NASA-CR-168993

19820017656

A Reproduced Cop_
t

oF

Reproduced for NASA

by the

NASA

Scientific

and Technical

Information

Facility

LAhL,_EYRESE'Ap.r-_
CENTER
LIORARy N ASA
HAMPTON, V!R'JlNIa

FFNo 672 Aug 65

.... ""

'

" .....

: :

3 1176 01327 5772

.........

" "

..........

-JPL PUBLICATION 8!-72

(NASA-CR-I,_8u931 S 1EC_U_{ALZg_,A_ICS

I18_-2_.
32

AIiAL_.SES_.ES_ZSG A_;Z CGS.iLA:&[._(JCt

"

Propu._.sio,.
Lab.}

59 _.n_;A6._/aAOI
CSCI 2.3_.

27962

, :, :
-:

Structural Dynamics Analyses

Testing and Correlation

T.K. Caughey

:.

May 1. 1982

!.

Nationa_
Aercnautics
and
Space Acminis:ration

Uacla s
G3/39

-.

Jet Propuls{on Laboratory

CaNornia ,N_;h_._
: -'""o of Tchnc:::Cy
_
Pasadena. Ca!ifornia

""

JPL PUBLICATION 81-72

Structural Dynamics Analyses

Testing and Correlation
T. K. Caughey

May 1. 1982

N,ttionai Ae:emlutics and

Space Adm:n;strab_m
Jet Propulsion Laboratory
031ifof['_3 If'uq',dLl[O Of Tec:_,,qo_c_y

The,rt_earch d("scribod
Lnbor,ltory.
Aetonauhcs

in this publrC_!_n W_S C.3.rr_,,.__ul by tho Jc! Pr_l.'_.Irs_cr_

C_tt.f('rm,,L Institute _,! Techn_oqy.
under c_t_trac! _,_th the t_,l!tC.r_,l!
and Space Admm_strahon.

!
I

.
!

. ACI_NOI_LEDG_,'E:_T

,_

The research

Applied

tute

1"

by _Ir. Sa=uel

i!
i:
!

and Space

L".

t.
I

!,echanlcs

n this
Division

of Technology

under

Venneri

Technology.

publicatLon
of the

Jet

NASA: contract

_::terlals

l.zas carried
Propulsicn

out

under

Laborato_,,.

._o. IIAS 7-100.

and Structt,res

Division,

the

auspices

California

Z_.e effort
Office

of the
InstI-

was supported
of Aeronautlcg

ABSTRACT
This report examines
the predictions
vibration
-

,-

some aspects

of analytical

modeling

tests on such structures,

of the lack of close correlation

of dynamic

between

strue.tures and the results

and suggests _ys

in which the correlation

may be improved.

of

.............................

7.

INTRODUCTION

1-1

II.

_21AI.YTICAL
.tIODELING
A._'I)PREDICTIO.N
OF RESPONSE.........

2-1

A.

...................

2-1

Systems .................

2-2

B.

C.

Structural

Dyna_aics

1.

Continuous

2.

Discrete

3.

Excitation

4.

Nonclassical

5.

Useful Properties
of Discrete Syste=s with
Classical
._ort_al ,_Iodes ...............

Systera

Systems ..................

2-3

of Pure :lodes ..............

Reductien

Normal Hodes

2-5

.............

2-6
2-11

.....................

2-12

1.

Theorem

.....................

2-12

2.

Theorem

I I .....................

2-15

Effect

of blodellng

Errors

on Predicted

1.

Case (a) - Short Transient

2.

Case (b) - Persistent

of Neglecting

tligher

Response

.....

Loads ..........

Excitation

D.

Effects

E.

Errors

in Eige;lw_lues and Eigenvectors

F.

Errors

in Fk>daI "F(wce" Coefficients

2-22

..........

2-22

Flodes ............
....

2-19

2-23
......

2-27

!"
~

7_
_.,.g

...........

2-28

Coh'rE.Yrs

III.

DYNAMIC TESTING

A.

Modal

B.

Errors

(Cont lnued)

........................

Testing

3-1

......................

3-i

in Modal Testing

.................

3-6

i.

C.

IV.

Nonclassical

Norma!

Modes

2.

Impure Modal Excitation

3.

Measure=ent

4.

Effects of Discretizatlon

Other

Identification

CONCLUSIO,_S

Errors

Conclusions

B.

Reco_endations

3-6

..............

3-6

.................

Techniques

t_{DRECO._DIENDATIONS

A.

.............

or Conden_atlon

.............

. .............

.......................

.....................

3-6
.....

3-7

3-10

4-1

3-1

_-.'

REFERE,;CES ..............................

R-!

APPL'NDIX :

A-I

AN F2L-_."!
PLE

........................

".i

Designing
most often
Since

a structure,

perforn.ed

"

mimic tile
modern
to

clude

behavior

the

use

cest

of accuracy,

of highly

accurate

try

are

more cost

craft

and aerospace

until

recently
of the

=ethods

rather

r.ethods

and the

the

are

desired

"qualify"
or

scientist,

is,

"given

ho_ accurate
the

dyna--.ic

aspects

analytical

use

of highly

the

must

response
problem

wlth

t,alance

the

it

found

that,

of the

to accept

due

analytical

very

the

_o the

is kzamm with

model be to obtain
This

to correlate
predictions

of

double

cut-and-try
models,
model of

testing
the

to

mathematician

an important

question

only

accuracy,

limited

examines

tim results
based

air-

cut-and-try

analytical

'adequate'

report

conplex

analytical

unlil:e

cost,

cut-and-

decades,

bet_.'een

progra_

if
coat

the

to use

crude

engineer,

against

for

efficient

a fairly

tl:e

While

expensive,

to a limited

structure?"
trying

be that

To co=promise
but

Clearly,

may _,'ell

modeling

environment

analytical

of

then

system

may pre-

leith

analytical

rigor

a physical

co.-parable

used

Since

g'hlle

system.

methods.

accurate,

model,

dynamic
the

bec,_es

the

analytical

modeling.

and to resort

accuracy,

th.m

Industry

wtlling

must al_'ays
that

system

of

prior
mode

or tiv_e constderatio:ls

system,

have

structure.

analytical

of modeling

models

ts

environment

the

reasonable

b_dy, it _.'as more cost

than

analytical

of the

a structure

automobile

structure

the

effective

automotive

frequently

the

financial

physical

design
that

capability

analytical

industries

the

_'ith

envtron'_ent

model of the

therefore,

the

modeling
the

dynamic

to their

system

have

degree

anti testing

eng!neers

be subiected
important,

technique.q

of building
methods

not

is very

of analytically

a prescrlb,-d

by nsin;_ an analytical

of tile physical

analytical

curvature

_,,ill
it

any desired

the

no_,.tdays

many structures

to co_issiening,

to sur_,ive

on rather

predtct'.'ons
a number

of dynamic
crude

of

of

testing

modeling.

of.

SECTION II

'

A_ALYTICAL HODELI_G

:_ :

Almost all structural

systems;

systems are distributed

this is particularly

distributed

throughout

parameter

results in a design with mass and stiffness

The complex

the system.

geometry and boundary

tlons in space vehicles seldom permit exact solutions

ferential equations describing
this reason,
techniques

finite differences,

finite element,

are normally used to discretlze

Rayleigh-Ritz,

or Galerkin
the problem

these techniques differ

they all have the same general properties.

For

They attempt

in

to approxi-

mate a space-continuous

system by a discrete system having a finite number

degrees of freedom, R.

