
合
を考
え一
つ の閉
じ た力
学
系 を作
ろ う と す る立 場 を採
っ て い る。
本報
告
は これ まで構 造 解 析
の分 野で あ ま り注
目 され て い ない こう
し たGTeenspan
の考
え方
に近
い立 場
か ら,
改 め
てNewmalk
の β法
を見 直
し,
そ れ を線 形
お よ び幾
何 学 的 非 線 形
の構
造モデル に適用
した場合
に成 立 す るエネ
ルギー
バ ラン ス公式
を導
く。§
2.
線
形
モ デ ル に お け るエネ
ルギ
ー
バ ラ ン ス 公式
運
動 方 程 式
が次
式で与 え ら れ る系
を考
え る。
[
M ]
An
+[
c
]
Vn
+[
K
]
Dn =En
†日本 建 築 学 会 大 会 学 術 講 演 梗概 集
(
昭 和54
年9
月)
,
お よ び第
7
回ヨー
ロ ッパ地 震工学 会 議 (昭 和57
年9
月)にその一
部 を発表。
*
大
阪大学 助手
・
工修
(昭 和60
年5
月8
日原 稿 受理} ただ し,
[
M
]
,
[
C
]
,
[
K ]
は そ れ ぞ れ質
量行 列
,
粘 性 行 列
,
剛 性 行 列
とし,
An
,
Vn
,
Dn,
En
は そ れ ぞ れAt
×n
時
間 後
の加 速 度
,
速 度
,
変 位
,外 力
の各
ベ ク トルと す る。 また,
[
M ]
,
[
K ]
は対 称 行 列
と す る。
こ の
系
にNewmark
のβ法
を適
用 し た 場合
,次
のエネ
ルギ
ー
バ ラン ス公 式
が成 立 す
る。【
定 理
1
】
エネ
ルギ
ー
バ ラ ン ス公 式 (
1
)
n−
L n−
1(
τ广 ア。)
+Σ
山
+(
嗾一
既)
+(
en
−
e
。)
=
Σ P
卍 k=
¢ h=
o た だ し,
Tn
=1
/
2
V
:
[
M
]
V.
:(
運 動
エネ
ルギ
ー
)
d
.=1
/
2
(D
..、− D
,)
7[
C
〕(
γ,.1+囚
:(
逸
散
エネ
ルギ
ー
)
Wn
=
ユ/
2
以 [
K
]
Dn
:(
弾 性
エネ
ルギ
ー
)
en
=
At2
/
2
(
β
一
1
/
4
)
A
:
[
M
]
A
.:(
補
正項 >
PiC
=
1
/
2
(
D
κ. 、− D
,)
丁(
E
,.1+E
,)
:(
外 力
によ
るエネ
ルギ
ー
)
(証 明は本 文 参 照〉
こ の
定
理 か らNewmark
のβ 法
で は(
en
−
eo
)
がエネ
ルギ
ー
補
正
項
の働
きを
していること が 分 か る。 ま た そ れ は初
め と終
わ りの加 速度
だ けに依 存
し,
し た がっ て徐々 に増 大 す
るの で は な く,
加 速 度
に応
じ た周 期 性
を有
して い ることも分
か る。
【
定 理
1】
は また,
Newmark
,
Nickel1
ら に よ り示
さ れ た次
の結 果
を包 含
して い る。[
C
]
=
[
o
]
,
β
=1
/
4
の一
自由 度
系
の自
由
振 動
においては 振 幅 誤 差 が
生
じ ない なぜな ら,[
C ]
=
[
0]
,β
・
!
1
〆
4
の自 由振 動
であ
る ことか ら【
定 理
1
】
よ
り直 ち
に次
の関 係
が導
か れ る。
T
π十Wn
=
To
十Wo
し た がっ てTn
+Wn
は常
に初 期 状
態 か ら定
ま る一
定 な値
を と り,Tn=0
の と きに生
じ る一
自
由
度
系
で の最 大 振 幅
も ま た一
定 値
を保
つ。Newmark
やNickell
はこ の性 質
を微
分 方 程 式の解
か ら 導い て い る が,
【
定 理1】
は よ り一
般 的
で あ り,
n
一
自由度 系
に対
し て のエネ
ルギ
ー
的
な外 枠
をNewmark
のβ
法
に与
えて い る。こ れ らの ことはまた,
別
の重 要
な問 題
を含
ん で い る。今
,
ある解 析 結 果
に対
してネ
ル ギー
保 存 則
の チェ ッ ク を一
58
一
N工 工一
Eleotronio LibraryArchitectural Institute of Japan
NII-Electronic Library Service
Arohiteotural エnstitute of Japan