• 検索結果がありません。

Newmarkのβ法におけるエネルギー保存則について(梗概)

N/A
N/A
Protected

Academic year: 2021

シェア "Newmarkのβ法におけるエネルギー保存則について(梗概)"

Copied!
10
0
0

読み込み中.... (全文を見る)

全文

(1)

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"g

S*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 many

time

integration

algorithms

in

structural

dynamics,

For

example

we can

give

the

Euler

method,

the

Runge-Kutta

method'}

(so-called

explicit

method), and

the

Newmark's

B

methodZ),

the

Wilson's

e

method3} and

the

Hilber's

a

method`}

Cso-called

implicit

rnethod).

And

when we use

them

practically,

the

numerical stability

and

th6

accuracy come up

important

problems,

Nickell5),

Bathe6),

Hilber')

discussed

the

stability

by

using

the

concept

of

spectral radius of matrix.

These

are considered as

extentions

of way of

CourantS)

and

yon

Neumanng).

And

also

FujiiiO),

Oden]i),

Belytehenkoie)

discussed

the

stability

from

the

point

of view of "energy rnethod" which was

developed

by

Richtmyer

and

Mortoni'}.

But

as

to

the

accuracy,

there

are

many

points

which must

be

clarified,

In

general,

next

three

methods are used

to

error estimations.

1)

Comparison

of approximate solutions with

the

exact

solutions

by

taking

a single-degree-of-freedom

system

in

free

vibration.

2)

Comparison

of several solutions

by

changing

the

value of

discrete

time

interval

or

the

constants

fi,

e,

or a.

3)

Comparison

of

the

energy

of

the

system

with

the

energy worked

by

externaL

forces.

The

first

method

is

useful

if

we want

to

know

the

rough estimate of errors.

And

it

looks

only

this

method with which we can appteach

theoretically.

With

this

method some censiderable

developments

are

done

by

above

papers2L5)"T)

and

by

Browni4),

And

by

introducing

mode superposition,

these

discussions

can

be

applied

for

a

multi-degree-of-freedom

system.

But

they

do

not

inform

us of

the

aecuracy

for

each nonliner,

forced

vibration

problems.

The

second method

is

very

practical.

And

it

seems

the

most reliable

method

in

present

techniques.

But

if

we make

time

interval

short

in

oder

to

decrease

the

di$cretized

error,

the

round-off

error

must

grow

up

because

a

large

number

of

time

steps are needed.

So,

we

cannot

decide

which solution

is

more accurate.

The

third

methocl,

known

as "energy

balance

check",

is

very easy

to

use.

But

when we

get

the

good

balance

between

input

energy and

output

energy,

how

should we consider

this

good

balance?

Can

we say

that

the

discretized

euors and

the

round-off

errors

are small or only one of

them

is

small, or nothing we can

tell?

In

order

to

clear

this

problems,

we need

to

know

the

energy reiations which must

be

hQlded

in

those

numerical methods.

About

these

energy

problems,

Greenspan's

works]5)'i6) as

to

the

Newtonian

n-body

problems

are very suggestive

for

us,

however

his

woiks

have

not

been

taken

in

structural

dynamic

fields,

He

showed

that

the

energy conserving

law

is

still satisfied

for

the

classic

Euler's

method

by

modifying

the

equation

of

motion.

In

previous

references,

the

finite

difference

is

introduced

only

to

telations

between

displacernents,

velocities and

accelerations.

And

the

equation

of rnotion

is

not changed

its

form.

But

in

Greenspan's

way,

the

equation of motion

is

also modified

in

order

to

eliminate

the

vagueness

in

the

sence

of

energy

conservation.

In

this

paper,

from

the

same

point

ef view of

Greenspan,

we

prove

two

new energy

balance

equations

for

the

Newmark's

P

method.

One

is

for

linear

structure

models

and

the

other

is

for

geometrical

nonlinear

truss

models.

However

only

for

the

latter

case we need

to

modify

the

fo[m

of

the

equation of rnotion.

'

Research

Associate

of

Osaka

University

Maouscript received

May

S,

1985

(2)

-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

a

positive

mteger and

Dn

be

'a

displacement

(oJ

generalized

cobrdinate>

vector of structural system at

time

of

nXAt.

Similarly

let

Vh,

A.

and

E.

be

a velocity vector, an acceleration vector

and

an

external

force

vector, respectively,

Then

the

equation of motion'for a

linear

structura! system can

be

written as

follows.