A co_on

schemes

feature of such discrete

modeling

the first H modes of the continuous

the larger N is, the larger H may be, and the better
tion in the lower modes.
approximations

[:aturally,

the degree of approximapoor

of the continuous =odes, even when N is large.

atives of the deformations,

structure depends

in many important problems

than N.

the number of

For this reason

reduce the size of his model to more closely

spond to the number of active modes.

system

of the spatial derivatives;

in structural dynamics,

active modes, H, is much smaller

analyst will frequently

on tilespatial deriv-

the order [_ of the discrete approximating

should be large to obtain accurate approximations

dynamically

system.

chance of

The higher discrete modes are, in general,

Since the stress in a continuous

however,

This practice, unfortunately,

the
corre-

reduces

the accuracy with which stresses and forces in the structure may be determined,

A.

particularly

exact

origin

built,
For

in

the

ease

of

transient

motions.

STRUCTUP&L DYN._H CS
Damping

in d:._amic
and

form

in additic,n,
these

tees

of

approximating

is that only the first H =odes, M = I/3 N, have a reasonable

accurately

of the vehicle.

the system and reduce

_ile

condl-

of the partial dif-

the dynamical behavJors

to that of a lumped parameter system.

detail,

(continuous)

true in the case of aerospace vehicles where

the desire to minimize weight

'.

_RD PREDICTION OF RESPONSE

ns

structures
is

seldom

in we!!
it

is

::s::al

is

u_ually

a parasitic

kn.:_,m accurately

built

before

structures,

to a::g,::_e

!hal

lh:

effect
the

because

structure

t!_eda::p!ng

fs

_tructure

has

usually
vis,:eus

its
is

small.

.-

damping

and admits classical normal modes.

da=ping
is small
and the eigenvalues
approximation, at least in the lo_r

i.

Continuous

well
separated,
modes.

this

is

a reasonably

good

Systems

To illustrate
_

It will be shown that, if the

the techniques,

tlvely simple continuous

let us restrict our attention

structure

to a rela-

such as a beam, a plate, or a shell that

can be described

by Equation

2-1:

O(x) utt + LlUt + L2u = f(x,

t) on D

(2-D

u(x, 0) = ut(x, O) = 0
with

Bu = O on _D O(x)

that Equation
are m_t:

(I)

(2-1)

> O.

admits

Caughey

classical

and O'Kelly
normal.modes

L I and L2 are self-adjoint

(Reference
if

the

2-1)

have

following

sho_'n

conditions

spatial operators.

I
(2)

o_x) Ll and --_x) L2 eor_ute.

(3)

_le

boundary

conditions

operators

L l and

plete
such

of

set
that

L2.

linearly

prescribed
Under

these

independent

on _D are
conditions
eigenfunctlons

s(x_)XiX
j dx = 6ij

i "i_ij

_D XiLIXjdx= 2w _

":.L._X. dx =

_ 5:.

compatible
there

with

exist_
Xl(_)

the

a comI _[1,

_,)

(2-2)

(2-3)

(2-4)

2-2

......

2:

Thus if

u(x, t) =E

Yi(t)xi (x-)

(2-5)

i=l
k

"!

Equation

(2-1) reduces

to

"

Yl + 2'_f_i'_'i

+ _iyi2 = qi(t

where

(2-_,)

qi(t)

For ho,.':ogeneous initial

X(_x, t)f(x,

t) dx

- r)

-.

data,

Yi(t)

exp

-_j;i(t

qi(T)

dT
(2-7)

_
2.

Discrete
given

Systems
the

system

Hu + t)6 + Ku = f(t)

)
(2-8)

u(O) : 6(o) = o

Caughey

and O'kel!:"

(Reference

classica!

normal

iff

--des

2-1)

have

M, _,

sh,_-n

K-

that

'1-' *;" and :-,'-1D ccrm-t,te.

:: _ x

Equation

(2-8)

admits

If
least

H, D, and K are

positive

vectors

_(i)

s)_etrfc

senidefinite,
such

with I! positive

there

exists

definite

a complete

set

and D and K at
of ordinary

eigen-

that:

(I)T

rt "(j)

= _i

(2-9)

.(t)T

D _(J)
= 2wilt 5ij

_(i)T
_

K "(j)
_"

. (2-10)

2
= wt,51j

(2-11)

I _ [l, N1
If

we _.-rite

u(t) = _y(t)

(2-12)

_=[4_(i),
,,_(2),
...(S)]
then

'I*

_" _ :"1 +,,2

'-i:'i:
"i'_
i

= qi (t)

y!(O)--5,,t(o)
=o
where

e.(t)

= CTf(t)

(2-13)

.........................

;
'a

ORIGINALPAGE|S
OF POORQUALITY
For homogeneous

":

initial

:.Jo

data,

f]

o_p(-_jt _ _))_n _(t_- "_%(Od_

)
{2-14)

i -_[i.
"q
3.

Excitation of Pure Hodes

If in the case of Subsections II-A-1 and II-A-2 the forcing function is

given by
\

fCx,

L) = O(x)Xi(x)P(t)

f(t) = M _(i)
;
p (t)

or

I (2-15)

we see that

= p(c)Sij

qj(t)
ltence,

in the

case of Subsection

v.(t)
-J

--"
t

h (t
j

(2-16)

II-A-1,

- :)

p('-)

d_ 8..
1J

.I 0
:

."

{
sin
h.(t)
3

ti-::._ u(x,--

= exp (-..<.t)
.1 3

t) = X.(x,.'.(tJ.
]. -

2-3
-.

7t
3

(2-!7)

-....

"4

| :
i

;1

ONIGINALPAGT.IS

"

OF POORQUALITY

In the case of Subsection

II-A-2,

i
i
!

f0t

i
i

yj(t)

hj(t

_.

x)p(z)
sin

6i. i

(2-18)

o t

= ep (-=j _jt) -----J-g.

hi(t)

dr

i
I

thus

tj(t)

= (i)vi(t)"

llence

,re see

In

particular,

if

that

p(t)

in

both

= PO cos

cases
_t,

j "

then.

PO cos

i!
L
I

u(x,

t)

= Xt(x)

a pure
as

nor':_al
t .

=,

mode

is

excited.

we have

(-t - a i)

N](.j22 _ _.,2) 2 + (2wi_i)2

(2-19)

or
I

Pn cos (,2t - a i)

= 6(1)

u(t)

;
!

--

,J _ 2

._
r

where

-1

~*i-*'i

a. = ta._ -f--'f

i.

_i-

(i)
"'

Using

4.

the

result

of

Nonclassical

Equation

(2-19),

we can

d_termiue

_)i"

c'i'

and

Normal :.'edes

For simplicity,

we sha]! r,_.strictthe discussion to discrete, visceusly

,:-l..
If in .qt:bsectionIi-A-2 M-ID and . _,do not co_ute,
then

dam?ed sXstcms.

classical neff_at modes d_- not (-::ista'.:dit will be she,:..--,

thac it is i_pessiblc
-"

to

should
"311 _'t "t'

excite

be noted
i'J

J'JTI

"r]_) _ '-_ s

pure

eigen_odes

that

_,vcn

"

in

hy any
this

choice

c-_-se it

is

of

real

possible

forcing
to

excite

functions,
"fairly

it

--_--_

_--

" _-

., .....

_._z,

_,.-_,_:_:'_-;.T_=_'_

'_

ORtOINAL
p;,,GZ
_s
oF pOORQUAUTY.
TI_- for_l,lation of 2N ._pac'e

!_ as

I.e_

(Z-S)<'.m

z "(_,

then

I'quation

folhu,'s:

._,

re,-ritte,,

h,

the

form

- A.-.
+ b(t)
.-..(o)
.o
'b,'h
t"r 1,

and

(:-20)

-o)
._I-I
f it)

If tl,,,
that :

.".._rix

A is

r._c,b[_..f_,,5.liy.e,

T-1AT

_h,-re

,,xfs:s

a n,-usingul,lr

. .,' =

matt

ix T such

(2-2I)

4h(_
re

0
"I

"

L \1"2

..

.Nl

_2-22)

- _.._.z._._

r_._.I_=_-_

_.

." _

ORIGINALPAGE 15
OF POOR QUALITY

is a diagonal

matrix

of complex elgenvalues.

The matrix

T has the

structure:

[o_.l
---Lo_Ti
....
o,,:j
I

(2-23)

where

@=[_(1)z, __(2),...,__(.'I)]
and

H + ),iD+ I

In general,

__

tile_(i) are conplex N vectors.

The inverse of T is given by:

where

c3 = Cl
c_ = _,-I
-

_"

C.,= C_

_4,,.
L.

-_

OR_JN/_t.P/_GEIS

GF POORQU[_t.ITY
Let

z
m = Tv

(2-25)

Then

;
i

,.v= Av_+ qCt)

""

v(0)
= _0

fl(t) = T-lb(t)

(2-26)

T-lk(T) d:

..
Using

,,(t) --.,o [
Equations

_x(t)

(2-24)

and

exp(Xl((2-25),

we have

exp(,',l(t - :)14-1[_,'_10-1 - :.1-i] 1 M-If(:)

2_',e
o _0

(2-27)

dT

(2-28)

..

x(t)

Im(O exp(.'.. (t-_))O-])[Im(_,_l_b-1)l-l,'l-lf(,*)

dr

(2-29)

_0 t

Since .Y-If(t)__ Is a real vector,

and Im[_ exp(.'.lt)O -11
.-..:_.trices. no choice of the forcin,l
function
f(t) will

and Im(O'l O-l) are real

result
in the excita-

tion

of a pjlr3- eigen---_de.

c_ci:,,
5

.: pair

of ;-_-.-plex

of Subsections
,i,c,,rat,,ly
.,a:_:;-<_i

_dt'iltifv

II-A-!

and

It mav al.,;o be shown that

t'-:il]ll}:ii_e

:I-A-2,

p_llI'5"-

modal

the ei::,.nvalues

:_.,_rv.:il z;',!_,

i,,:.-pit_- "i:i_

.-.:_r!:-: .' i'_ :<':,:If ,m2 :.h_..,_.:,.r:v...'u,:s

i

it

_s impossible

eig_,nmo:_es.

Tim--, unlike

te'Jtin:;

llot enable

us to

of s::stems

-'ith

does

and t, igenvectors
fact,
'-,-I]

if

-n Equa:ic-.

s,4,:<at_-d,

(2-:.)

E::::-t_en

to

_i:,_".qvstem

:h,

ntm,,iamp_n-"

(2-2._1 ,:au be

ORIGWJ_L PAGZ IS

OF POORQUALFL'Y

Let 0 be such

that

:i_,
o= I;
L

If

oi:_o=

_-e set

f(t)

and

in

Equation

= ._t_O(i)p0 cos

after

the

(2-31)

(2-30)

[r.ff/_ij [ << "J2i

then

_t

initial

transients

die

Vi,

_(9

!'o*'

_i - _+ _ 1 i'"
_)2
(,
)2

N ,a't-_ki
+

ctj

tall

l/

(2-32)

out,

x(t) ....................
-

sin(,-'t
_

cos (_'t--_.)

t-

- a i - v_k)

_\2

_ _0

(2-33)

-1 ____i.1__
2

if

.u-d ..... md .c. arc

di..;tin,:t

a'_al '-el1

":T,.:t:ati,'n
(.:-_'.') ti,.t, rz.ain (.)l,-t't

,.

t,(

sc:'._.r._t'-',1
p_-:,cl.;,_lcal

2-!()

and

" ..

:}:,.n fx,,:_

d.!.-api::..: i:; _." ca_::.c l"."..q,"

.:

.-

.--

- ..-, .....

,7. ....

_-

-_-;

=_w-_.-,

......

- ' _-..

........................................

_z

Tr7 _-_
i

ORIGIP/_.L
PAG_iS
OF POORQUAL|TV

shifts

in the

equilibrium
case
that

for
if

response

portion

vector.

at tile

classically

F_eh __.ass no longer

same tize

damped

as all

systems.

the

other

In particular,

'_'k

l _ki I

2 1_k
_2 I

_k

<<

passes

through

its

masses,

as _as

the

Equation

(2-33)

sh_as

then

X(t)

Thus,

for

s=all

damping,

damping

increases

effects

of nonclassical
approximately,

5.

Useful

(2-9),

response

damping

and

(2-8)

is almost
between

mode

Systens

in the

(2-11)

case

with

= I

.i

(_-3.4)
_ ,

nor=ml mode.

eigenvalues
and

decreases,

tile response

Classical

:,ormal

of Subsection
in the

As the
the

is no longer,

2-2).

II-A-2,
form

#,]

"1"

_- cO =

a pure

(Reference

can be rewritten

+.
_'_

tile

become stronger

norr_l

of Discrete

to Systc_
(2-10),

." (i)
"-0

separation

in a pure

Properties

Returning

P0 cos(,.,t - a.)

--"

the

and the

even

ties

'

=_ =_

tkMes
,*he Proper-

Since

thL" vector.-;

nonsingular:

therl,

i)

are

_linearly

|halt.pendent,

the

_,atrix

_ I:;

fore:

,'t = (,I,-]) T (.-i)

o=
..

K = H_'

B.

"i

_'TH

"

SYSTI2i REIII'CTION
_,'t, shall

1.

now prove

two

intt, rest

in?

theorem._.

syster_,

.,tlt'h as

Theo rt-'_ I
A contim:ous

given.

It

_-l.,ssicnl

dynaic.i|

exhiblts'a
normal

complete
=odes

set

having

of

that

linearly

eiBeBfullct

or" Subsection
llldt_'ndetlt

ion:;

Xi (x),

I1-A-I,

i.q

viscously

damped

And eip_.'nvahtes

"1" i t_ [I.
_'].
(:ivt'tl
c'usly
dn=;,d discrett_o,h,_
the

havint:

the

i th ei_cnvalue

;'ector

.t positive
inte|-er
N. ther,exists
an :_tll-order
vissystem
exhibiting
a complt.te
set of classical
nortaal
.tit
"property
that
Its
t
t,i_:ellv_lhll
orrL, sponds
exa_'tly
to
_i

of

corrt'::pond_:

",nit !ntwu_

sv.qtem.

Th._t

t|lo

eotltilltlou.q

s\-stera

to ;I proiection

and

lurtht,

r the

or" the" i th L-igt.rtiunctioB

i th
of

ei._etl-

the
I

is,

,,_.t = ,_,.t

t!3 l)

_ Xi(x j)

{2-37)
--J

Proof:
pc.'Ildt'Ilt

Si:;.-_,

t_t"

the
t'Utl't

t-i._:,-:'_functions

of

the

cotltitt::oti._

vr,_blt'm

art-

Iim'._rlv

i::&'-

i,m

?:1}_ = _

t.X, i(':/-

[ - -

{ 2- .;S)

OKi.G;N,'_
PAGE l_

OF POOR QUALITY
;

cannot

vanish

_!.

X points

xj

identically
j

c (i,

unless

._) such

::

_.

that
=

:(i)

are linearly

a. 1 =- 0
the

Vi

,"

c (1,

._1).

Hence,

there

exists

vectors

_ Xi(xj

i,

(1,

R)

(2-39)

independent.