''

''

[M]An+[C]V;i+[K]Dn=En''''''''''''''''''''''''''''''''''"''H''''''''''''"'''''''''''''H''"'''''''H''''''''''''';'''''(1)

where

[M]

is

a mass matrix,

[C]

is

aviscous

damping

matrix and

[K]

is

a stiffness,matrix.

And'both

[M]

and

[K]

are assumed as symmetric matricesi.

The

Newmark's

finite

difference

expressions can

be

written as

Vi+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, and

here

another con$tant

7

is

assumed

to

112

already.

For

these

equations

the

next

theorem

can

be

proved,'

'

'

Theorem

(

I

)

If

we apply

the

N'ewmark's

finite

difference

expressions

(

2

),

<

3

)

to

the

linear

structural

system of which

the

equation of motion

is

given

as

Eq.

(

1

),

then

the

next energy

balance

equation

is・,

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 of

Eq.(

O

as

k+1

and

k,

and

by

adding

the

two

equations

we

have

[M]<A..,+A.}+[C]{V;,.,+mp+[K]{D..,+D,)=E,.,+E.・・・・・・・・・・・・-・・・・・・・・・・・・・・・・・・・・-・・・---・---(5)

Multiplyipg

by

{D,.,-D,)'

from

left

,to

this

equation we

obtain

(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)

and

Eq.C3)

the

fi'rst

term

of

Eq.(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

matrix

V:.,[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 as

follows.

,

,

'

{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

summing

up

th{s

equation

from

k=O

to

h=n-1,

we

have

(3)

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 as

the

error

term.

Belytschenko

also

introduce

the

last

term

of

Eq,

(

7

}

in

his

Eq.

(23)

of reference

12>.

But

the

key

equation of

this

proof

is

Eq.

(10).

With

this

equation we can eliminate

terms

of'Ti・・・T;tTi,

wr"'PiCi-h

er''en-i・

By

restricting

our

problems

within more

simple

case

Theorern

(

I

>

is

rewritten

as

follows.

1)

For

the

case ofB=114, and all

initial

displacements

and velecities and all

damping

coefficients are zero,

Do

==

VL=O,[C]=[O],

then

we obtain

n-1

Z,+-l,=Zp,---・・--・t・--・-・-・-・・・---・-・---・----・-・-t・--・-・・・・・.・--・-・・---・",.-.".-,"(]z)

t!o

Here

we need onLy results at

the

last

time

step

in

order

to

calculate

the

left

terms,

2)

Even

though

P=#114

in

the

case

1),

by

adding

the

condition

A,!O,

we obtain

'

n-1

7;,+;e;,+en=

£ pk・-・--・-・・・-・・・J・・-・・・・・・-・・--・--・・・-・・・-・-・・・・・・・・・--・--・・-・・-・・・・・・・・・-・・・-・・-・・-・・--・・・・・・・・・・・・{13)

ic=o

So,

in

the

case ofBtll4

the

error

term

of

e.

is

related

to

the

energy

balance

equation.

However

it

is

noticeable

that

e.

has

not cumulative nature

in

its

form.

3)

For

the

case of

B=l14,

[C]=[O]

and

E.iO

(free

vibration), we can obtain

the

following

classical energy

conservative

law.

Th+;ICi=To+;K''-'--"--"'-'''-'m'''"--m'--"''-"-"'-m'-'-'""'----m-'-"'-""'""'"(14)

It

is

remarkable

that

this

equation

holds

regardless

the

time

interval

At.

Newmark2)

and

Nickel15)

showed

the

fact

that

in

free

viblation of a single-degree-of-freedom system,

decays

of amplitudes

do

not occur

for

the

case of

B=1/4

and

[C]=[O],

This

characteristic nature can

be

interpreted

generally

by

Eq,

(14),

In

the

single-degree-of-freedom

system

the

maximum

amplitude

would

be

measuredi

at

the

time

when

the

velocity

is

zero.

So

in

that

time

following

equation must

be

satisfied.

14L=TVb・・---・---・--・---・・-・---・--・----・---・----・-・-・---・-・----・--・-・-・--・・-(15)

Since

the

value of

wr

and

;IC,

are

deteTmined

only

by

each amplitude, so

the

maximum amplitude

keeps

the

same

value.

If

any change of maximurn amplitude appear,

those

must

be

derived

from

the

round-off error

but

not

from

the

Newmark's

algorithm.

4)

If

the

calculation

is

done

for

relatively

long

duration,

the

motion of

the

system will

gradually

come

to

a stop.

In

that

case, we will

be

able

to

assume

D,,

Vh,

A,,

D.,

Vh

and

A.

be

zero.