Let

Since

the .(i) are linearly independent,

, ...

(2-'.,0)

the matrix _ is nonsingular.

Let

"*!
= a2('_-l)T(_-l)

_here

is

chosen

such

that

TrM = ._

Using
discrete
_

H, D, and K constructed

as

p(x)

dr.

in Equation

(2-42)

(2-41)

to for--

the

.':th-order

systc=,
.

:1_2 + Dh + Fu = _"_(t) = h'Cff. (t)

(2-'3)

,,,(o) = ,.',(o) = o

2-!]

ORZGINALPAGE IS

OF POORQUALITY
This

syste._ has the following

properties:

ordinary .
(I)

There

exists

a coeplete

set

eigenveetors

of

(i) i c (I, N) such that

if = [(I),
as in Equation
(2-40),
then

_TH_

q
_(2)

... _(ll

= a21
i

(2)

2w.l;.
l
t

,t

N
(4)

If

f(x,

t)=_ql(t)Xl(x)o(E)
i=l

(2-_d,)

then

If

qi(t)

in Equation

O(x)f(x,

(2-a3)

t2 =_,

t)

Equation

Yi + 2_l_'iYi

which
section

is

exactly
II-A-1

llence,

tile
with

same
f(x,

as
t)

X.(x)

the

i th

given

by

dx

(2-43)

is

reduced

to:

"-

+ _aiYi = qi(t)

mode

of

Equation

the

continuous

(2-:.5)

system

of

Sub-

(2-44).

sillce

"'lt)__,x

= _,-cO

(2-a6)

then

.N
:

uj(t)

=E
i=l

"J
,(i)yi(t)

(2-47)

=EXI(_j)Yi
1=1

is,

tile

solution
Equation

of the
(2-_4).

It
{x.}
i

,:

used

solution

should

_ [!,

[I,

N].

of the

continuous

d:screte

problem

be mr,ted that
Y] that
Thus

to mimic the

may be used

Theoren

any-

dynamical

positive
system

exist,
to define

is

II-A-I

the

projection

_,'ith f(x,t)

infinite

behavior

tt,e continuous

system.

It

at N points
in a continuotts
continuous
sv.qtem.

of the
given

sets

tile sets of vectors

.,th
.
e-any ,_ -oraer
models

infinitely

of

(2-38)

in general,

tht, re exist

gth-order
visc,uslv
integer

exhibiting

.!.}

by

of points

t0 (i)}
that

c;m be

is not

surprising,

system

do not

permit

dynamical
s vste_ exhibiting
a complete
set
damped classical
normal modes ,(i) , i :- [I,

"" < :;,

there

exists

an g_th-order

a co=F-!ete

set

of linearly

having

the

pr,_perty

c.i>
, j c [1, .X2] ,

nor__aI mod,,_; :_.

"

Problem

II

Given a discrete
linearly
independent
(,i\'en

(2-4.9)

of Suhseetton

there

therefore,
that observations
vnique
identification
of the
2.

(2-48)

uj(t) - u(x.,t)
-j

".

That

(t)

tt._

NI.

discrete

independent
that

of

classical

i th -'f.';vnvaitw

eorresp,:nds
exactly
to the i ill t_igtql\,;llue
s,I" tilt' larger
system and further
tim i th ei::,'nvector
(i)
"
'
.
cgrre,;.,,nd.q
.:::a-:oily to .-. proi,,ctio::
of t]_e ,th
(i)
elgcnv,'ctor
:
of t{_t. larger
syste.-:..

ORIGINAL PRG'_ 1_

OF POOR QUALITY
.
Proof" Since tileeigenvectors of tilegiven system are 11nearly independent,
the matrix

is nonstngular

that

is,

. o
tlowever,
less

if

than

of order

N.
N2.

this

is true

there

In partieular

there

"

must

nonvanishtng

exist

at

S ts an R2 - N matrix

= S _(i)

whose columns

__

that

appropria_'e

of all

orders

one nonvanishing

minor

i c [1", N2]

consist

O or distinct
unit vectors
e.," j [1 "'2] have
th
3
ro_, which has unity. The matrix

is such

least

minors

Let

_(i)

gqmre

must exist

(2-51)

I_1 is a minor
choice

of the

1)

,(2)

of @ of order

matrix

S.

""

(2-52)

of either
zero

the

entries

null
in all

vector
but

:;2 and so does

the

(2-53"J

not vanish

for

an

Let

H2

T ....
r'2

7 :'.
r

(2-33_
"
'

ORIG,N.ed.
pAGS!_
OF POORQUALITY

:
4

Using the H2, D2, and K2 constructed

order discrete
syste=,

in Equation

(2-54)

to form the N2th-

u2(0) = _ (0_ = 0
M2u2 -+ D2u-2 +-2K2u2 = f2(t)

(2-56)

Thls system has the follot:ingproperties:

(1)

There

exists

a complete

set

of ordinary

eigenvectors

3'

_T/12_ = a-I

(2) _TD2, =

(3)

--'j_j

_j

N2

(4) If f(t)

ni(t)qi(t)'

=E

then

i=l
:;2
-f2(t)

If E2 = _,then

;"

Equation

=E

(2-56)

t=1

.'I2-"(_)'_"
* qi(t)

is reduced

(2-57)

to:

zI + _-i_izi
_. r "" + _izi
_
= qi(t

(2-58)

[
_

zl(O)= 1(0)= 0

2-17

(2-59)

. ORIGI?ZALPAGE IS

OF POORQUALI_
Equatlon

(2-59) is exactly

the same as the ith mode of the large system.

Hence, since

_u2(t) = Cz(t)

(2-60)

N2
u-2f(t)

"'i

"'j'_"

(2-61)

1=1

Using Equation

(2-52),

we have
N2
u_2t(t)

= S E_x._(j)zj(t).
j=l

(2-62)

= Sx.(t)
--1

.-. _2(t) = S_(t)

That

is,

the

solutian

of

It

solution
the

be noted
minors.

the

the

lover-order

higher-order

should

nonvanishing
mimic

of

behavior

system

Is

the

projection

that

there

the

exist,
there

?;tt;-order

in general
exist

for

several

system.

It

is

there

exists

Th,-ore.'n..s
eter

.-.odel__

parameter

seT..e sper'ia]

I
that

or

a_:d
can

conti:_uo_ls

ii

structure

en_>'are
mimic

that

ezaetly

s)'.qteL'i

to

thore
the

_mder

the

exist
be!_v.,ior

apprc>priate

each

"2th-order
not

that observations
at ,
.,, points in a Nth_order discrete
general,
permit uniq_e identification
of the Ntn-order
or

of

the

system.

Therefore,
of

(2-63)

surprising,

N_ --" ;,", several

models

that

therefore,

system do not.
system, unless

in
2;_ : Y

system.

finite-di.'v..ens[on
of

a higln-r-order
" ", _""

l_.'mped
lu..-.ped

param-

,,',v_t
0 (r}.,_ia.,
,_,. PAG"-13
OF POOR QUALff'Y
forces
-

the

to the

number

structure

of points

at only
at

which _'e excite

response,

_'e prejudice

uniquely.

In general,

"%2points,
model.

we can uniquely

C.

,
1

a snail

the

outcome

if we excite
identify

finite
the

number of points.
structure

of any attempt
a structure
only

By limiting

and observe

to identify
and observe

an R2th-order

the

the
its

lumped

structure

response

at

parameter

EFFECT OF HODELINGERRORS ON PREDICTED RESPOXSE

Given

the

System

(2-6_),

rL,_+ n;:+ Kx =

:
i

!
_

x(O)

It

Is semetir:es

f(t)

t _ T

t > T

(2-6")

= _(o) = o

more convenient

to write

the

equations

in lst-order

form.

\
t g(t)

(I_
_=Az+
dt

t _- T
t >T

z(0) = o
-.'here

..

A =

_:I_ID
[_.,t-ll:

.:..:l.
d

, '! t;t) 1
- I

C_ ['_"OR QU._.LtTY
Suppose now that we have an analytical

dz

Ih(t)

dt
--- = By +

modei of System

(2-65).

t zr

t > T

(2-66)

I (o) = 2
where

h(t)

B = A

-- g(t)

(2-67)

_e wish to know what errors are induced in the solution by modeling

errors

in A and g(t).

Let

w=z-v

(2-68)

thus

dw

d--t= Aw + (B - A)l + ]6(t)

w(0)

- _h(t)t

(2-69)

= 0

(2-70)

thLls

_(t)

exp

A(t

r)

J_O Hin(t'T)

B'[(

A)v(_)

g(:)
(

- h(_)

dr (2-70)
)]

thus

t_'_t; i : fXtn(t.:)

: e:.:F A('. - -)i.' rII(_ - A) : "'-,-(:')

i
i',
"I

"t 7 )

;,t-

;'__711

ORIGINALPAGZ-I_
OF POOR QUALITY"

(2-72)

i.
:_

l l_':p (Boll <

"a 1, _2 > o
- M2 e_., (-a2t)
]lexp(AOII-<'_IeP(-slt)

Let
L'
L

._! = Max (MI, H2)

a = Min (a 1, a 2)
(2-73)

d = sup
II_(o - b(t)ll
t

[:

,]

k = M_x s pils(t)
il,supilh(t)
l
t

Then

ilw(t_ll
- Iu-':lsupll,'(t)ll
+ -- p

(2-74)

where

:'

t = t if t <T
(2-75)
= Tif

:(t)

c::p

-.

t _T

B(t

- T) 12(:)

d_

(2--76)

ORIGINALPAGE IS
OF POOR QUALITY

._p.(t-t . iIz(t)ll <Msupllh(:)[l

__ [c
<_.,>

..

(2-77)
q

II,:,c,:)ll ,,,,_
(2-78)

There

1.

are

two cases

of special

interest.

Case (a) - Short Transient

l,oads

If aT << 1, then Equation

(2-78)

yields:

s.pil-_'(t)t[ -<(',"r)"kll_ - All + :m'

(2-79)

_,-"_T
'- OCU
t_,,..,,
th,.,,:,-,,,:,.,=s
,_,._,',d
k lib - ,",1'
i arL'

of

t, qt_tlL

i=portal,ce.

e{,.,_,,,th_erro_._!IB- ajl and Ilk(t) - !2(t)l] ar., of eq-,.] significance

and the

systt.-m l,arnmett.rs

tt_c forcing

2.

Case
If

(b) - Persli_tcnt
then

the po:;sible

any ',,:i.<hcr acct, racv

t!:an

Excitation

l'quation

(2-78)

s,.vil.(t)l[

"-"'

in many structuraU

"vrm -_'/a bco_.:es v,,rv

k' IB - Xi!

not be known with

functions.

,T ""- 1,

Since

ileed

dsst,nt's
eff,','ts

yield,;:

"

()

It.- ,xl'+ : a

'd'.,'lla--.ics probl_,;.-.s tE._. d.t._pin_; is sm..aiI,

l::re," comp,_r,.d tt- unity.

mu,.'!_gle/lter
of

(2-80)

ir'pt,:'t.m('_,

FtSt',_/lllel'.

r:::.:_ ':'," .!,': _:_t'd ;.{,.Ix .l .-.':.'!; !:i';h,':"

It: ti_i:; c;_sc,

thdn

"_ht" f_.rst

tb.m

te:-r-

th," st.covld t_ r.-n d; th_:i

Thu.'; '-," .-.,,e t_:.it u-!,h,r l', rsi:z:,,n:

.:.'ct'r.,c;"

_'a_"

:L,'

i.,:,'ip,,:

:_hows

cx,'ta-

}_;_::'t':,,_:_..

11

'

ORZG;NALPAGE IS
OF POORQUALITY
ls interesting
similar

D.

to note that Chen and Uada (Reference 2-3) established

result using

perturbational

analyses.

EKrECTS OF NECLECTTNC

IIICIIERMODES

let us tirst consider

the case of an Nth-order

which

the forces fit) are basically

discrete

System

(2-8) in

low frequency in nature

+ _ m + Ku
= f(t)
-(2-81)

u(O) = _(o) -- o
,

where

._I-1D and

_!-IK

co:mute.

Then

u(t) = E,(i)
i=l

_texp
z.(t)x

zi(t )

(2-83)

(-_=i_itt

=JO

- :))

a.-1

sin

a.l

(t

- ;)

qi(;)

d;
(2-84)

u-here

q(t)

= _Tf (t)

_IY4'

"

-_. = ,_. ,_
r_
.....-"
I
l
"i

-_ (1,

"_"i"i

',;)

O,RIO!NMPAGE 13
OF POOR QUALITY

If

the

small

frequenEy
compared

spectrum
to

all

of

but

_(t)

the

flf

contains

lowest

< wi

In this case it is reasonable

only

frequencies

eigenvalue

for

_i"

i > N , N

to approximate

_f

which

are

then

< N

(2-86)

the solution u(t) by the

truncated series,

N
-u(t) :

UT(t).=E

_(i)
zi(t)

(2-87)

i=l

If

u_

look

at

a typical

term

In

tlle

zj (t) =

re_ainder

.h'] (t
t

T)

of

the

qj (z)

series,

de

(2-S8)

hj (r) qj (t - 7)

Expanding

qj(t

T) in

zj (t)

a Taylor's

series

fot

= qj(t)

hj (7)

_.(t!/"t
+-_"2---]0

dr

about

- _j (t)

d _.

z = O, we

fot

711.J (7)

find

d-

._

:'h.(T)

d_ + "'"

(2-89/

ORIGINI_.PAG_IS
OF POORQUALITY

If _._.t > 1, then

.1 .1

(T)d_ --!2
thj

_t'chj(z)

"

dr --"2r_J3
_j

T2h.(r)

dz ;

-]

.1

(2-90)

4 '
2 -,j.3t_

Thus
I-

-<

(2-91)

-j

If

C ' (t) i
_jlqj
max

< 1

(2-92)

i.e., qj(t) is Io'-"

frequency coz:paredto ,xj,then

qj(t)
z.(t)j.

..... 2

,
'

J _ (:;

_;;

2-23

4-. l,

';)

(2-93)

-.

ORIGIN_ PAGEIS
OF POORQUALITY

In thts case _e =ay improve the approximate

by addfng the additfonal

..

terns

u(t)

solutionof

of Equatton

(2-93).

=E_C(J)zj(t)
j=l

Equatfon

(2-87)

'

+ Z
$(j)
i
tl +i

zj(t)

(2-94)

Let
qi(t)
= _

-(s)(t)

i _ (I

s)

(2-95)

Equation

(2-9_)

u(t)

_ay be rewritten,

_J)

(t) - z.

(t

j=l

(t)

(2-96)

j=l

:_o_

(t)

= T_f (t)

(2-97)

..

L #\J

j=!

:_ut

=
_TK_
. .

. .

:.z:
E
t="

[ "',.*

t - = .

( s)(t)

= -:-i f(t)

(2-99)
]
_

....

= t2Cs)(t)

(2-[01)

..

._ .....

,_ .. _._,_,._

_*,-_-,--_'_

_-_

.......