Then

we obtain

n-1 n-1

Zd,==Zp,-・---・・・-・--・--・・-・----・----・---・---・-・-・-・---・--・-・・---・-(16)

ktiO k=O

As

can

be

seen

in

these

process,

we can say

two

different

sides

of

the

Newmark"s

method.

First,

the

Newmark's

method

has

a

trend

that

the

theoretical

errors

do

not

grow

up

in

the

sence of

total

energy.

And

by

Theorem

{

I

)

we can check whether

the

,calculation

is

done

validly according

to

the

Newmark's

argoTithm.

Second,

even

if

Theorem

(

I

)

is

satisfied exactty, we should not conclude

the

theoretical

errors of

D.,

VC,

ancl

A.

are small,

Because

Theorem

{I)

is

only a necessary condition.

In

order

to

interpret

this,

let

consider

a single-digree-of-freedom system.

(Shown

as

Fig.

1)

The

mass m, and

the

stiffness

ts,

are

given

as

(O.

312

rr)2and

1.

0,

respectively.

So

the

natural

period

equals

O,

3

(sec).

External

excitations

for

this

example

are shown

as

Table

1

(It

keeps

reproducibility of

this

paper,

).

The

linear

interpolation

is

adopted

for

estimating values

between

peek

to

peek.

Direct

integration

formula

of

the

Newmark's

method

is

used which

is

proposed

by

ChaniS),

and calcuiations are

executed

in

a single

precision

(36

bit>,

Fig.

2

shows

the

displacement

of

two

cases

(At=O,

Ol

and

At==O.

02)

forB=1/4.

In

both

cases

T.+

1-;,

agrees

n-1

well with

E]

p,

at

s(sec),

but

out-looks of

the

response are

quite

different

to

each other,

k#1

This

is

only

one

case

in

which

a

good

agreement

has

not

been

obtained,

but

it

shows

enough

that

the

energy

balance

is

not sufficient condition

to

assure

the

accuracy.

Next,

Table2

shows

the

energy

balance

for

various cases.

(4)

-52-Architectural Institute of Japan

NII-Electronic Library Service

ArchitecturalInstitute ofJapan 2

Cq)2.0

1.0

o.'-LO

-2.0

Fig.1

9

m

One-mass

system

Tnewil.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.941.001.07

Table1

'

I.5 l4

1v.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

GAL

lo.se

o.ge 15.59

-O.11

18.52

O.10 5.8B

-1.21

19.60-23.23 7.Sl 41.66 9.21

13.52-16.93-46.80-28.22-49.25-77,33-51.34

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

1.63 112.01

1.70

230.87

1.SO

139.94

1.86 174.15

1,92

-255.78

2,Ol

-313.06

2.21 289.30

2,27

258.13

2.32

-292.44

2.39 5.32

2.45

280.77

Model

Excitation

{Fig.])

(EL-CENTRO

SEC

GAL

2.52

-45,96

2.57

148.is5

2.65 203.56

2.71

106.53

2.77

-31,36

2.B9

101.33

2.98

-78.72

3.07

50.96

3.13

-151.61

3.21' 6.39

3.25

-2el.94

3.39

ISB.85

3.42

-91.e3

3,53 167.39

3.60

-35.25

3.67 59.07

3,74

-72.13

3.83

・30,.57

3.90

-179.69,

4.01 22.25

N.S)m

SEC

GAL

4.06

-42.63

4.11

21.18

4,22

-193.31

4,31

-172.68

4.42 143.0g

4.47

-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.li

213.93

5.20 26.22

5.23

122.72

5.30 126.42

5.33

106.72

5.34

'

-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 114

116

1181112 1.S41822O.7330948

D.4446092

e.24I7114

Output

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.27e4247

O.2271959

HIrEIIig6--'-677---'IIEII555--"

O.7400260

-O,O06927q

O.4SO0696

-O.O054570

O.2449881

-O.O032729O.7330986

O.44q6126

O.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 see

that

input

energies

keep

good

balance

witti output energy

fer

every cases.

The

numbers

in

the

last

column should

be

O

by

Theorem

(

I

}.

So

these

errors rpust

be

arlsed

frorn

round-off

effect.

It

iF

reasonable

that

the

eJror

in

At=O.Ol

is

larger

than

At=O.02.

3.

Energy

Balance

Equation

for

Nonlinear

Truss

Structure

In

case of

large

displacement

problems

the

equilibrium equation

must

be

considered

the

movement of structural elements.

So

the

stiffness

matrix

[K]

is

not

independent

to

the

displacement

vector

D..,

,

'

In

this

section, a energy

balance

equation

for

these

geometrical

nonlinear

truss

structures

is

introduced.

Let

M,.

be

a

posltion

vector

of

noclal

p6int

r

(which

is

considered at

the

joint

of members) at

time

of

At

×

n.

Similarty

Iet

U,.be

avelocity vector,

A.,.be

'an

ahcerelation vector

and

E.,.

be

an exterfia.1

foicevector

of nodal

pomt

r,

.

,

Then

the

equation of motion

for

the

truss

structure can

be

written as

follows,

MrAT,ic=i..

£

d"n

AtEi

!Lkit,oLt'"

\'it

".X}'k

-Er,t'''''''''''''''''''H'''''''''':''''"''-'''-''''''':'''',-'''''1"''H'''H(17)

'

where

m.

is

a

lumped

mass on

the

nodal

point

r,

adj<

r)

is

a

set

of

member's

number which

is

jointed

to

the

nodal

point

r,

At

and

E,

are

section

area and

Young's

modulus of member

i,

respectively,

and

Lc,.

is

a

length

of

member

i

at

time

of

AtXn.

L,,k

can'be expressed with

poFition

vectors as

.

.

.

L2,=(M,,-X),Dt(M,,-X),D・・・・-・-・・・・-・:--・・・・・・・-・・-・・・・・・・-・・・-・-・・・.・・・・・・・・・・・-・・・・・・・・・・・-・・・・・・・:・・・・-・・・-・・・-・・・・(18)

'

(5)

-53-NII-Electronic Library Service

where

indexes

r

and

j

mean nodal

point

of

the

member

i.

In

previous

section

Eq,

(10)

makes us

enable

to

eliminate

the

term

of

Ti・・・Zt.i,

VZ・-・l-;t-i

and

ei-・・en-i.

But

we

cannot

reduce

Eq.

(17>

to

recurrence

formula

like

as

Eq.

(10).

So

in

this

section we

introduce

amodified

form

ef

Eq,

(17)

as

foilows.

i

Er,ic+i+Er,h

Aak+t+Ar,k

-

£

Fr,h=

"-""""""""-,"-"-""M""-"---・-・---・---・(19}

Mr

2

tcadJcn

2

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 replaced

by

their

mean values at

the

time

of

Atxle

and

Abx(h+1).

The

Newmark's

finite

difference

expression

for

these

nodal

pointwise

vectors can

be

written as

follows.

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

assumed

to

112.

Then

the

next

theorem

can

be

prov,ed.

Theorem

(

ll

)

If

we apply

the

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 nodal

points

r

X

:

Summation

for

all members

t

Tr,n=t

V;,nMrUr,n

wr,n-2

1

(A`E`

)

(L".-Lt,o)'

1pr,k=2

er,n`=At!(B

Lt,o

-1)A;,nfTlrAr,n

(20),

(21}

to

the

geometrical

,

then

the

next energy

balance

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)

we

have

(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

summing

both

side of

this

equation

for

all nodal

points,

then

we

have

\{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

of

left

side of

Eq,(24)

becomes

as

follows,

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 equation

is

obviously satisfied.

4[iEa

£

d,m

fL'1=\'

(rE

n

£

odui

't}"]'''H''H''H''H''H''H''''''''''''''''''''''''--'''''''''''-''H'''--'''''''-'''''''''''''''''(26)

where

nod(

i>

means anodal

point

set which

belongs

to

member

i

and

.fL.is

ascalar

term

using

indexes

i

and r.

This

transformation

can

be

understood

by

thinking

as

follows.

In

the

left

side, .tl,. are summed

by

paying

attention

first

to

each nodal

point,

while

the

right side,

those

are summed

by

paying

attention

first

to

each member.

So,

by

setting

nod(i)=lr.jL

the

second

terTn

of

left

side

of

Eq.(24>

¢an

be

wTitten as

(6)

Architectural 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

of

F;,,,

the

internal

part

of right

side

can

be

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

and

by

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

reduced

to

'

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 equation

from

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

appeared

that,

for

the

geometrical

nonlinear

truss

structure,

we

can

say

the

similar

mentions

described

in

the

previous

section.

(In

Teorem

(

ll

)

the

damping

coefficients

are

not

considered

in

but

we

can

easily

develop

Theorem

'

{ll)

in

the

same manner as

Theorem

(I).),

The

main

difference

in

the

deriving

process

Theorem(I

),

(

ll)is

concerned

to

Eq.

<10)

and

Eq.