L-_ _---_7._,

,
I

t_

where

f(t)

_u(S)(t)

is

the

"static"

response

of

the

system

to

the

applied

forces

If we write
N

u(d)(t)

=E

1[ zi(t)

- z_S)(t)]

(2-102)

i=l

then

u(t)

= _u(d) (t)

+ _U(s)

Thus we see that the total response

terms; _(d)(t)
"static"

consists

response

whole structure

(2-103)

is approximated

by two separate

of the d)mamic response of the active modes less the

in these modes and u(S)(t)

to the applied

what im sometimes

(t)

called

forces f(t).

the "static" response

of the

These results are identical to

the mode acceleration

method.

As a measure of the error in neglecting the higher modes we have

II (t)li - lluT(t)il
n =

ilu(t)]]
--

(2-104)

This error tends to be smaller for the case of persistent excitation where
resonance

may occur,

and higher

for short transients where resonance has no

chance of occurring.

E.

ERRORS IN

EIG_;VALUES

From an analytical
""
[

are usually

the

discr;tization

AND E!CE'_I'ECTORS

point of vie'.-,errors

-_1 of modeling
resu_,
of the continuou._

to model the lower eigenva!ues

-

in eigenvalues

errors or the use of too crude a level of

system.

In practice,

and eigenvectors

;m a,_curac,-of five to ten perc_.nt, or better.

nr_re

accuratelv

mp_-:-ied

;t Pr::b
1.
,. "_. related_
c[-nva'_-:es

and

t<_ the

than

eit,:n'.'t-ctc.rs.

Ti:is

encountercc;_
.,'i;:je_

2-27

it is us,ally

of aerospace
Th_

the ci_7_nvectors t_',',:_h

diffic,:i/!_._

and elgenvectors

in

possible

structures "with

eigenva!uL.s are usual!v

this
tostin;;

r:a':

b,-. :_im.',17
.,.,a'

..-.casuring

t :.,';_l,, }:c _.rea_u__"_ in .-._-re deuni

_.

_7_

_ _ _-7 7

O_t_tNAL
PAGEiS
OF FOOR QUALITY
%..-

F.

ERRORS IN HODAL "FORCE" COEFFICI_,'TS

Errors

modal

in eigenvalues

"force"

and

coefficients;

eigenvectors

hou'ever,

of the first N =odes are Imo_

even

ference approximation
the discrete
9

Thus

..

if

elgenvalues

effects.

For example,
the curvature

Un =

and

on

the

eigenvectors

a central

dif-

of a beam using

of the beam.

3"u/_x" Is

approxir,ated by

Un+ I + Un_ 1 - 2Un

h

where

effect

in the modal force coefficients

may be used to calculate

displacements

a significant

the

exactly, errors

will still arise due to discretizatlon

t

have

u(xn) ,

xn

= nh,

and

h is

the

mesh

(2-105)

s:.'acing.

Now

-}

Un+l

+ Un-i
h2

2a n = ._-tt
9 + h122 3"u; + O(h 4)
_X3x _

(2-106)

Thus

Un+l + Un_ I

2u

- h2 9241

max

!_l

L_ax

If

u,x,.
A
I
iUn+l

* Un_1
- 2un ....h 2
.

iI
"

"_ i =-u I
_ 21
i3:.- I

,2_10 ,
"--u!

;;!

2..,2---

':< i

5 "_--/_ 2)

:z.:::.:

'_

-L.

(2-!0,9"!

OF POOR QUALITY
a.

Now

Xlh =

"
Im

the number of mesh points per wavelength"

2
..

_.x
2

.
l Un+

h2 1I_
_2 u I
+ Un.. 1l_x2i
- 2u n_ax
- h 2 _2u I{max

m
3N2

(2-110)

since

_2 u
oB = E z-_---_-

(2-111)

d_

The relative

error in the bending stress

is also given by Equation

(2-110).

Thus, the relative error in the bending stress increases rapidly as Nm, tlle
nunber of mesh points/wavelength,
is decreased.
Since we have shogm in
Section

II-B that it is theoretically

possible

to construct

discrete models

\
whose eigenvalues
ous system

and tbmt eigenvectors

the continuous
modal "force"
values

agree exactly with the first :g eigenvalues

are projections

of the eigenfunctlons

system, it shoulg not be too surprising

coefficients

is usually

of the continuof

that the accuracy of

lower than that of either

the eigen-

or the eigenvectors.

In addition
errors

to the dfscretization

arise because of modal spill-over

in obtaining

pure r_odal excitation,

not admit classical

detail

errors dlscussed

in Section

normal modes;

caused

abovr,

additional

by experimental

diffic_Itles

and the fact that the real structure may

these errors will be discussed

III of this report.

2_21o

in more

\
\

SECTION III
DYNAMIC TESTIXG

Though

it is possible,

to model the dynamic

accuracy,
without

using modern analytical

behavior

few engineers

of a structure

to any desired

would be happy to co=isslon

at least some limited

d)mamic

conser_'ative approach

are:

structure

it is usua!ly

accurately,

testing.

engineers

nonlinearities

require

the analytical

ment between measured

tunately,

values,

the engineer

model and so increase the precision

Thls is a vet'.,useful

First of all, it must be pointed

parameters

of the co_-plete structural

ture.

this approach

tive capabilities

A.

the

It c_n, however, be used to

frequencies,

damping

factors,

and

of these modes to the response of the struc-

Is useful

in improving

capabilities

structure,

the analytical

and predic-

it cannot help improve the

for new and unbuilt structurt.s.

._:.3
DAI.TESTIXG
As sho'-,_

i._g classical

,'-._4sho_'n
=.al

out that due to the non-

of modes observed, and hence to make better

for a given physical

and predictive

predic-

it cannot be used to identify

system.

estir.etes of the natura!

of the contributions

analytical

process,

of the finite numbers

While

of the analytical

technique and can yield good results if prop-

of the iden=iflcation

estimates

is left

that the test data be used to update

uniqueness

v_ode shapes

is happy; unfor-

is poor, as it frequently Is, the engineer

the analytical

obtain updated

testing to "qualify"

of testing yields good agree-

It has been suggested

erly applied.

For these reasons, most

of d)_a_le

In a quandry.

tions.

to do

effects such as geometric

the system.

If tile limited program

if the agreement

in theory to model the

too costly or too time consuming

in modeling

and predicted

degree of

a new space vehicle

(i) while it is possible

at least a limited program

model.

techniques,

The main reasons for this

so; (2) It is very easy to omit some significant

or material

and numerical

=:,Je5

in

in

Subsection

norr.Ji

II-A-3,

.--Odes

St,bs(-ction

II-A-4,

canno t be excited

_a'?:_-l'- ' (?:,..:,,rznce

2-2)

ar*_

discrete

capable

discrete
in

pure

of

and
being

systems
normal

!,a:_ ._!,oa._ that

if

=odvs.

continuou.';
e:-:cited

exhibitim;
Despite

tP._. d;c:pi--.g

in

systems

in

pure

exhibit-

nor:-._;l! m_.des.

nonclassical
this

nor-

fact,

a ,:tructure

i_

small

O_;-_INAL
PAGE I_
OF POOR QUALITY
and the elgenvalues
-

yell separated,

structure can be excited

In structural
classical normal
many aerospace

in rather good approximations

d)mamics,

modes,

it is c_only

assumed

h_ile this is seldom

structures

be excited, particularly

Consider

the lower modes of any viscously

to pure =odes.

that the system possesses

strictly

is often quite small,

damped

true, the damping

in

and so fairly pure nodes can

in the lower =odes.

the Nth-order

discrete

system

that exhibits

classical

normal

modes:

nu + D_ + Ku - f(t)
(3-i)

u_(to) - G(to) _ o
Let

f(t)

= C cos

_t,

t o "+ -_

(3-2)

Since Equation (3-I) has classical normal modes, the solution of Equation
(3-1) vith Equation (3-2) can be written

u(t)

',qlere_

is the modal r-atrix

R =

(2-12)

,/_ 2 exp
21 (-Jai)
+ (2,1r.i,_)2

_ere

l.t_
t

, ts

defined

in

Equation

(3-3)
m_trix:

(j,_t))

_nd R is the response

J
_

= Re(_>R@Tc exp

(2-19).