(29).

Eq.

(10)

can

be

deduced

froTh

Eq,

(

1

)

by

consid'ering

time

step

k

and

k+1,

while

Eq,

(29)

cannot

be

deduced

from

Eq,

(17>

in

the

cause of nonlinearity.

So

in

Thorern

(

fi

)

we nged

to

change

Eq.

(17)

to

Eq.

(19>.

However

we can say

that

it

is

better

to

use

Eq.

(lg)

than

to

use

Eq.(17L

if.we

need

the

theoretical

clearn6ss

in

the

sense

of

total

e'neigy.

,'

Next

let

considei

the

Timoshenko's

modeli9} as a

geometrical

nonlinear

free

vibration

problem

<Shown

as

Fig.

3).

In

this

rnodel

the

cross-sectional area

At,

the

Young's

rr}odulus

Ei

apd

the

initial

'Eolizontal

displacement

are set as

1,

and

the

mass

m,

is

set as

O.

OOI.

For

the

nonlinearity, we cannot use

direct

formula

for

the

Newmark's

fi

method, so we use an

iterative

scheme.

Calculations

for

each

tirne

step

is

done

unti・11

1ar"i-a?,,L$O.OOI.

Where

a?,.

is

the

h-th

assumed

holizonta.1

acceleration

in

the

h-th

discrete

time

step.

To

make clear

the

characteristic nature

of

the

Newmark's

B

method, we

'

'

'

/t

t.

/

,

Table3

Ratio

of

Energy

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

modei

rimeRatioofEne

tsec.Newnark'sMet

At.o,olAt

o.'o.

o 1.00-O,OQOIo 1.00-O.OOOIo

2.00-o.eoo4o

3.00-o.eoon-o 4.00-o.eoo7-o 5.00-o.eoo7-o

Max.R-O.OO09-o

atCt.4.55](t

Y=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

.

rneans

iTr,n

i・l. Vrtn

(7)

-55-NII-Electronic Library Service

Table4Comparison

of

theEnergy

Error

in

caseof

Eq,(17)

andEq.

(19)

R

z100

Time

sec.

o.oo1,OO2.003.004.005.00Max.

R at o9K

so30

+BO

60'I30

1

40

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 NewnarktsMe

At=o.04

ngEq.(17) Vs±ng IX6S[118B=1!12B=1!4B=116

oooO.OOOo.oooo.oooo.ooo

78t11.12322.186o.eoe2.462

66316.4885.668e.ooo-32.2e5 3446.05816.S09O.OOI4.8S7

O0213.63713.B7eO.OOI5.2S9

5e514.3687.905O.OOI1.434 O0319.55722.297O.OOI-37.077

oo]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 7.641

9.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}

t

xx

tf'A

rt"

XYJ.-f-"-"

v

'tt

et--

-x

,--Fv -"

"

-"r

ll

'nt

t

'

'd<`AV

1lii,Ab

sfk,

x'"Newmark's

Method

'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,

energy

eTrors

in

Newrnark's

B

method

those

are

almost

Next

Table4

sriows

the

comparison of energy equation

of

motion.

From

Table

4,

energy errors are not zero

in

not

increased

monotonously

like

as

In

Fig.4

these

facts

are

interpreted

more

4.

Conctusions

first

one

is

for

the

linear

structural models.

An

the

energy error

term

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

of

the

energy error

(At=O.04)

of

At=

O.Ol,

At=O.

02,

At=O.

04.

the

Runge-Kutta

method are

increased

monotonously, while

in

the

negligible.

errors

for

the

case of using

Eq.(17)

and using

Eq.

(19)

as

the

case of

Bt=:114

if

Eq.

(17)

is

adopted,

however

it

is

noticiable

those

are

Runge-Kutta

method.

apparently,

Two

theorems

concerning

to

energy

balance

are

proved

which

should

be

satisfied

in

the

Newmark's

procedure.

The

d

the

second one

is

for

the

geometrical

nonlinear

struss

models.

And

(8)

Architectural Institute of Japan

NII-Electronic Library Service

ArchitecturalInstitute of Japan

J

'have

the

cumulative nature.

'

With

these

theorem

we can

grasp

the

new aspect of

the

Newmarkis

method

from

the

point

of view' of energy

conservatlon.

'

Acknoytledgements

,

,.

,

The

author

wish

to

thank

the

late

Prof.

K.

Washio

for

his

valuable suggestions and encouragements.

Also,

the

author wish

to

thank

Assoc.

Prof.