_-2

(3-4)

ORIGINALPAGE [$

OF POOR QUALITY
Then

If

_3= qkek

(3-7)

then
J

u(t)_

Condition

(3-7)

= (k)qk

cos

requires

(.:t

_k)

2 _ ,_2) 2 + (2_k_k_,)2.

that

C = qkM_e k

However, _

is unkno_

We observe

at the beginning

that if-

and well separated,

of the test.

_i are distinct

then iR(k)!; >> !R(i) I i = k; thus:

cos (_t-

Using Equation

(3-9)

_ _:k, _i << Vi and the eigenvalues

u(t) = (k)q k

This suggests

(3-8)

that an iterative

_k )
(3-10)

,f( ?
scheme can be used to generate

(2-6), we can define

an Iterative

scheme.

Uk(t) = ?e _k(tJ

_, (t) = #vP.
_;>:ptj,:t)
=k
q-k-I

ik(:> = -:,.._k(t)
--, co,_stant
"

2:.

"':'!P.

c,.

._--J

pure modes.

(3-11)

l-,

,.')

_9-_.)

O.,IG,d_,L
PAG,e "
|.!
o"

OF POOR QU,_d.I,'I_

Usir$

the

properties

of classical

normal

modes,

"

we have

_I,
TN @ = I

(3-13)

"

qk " Rqk-I

{3-14)

_here

_.

_ exp (-J'*')l

Ril

_( iz
2 _ _2)"
(2_,i;t._,2"
+

(3-15)

RI(

= O,

i #

Thus,

_I = Rq-o
%

%9

q2 = R_I = R-90
(3-t6)

q
-n

_n-1

If

(_2l - u2)+

.-:.''.,, [;

"_',:":_clect

r ,:'x9 (-j,',.;

(2_i.-.1_.)2

<< 1

such that
%

- 2 -:

Vi # ;

(3-17)

 ._,_ "e_

,R'_"..IAL P ..... _"

OF _C'C/_ 'al';'L:_"
:',
then

(3-!9)

liu R = Ei
N-_

Where

E.l are

nonn_ative,

definite

matrices

sat.fsfy

tile

following

(3-20)

EiEj = Ei_lj

Thus,

using

Equation

(3-16),

-'e

}lave

lira q-X = liw R"q--O= qoiei

N-_
N-'_

Using

Equations

(3-21) and

property:

(3-Ii), _'e

(3-21)

find

lim Uk(t) = lim

exp (j_t)
k -_
k_ _Rqk-I

.(l)

e::n [j(,_'t
"_

".

lira !lk(t)

- ai)l

O---)

= Re li.-. __{(t)

(3-23)
= :(i)

cos
._. -

Thu_. l!,--,
iterati'.'e
process:

(=t - a i)

:)

+ (2-: 1 " I -_"

cenverges :o .l ._olutlonproI)ort
i,.zalto the pure

-_dv :
, ."rt,n
"_hichthe _atural fr_-<:.uenc>'
.: and IL,"dar:vir.<
::ar_::_:er
'i
:i)
r_lv b.." obt,:ined
in additlt,n
to :

It
L -

sheuld

System

be noted

(2-8)

is also

tlna,_us

or discrete,

Iterative

schL_e,

in passing

that

true

of the

continuous

that

exhibits

classical

like

that

above,

uhat

which

has

been

System

said

for

(2-1).

normal

the discrete

Ally systt_-a,

modes lends

cotaverges

to a pure

con-

Itself

to an

mode.

,..

B.

ERRORS IX ,HODALTESTING

1.

Nonclassical
Even if

_he system, under

discrete

syste-n

possess

classical

of forcing
cut,

if

it

small

tures,

modes.

that

excite

damping

eigenvalues

It:pure

in :ill

syste:=

systt_

requires

proportiona_

._o that
t3-1"_,

.!t|lit,re
vid,'d

this
that

seldom
t2.1tes;

we

case

a-':.-I at

test

a'ltTa

no choice

well
the

as often

separattq,
system

happens

even

pointed

da=ping

is

in rea!

struc-

pure

modes

relatively

modes.

can be regard,,d

of a pure

each

in the

by a force

mass

r:odal

system

mass matrix

prek'ise]y,

an adequate

we often

e\'t'll

,'_.ort-

have

mass.

In

can he used

to

excitatien

approxinati{m
task

_21ile we will

.tth'qu.ltely

availaable

diif;.,rult

=ode

";ystem is kno_,-n and pro-

transducers

d i:;cr_:te

tllat

to each mass.

of a Ct_lltillu,3tlS syst','2, =odal

least

of

technique

of the

ft, rvt.s

hay<. v-force

be excited

displaccmcnt

an iterativc

tile _-eans to apply

seldc_

as an N-degree-of-freedo.'a

good e.<;tictmduct

to

should
t_

tilt,

a!st_ be
i,'lllOll,;

co:it

to acct,.-,plish,

}_:"rO F.q

o: lrc_.','.-a
_':

exists

nor_..al re..odes, excitation

As,_;'-'ui:_;:. as in Sub,section,,;

7"

will

classica!

the

05 r.'tlrccs----dll

._'.t'/1: u F _.7.<_ t

d,.,:r,,,,

system

As previously

If

to excite

lowest

sho_,-n that

provided

_,e have

In the

di-._ribution

spaced,

mass alld to the

kno'_ t.-._" mass raatrix

t'O.'ItillllOUS,

_.

that

",'e have

end,

]lOWt,ver.

tt,:;t.

mode.,;.

may be excited.

very

under

possessing

of vibration

}_quation

there

and tile eigenvalues

In practice
the

the

}',,-dal Excitation

Even if the
dis.:rete

but

that

theoretically,
normal

closely

it may be impossible

2.

pure

Is small

as an N-degree-of-freedom

in practice,

Thus,

modes of vibration

and the

of vibrations

can be regarded

unlikely

norr_al

tilt: system
pure

test

is highly

functions

relatively
not

Norraal Hodes

:-'-'-::tiFi'1_,

III-_;-1

and

is .t r;,'..,l r.a_del of the

',};t"

ilia_)].ltt'P_-t'r]t:,

,:_

I!l-B--'.

s'..s_t.n0
"1

i _!_:][:_

th,i_

tht.re

.';till

{tie

t+',l_'2i

til__,di:;crc'_rt.2ains
[F_."!ilt':i,''."

r.ht.
.

711_,,_

g displace=_'ut

transdvcers

N0 - X transducers
consur_ing procedure

there

errors

in phase

for

a test;

aud move them around

and greatly
increases

if ix'_ have Y transducers

modes,

available

still

available,

exists

the

_easurements,

which

it

is possible

the structure,
but this
tile cost of tile test.

and even

question

true,

if _,'e could

of t-.<,asurement
are

notoriously

is a tl--e
However, even

excite

error,

to use

pure

no_ll

particularly

difficult

to make with

accuracy.

4.