Y,

Inoue

of

Osaka

University

for

his

supports on

the

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

ofcemputatiop

fg[

structural

dynamics,

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

of

Approximation

OPerato[s

Ln

P[oblems

of

StructuraL

Dynarnics,

Int.

J.

Solids

Structures,

VoL7,

(1971)

pp.301-319,

'.

'

6)

K.

J.

Bathe

and

E.

L.

Wilson

:

Stability

and

Accuracy

Analysis

of

Direct

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.

vort

Neurnann.

R.D,

Richtmyer

:

A

Method

for

tlte

Nllmerical

Calculation

of

Hydrodynamic

Shocks,

J,.Appl.

Plty.

,

VoL21,

(1950)

pp.232-237

,

-

'

lo)

H.

Fujii

:

Finite

'Eiement

Schemes

;

Stability

and

Conyergence,

Advances

in

Computatinal

Mettiods

ig

Structural

Mec'hanics

?:g7:.}eSlgp!

281

-diil・

iF・

Oden,

R-

W・

Ciough

and

Y・

Yamamote),

The

Univ,

of

Atabama

i,n

'uuntsrilte

press,

Huntsvine,

10

J.

T.

Oden,

R.

B.

Fost

:

CorivergEnce,

Accuracy

and

Stabllity

of

Finite

Elrnent

AppTokimations

of a

CIa'ss

of

Non:linear

Hyperbolic

Equations,

Int.

J.

Num..

Meth.

Engng.

VoL.6

(1973)

pp.357-365

,

'

12)

T.

Belytschenko,

D.F.

Sehoeberle

:

On

the

Unconditional

Stability

of an

Implicit

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

of

Newmark's'Method

for

Numerical'

lntegratio]

ifi

StTuctural

Dynamics,

Earth.

Engng.

Struc,'

Dyn,;Vol.13,

(19855

pp.43'51

''

-

'

'

'

15}

D,

'Grtienspan

:

Numerical

Studies

of

the

3-Body

Problem,

SIAM

J.

Appl.

Math,

VoL.20,

No.1,

(1971)

pp.67-78

16)

D.

GTeenspan

:

An

Algebraic,

Energy

Conserving

Forrnulation

of

Ctassical

Molecular

and

tyew,tonian

n-Body

Interaction,

Bulletln

ef

the

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

of

Ferced

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

/

lt

lt

/.

/

t

'

t

/

'

'

-

57

NII-Electronic

(9)

NII-Electronic Library Service

1

 

UDC  :624

042

7 :620

1 :531

6 日本建 築 学 会 構 造 系 論 文 報告 集 第

359

昭和

61

年1 月

Newmark

β

お け る

保存則

つ い

梗概 )

† 正 会

 橘

  §

1.

 

 

Runge−Kutta

1) , 

Newmark

β 法

z}

 

Wilson

の θ

S)

代 表

さ れ る

数 値 積 分 法

地 震 外 力

を想 定

し た

離 散 系

動 的

解 析

に しばしば 用い られ る

 

そ し て そ の

場 合

発 散

し な い た めの

数 値 安 定 性

stability

)と精 度 (

accuracy

)が 現 実 的

重 要

問 題 と

な る

 

こ の

数 値 安 定 性

問 題

に n ス テ ッ プか ら

n

1

ステッ

問 題

帰 着

さ れ,

安 定

の た めの

条 件

Nickell5

 

BatheS

藤 井

7〕に よ り 理 論 的に明 ら か に されつ つ ある

 

し か し

も う

精 度

にっ い て は

解 析

全 過

題 と なる た め か

定 性

議 論

ほど

ら か で は な い

そ れ らの

く は 正

ま る 簡

なモデル を

対 象

と し た

か, も し く は

値実

験 的

じ た も の と なっ て い るQ

 

とこ ろ で こうし た

問 題

して の

視 点

Green−

span の

っ た

研 究

15)

16}の

すこと

で き る

 

Greenspan

引 力

分 子 間 力

作 用

す る

n

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

に よ る

差 分 解 が 完 全

にエ

則 を満

た すこと

してい る

 

従 来

す れば

運 動

程 式

方 程 式

上の

問 題

差 分 解

安 定 性

などの

問 題

と は

々 の

分 野

じ られて いる が

,Greenspan

は そ の

両者

系 を作

ろ う と す る

立 場 を採

っ て い る

 本報

は これ まで

構 造 解 析

の分 野で あ ま り

目 され て い ない こ

し た

GTeenspan

立 場

か ら

改 め

Newmalk

の β

見 直

そ れ を

線 形

お よ び

何 学 的 非 線 形

造モデル に適

した場

に成 立 す るエ

ルギ

バ ラン ス公

く。

 

§

2.

モ デ ル に お け るエ

バ ラ ン ス 公

 

動 方 程 式

式で与 え ら れ る

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

9

お よ び

7

回ヨ

ロ ッパ地 震工学 会 議 (昭 和

57

9

月)にその

部 を発表

 

大学 助手

  (昭 和

60

5

8

日原 稿 受理} ただ し

M

C

K ]

は そ れ ぞ れ

行 列

粘 性 行 列

剛 性 行 列

とし

An

 

Vn

 

Dn,

 

En

は そ れ ぞ れ

At

×

n

間 後

加 速 度

速 度

変 位

外 力

ベ ク トルと す る。 また

M ]

K ]

対 称 行 列

と す る

 

こ の

Newmark

β法

用 し た 場

のエ

ン ス

公 式

成 立 す

る。

 

定 理

1

バ ラ ン ス

公 式 (

1

      n

L             n

1

    

τ广 ア。

Σ

en

e

Σ P

卍       k

¢                                       h

o た だ し

    

Tn

1

2

 

V

M

 

V.

運 動

    

d

.= 

1

2

D

..、

− D

7

C

〕(

γ,.1+

      

     

Wn

2

以 [

K

Dn

弾 性

    

en

At2

2

β

1

4

A

M

A

.:

項 >

     

PiC

1

2

D

κ. 、

− D

E

,.1+

E

外 力

るエ

      

ルギ

(証 明は本 文 参 照

 

こ の

理 か ら

Newmark

β 法

で は

en

eo

がエ

していること が 分 か る。 ま た そ れ は

め と

わ りの加 速

だ けに

依 存

し た がっ て徐々 に

増 大 す

るの で は な く

加 速 度

じ た

周 期 性

して い ること

も分

か る

  【

定 理

1】

は また

Newmark

 

Nickel1

ら に よ り

さ れ た

結 果

包 含

して い る。

    [

C

o

β

=1

4

自由 度

振 動

におい

     

ては 振 幅 誤 差 が

じ ない なぜな ら

C ]

0]

β

1

4

自 由振 動

る ことか ら

定 理

1

直 ち

関 係

か れ る

     

T

π十

Wn

To

Wo

し た がっ て

Tn

Wn

初 期 状

態 か ら

ま る

定 な

を と り

,Tn=0

の と きに

じ る

で の

最 大 振 幅

も ま た

定 値

。Newmark

Nickell

はこ の

性 質

分 方 程 式の

か ら 導い て い る が

定 理

1】

は よ り

般 的

あ り

n

自由度 系

し て の

外 枠

Newmark

β

えて い る。

 

こ れ らの ことはまた

重 要

問 題

ん で い る。

ある

解 析 結 果

して

ル ギ

保 存 則

の チェ ッ ク を

58

N工 工

Eleotronio  Library  

(10)

Architectural Institute of Japan

NII-Electronic Library Service

Arohiteotural エnstitute  of  Japan

それ がか なり

た さ れ て い たと す る

も し

差 分 形 式

か れ た

近 似 解

はエ

ル ギ

保 存 則

た さ ないと

漠 然

えて いた な ら

解 析

精 度

を か な り

め に

評 価

す るで

φ

ろ う

しか し

1

11

そ う し た

判 断

早 計

で あ るこ ど を

して いる。

静 止

か ら

振動

が あ る

お さ ま る まで

解析

が な さ れた

場 合

初 期 加 速 度

A

。は

0

最 終

ス テッ

で の

加 速

An

さ い

で あ

え ら れ

補 正 項

e

e。

も 小 さ くな

る。 し た

っ て

Newmark

β 法 が 忠 実

実 行 さ

い る

り,

化 誤 差

いか に

大 き くと も

バ ラ ン ス の

く あ うこ と が

保 証

され る

ン ス はそれぞれ よ く

して い るが

変 位 応 答

相 当 異

なっ て い る

解 析 例

Fig

2

した

これ

性 質

Newmark

β 法

有 効 性

の で は な く

ルギ

をマ クロに

え る た め

む し ろ

利 点

も言 う

べ きであ ろ う。 た だ そ う し た

を 知 ら ない と

まっ た

判 断

を な す 恐 れ が あ る

 

§

3.

幾何 学

的 非

トラスモ デル に お け るエ

ル ギ

     

バ ラン ス

公 式

 

次 式

運 動

程 式

之.

ら れ た

幾 何

的 非 線

形 ト ラス モ デル

え る

    

mrAr

iC

ZA

E

LLit

Lt,

o

∂/(

Lt,

Ltit

      i

     

iEr

in た だ し

   

Mr

r

中質

    

A

‘:

i

断 面 積

    

E

‘:

要素

i

のヤン

   

Lt,

k :

要 素

i

At

×

k

時 間 後

部 材 長

    A

k :

節 点

r の

At

×

k

加 速 度

ベク トル

   

X

k :

節 点

r

At ,

×

k

時 間 後

位 置

ベ ク

    E

h :

節 点

r の

AtXk

時 間 後

外 力

ベ ク トル こ こ で

Grgenspan

に 砺 κ

 

Xr

rc

玩鬼

 

E

k を

平 均 値

An

κ+1十

Ar

h

2

丿

rr

1+L十 丿

rr

2

Lr

k“十

Lr

D

2

 

Er

k+1

十 

E

 it

2

置 き

か え

 

こ の

き か え に よ り

修 正

され た

運 動 方 程 式

New

mark

β 法 を適 用

し た

場 合

関 係 式

か れ る 6

  【

定 理

2

ンス

公 式 ([)

    

Σ (

τ。

− T

Σ 儡

− W

Σ (

e

.バ

e

      r       l       r       n

1

    

Σ Σ

ρ

il

     

        T   陀

0 た だ し

    

Σ

べ て の

節 点

関 す

     Σ

はすべ て の

要 素

す る

      t

  

Tr,

n

1

2

 

V

nMrVr

n

  

Wl.

n

1

2

A

Es

Li、

o

L

n

− L

2

  

e.

At2 (

β

1

4

縄 。ん1η

   Pr

t

1

2

Xr

t+1

X

D

Er

h+i十

Er

k

 (

証 明

本 文 を 参 照 )

この

21

に より

前 節

同 様

の こ と が

幾 何 学 的 非 線

トラス の

場 合

も言

え るこ とにな る

こ の

場 合

簡 単

の た め

略 さ れ て

る が

拡 張

であ る

そ して

1】【

定 理

2

を 適 用 して猶エ ネル ギ

が バランス し ない

分 は必

然 的

に round

off に ょる

誤 差

と なる。 そ れ ら

を 説 明

す る

簡 単

例 題

Timoshenko

い た モ デル に よ りなされ てい る

本 文 参

  §

4.

  Newrnark

β 法 を用

いた

場 合

成 立

す る二 つの エ

保 存 則

する公

いた

。一

つ は

モデル に

する

の であ り, も う

つ は

幾何

的非 線

形 性 を有 す

ラス モ

ル に

対 す

ので

差 分 化

の た

φ

保 存 則

に形

t

要 と なる

正 項 は

め と

り の

時 間

に お ける

加 速 度

だ けに

依 存

し, した がっ て

集 積

し な が ら

増 大 す

性 質

の ない こと

かっ た

ま た

Newmark

Nickell

ら に より

見 い 出

さ れ てい た とこ ろの

1

質 点の 自 由 振 動で

β

1

4

を と れ ば numerical  

dumping

じ な と も

そ れ ら公 式 の

特 殊

場 合

にす

ない こと が 明 ら か と なっ た

通 常

般 的

力学

理 論で の問 題 と 数 値 解 析 上で の問

題 と が分 離

し て

じ ら れ る

傾 向

にあ る が

で の

両 者

してマ クロ に

め よ う と す ることによ り

られた もの で あ る

t

59

N工 工

Eleotronio  Library  

参照

関連したドキュメント

All (4 × 4) rank one solutions of the Yang equation with rational vacuum curve with ordinary double point are gauge equivalent to the Cherednik solution.. The Cherednik and the

We believe it will prove to be useful both for the user of critical point theorems and for further development of the theory, namely for quick proofs (and in some cases improvement)

It is well known that the inverse problems for the parabolic equations are ill- posed apart from this the inverse problems considered here are not easy to handle due to the

Log abelian varieties are defined as certain sheaves in the classical ´etale topol- ogy in [KKN08a], however the log flat topology is needed for studying some problems, for example

Infinitesimal actions of quadratic forms is computed in Weyl ordering and normal ordering, and these define involutive distributions on the space of exponential functions.. We

Touchdown Total may be applied as a spot spray in peppermint and spearmint. Apply spray-to-wet with hand-held equipment, such as backpack and knapsack sprayers, pump-up

一方、介護保険法においては、各市町村に設置される地域包括支援センターにおけ

用局面が限定されている︒