Effects

of Diseretization

At.r,-spat'i" structure:;

"

very-high-order
ing,

discrete

normal

if the

m.,des,

eigenvalue

tile

classes

erigina!

o_ excitation

the

structure.

tit:at the

model

is no= one

structure

we shall

(2-43),

4.,:-discrete,

very

to one With
testing.

we assign

let

the

Let
to it

.t:ld apply

t:(tl

be given

classical

model whose

or tile original

shown that

model exactly

us

,_; c:ordlnates
l'orcea

b':

only

certain
that

_t should

structure.
suppose

or

for

mimicked

properties,

origillill

test-

it was

of tile eigen[une_ions

useful

best,

high-order

III-B

discrete

at

and

exhibited

.R e:.genvalues

of the

these

m.lkemeasurements

l,n Equation

In Suhsection

It was furtiler

response

or,

of aualysis

an .R-degree

first

system.

in m_dal

a,nd that

th_

llattlre_

or _2ondellse the

system-

were projeetious

Despite

shows up immediately

structure

eonti'mous

with

|n

for purposes

to construct

exactly

the

origina!

tim,ous

system,

eigenveetors

of

t, entinuotlS

however,

eOlltinuolls

it was possible

s.x'stem and whose

always

a m_inageable

original

coincided

eienve.-tors

almost

s\-stcms;

sv-;tem to obtain

sho'-'n th.lt

tt:at

are

we .-.,,_st diseretime

discrete

or Conden.-'at|on

that
-1
X.

Ibis

of tht'
be clear

filet

we have

a con-

i .. (1,

N_, and

at these

.N points.

CF p_5_ ..-L,,L:.

The points

of application

of these forces correspond

x. i c (I, N) of the continuous

system of Equation

to the points

(2-])

Thus

N
f(x,t) = EtJi(t)
i=l

-'- qi(t)

5(_ - xi)

Xi(x_)f(x,t)
dx .

(3-25)

(3-26)

:1
= EXi(xj)
j=l
With Equations

(2-39)

and

(2-40),

if

_.(t)j

i c (1,

N),

(3-27)

then

q_

q(t)

=_rf(t)

(3-28)

bd_

}'(t)
_
= }_q(t)

(3-29)

thus

_(t)

.'.
Dr,,;idt.d

!_, ;"

= _T_(t)

_.._.(t)
= _(_)

= _)T:lq(t)

(3-30)

7_rJ.-.lL ........

"-i: 7-.'_-- -_ ...............

_le note

OF FOORQUALFW

that

?;
_1_(t) = Z

X_ (xj)_j (t)

(3-31)

j=l
lc

IN+

1, =1

In general,

(3-32)

q.(t)
#0
k
For

example,

(3-32)

if

f!(t) = qk(t)Sk

k [I, Thus

q_(t)

= qt(t)._k

t,k

[1,

S],

but

:i

qt(t) =

X_(xj)fj(t)

_ > N

(3-3_,)

j=i

# 0 in general

Thus,
there

for

exist

the.

than

X,

response

of

these

t.-.,:de,

to

tha_

then

of
of

the

one

modes,
the

if

the

modes
rh,:

k th .-.._de.
of

only .

"ali_ed"

in
is

1: th mode

h,,'_'e'-er,

are

of

is

is

c,_:p,=r_-d to
*'>:citation

th(,

n_.,.i,_,s, seri,,_._s

excited.

excited.

eigenw_lues

small

frt,cu_'r:cy

If,

the

" ?;, which

separaticm

"al!ased"

particular1_."

frequency

N modes,

higher-order

smaller

h th

first

fr_,Ivency
errors

If
usually
the
is

of
can

However,
k

is

such

much
that

response
clo._t,

to

excitation
result.

the

of
the

the
natur,ii

is

c!oye

C.

OTHERID_IFICATIOY TECI_IQUES
Since modal testing

classically
reasonable

damped

as an identification

technique

systems, or at least systems with small damping,

to ask if there are other identification

do a better

Job.

is restricted

techniques

There _xists a variety of identification

parameter and nonpara=eter;

however,

if one wishes

stiffness,

and damping matrices,

the number

of points N 2 at which measurements

to
it is

that could

techniques,

to identify

both

the mass,

one is faced with a fundamental

limitation:

are _ade must, in general,

be

equal to N I, the ntmber of degrees of freedom of tilestructure.

Unless this
is done, the solutions obtained are not unique.
Since, as already pointed
out, aerospace

structures

are almost always centinuous,

large number

of degrees of freedom, unique

is virtually

impossible.

structures
fication

identification

One has to conclude

with small damping,

nodal testing

technique as any available.

or at least bare a

therefore
is probably

of tilestructure
that, at least for
as good an identi-

SEc'rZON !V
-

CONCLUSIONS,\X_ RECO._i_I_XD,\TIONS

The
and
the

A.

object

testing
results.

of

of

this

report

aerospace

has

been

structures

and

to

exanlne

tile

proble.'_s

the

dlfficulti,__s

of

have emerged

from this

study:

of

analyses

correlat,.'ng

CONC.i.US
I o,_s
The following

(1)

Hodal

conclu,;ions

testing

(and

as a =ett:od
stiffness

for

art\" other

uniquely

,"_trtces

in nature,

provide

impr,'vtng

of

tool

response.

for obtainin_

vectors

of

the

correlated
crete

models

analysis

Analv:ical

teclmiques

capable

with

il higLer

degrt:e

accuracy

cocff_cient
._tr,'g'at's

can
.ITi.!

than

_-'ill

,",it,-

._:'._ivtic.tl

lt_r,zt,

the

1.'_I. 1

prt'dicti,,r:
i,r,'di_:t,mr-

,'ff,,r:

be ac,-,,p', ed.

of accuracy

dynanic
that

with

Accurate

l't?-I_!'-'."_"

liP,

o:" e-gtmvnlu,.s
oi

,-'.nd c_.7,.i::_ ,. ,,i

:.t.t','._-,

tl:;i_;

with

t.'hieh

]eve[

ft_r dis-

useful

of

the

a higher
_,d,tl

force
of

dlSdl't'L[-.lt[Oll

,lrt-

,alld

t_
cm_

pr,'diction

fortt.s

fit_._'r

in the

structures

and ei._:c,w,:ctot's.
.Itld

can be

can e|ge:l-

analytical
dr

and ,'i,_ien-

eigenvalues

than

call be predicted

be prt.dicted.

useful

tile basis

be extremely

is cleat"

til;m ".hat accuracy

deter-

The,-ae measures

of =odeling
It

anti th,l_, t'lgenvt.dtors

of

.ldd'-'t',,n.:l

arc

dt'_-ree of ac,'ttracv.

be pr,_ilcted

de_;re

contr,_l.

eigenvalt:es

or provide
nay

cannot

for

is an extrer._,ly

of the

which

a.').!

_,'hich are

technique._

structure.

results,

structure,

stability

amy desired

measures
of the

analytical

of the

of

vectors,

accurate

and

test int: thereby

analytical

,_I_da! testing

lo_.'er =odes

with

r:odal

be used

damping,

structures,

,rod the
the

ca!mot

tilt, mass,

aerospace

usua! [v cont inu,-',ls

a reans

technique)

dett.r=ining

of rt.al

ralni_ g dynamics

(2)

indeed

! illt._

If

l-t.c.uir_,Q.,

:_t..-.]_.:;

accutilt.

flu:eL

(3)

Accuracy

of modeling

modelln}.:

"Given

accurate

need

response7"
ll-C,

topic

and

the

data;

resonance,

degree
tent

for

_,'lth

data

has

in all

only

be to obtain

finite

acceptable

_,'lth at

sore

mathematical
accuracy,

lu.v

accuracy

in tile

length

in Sub-

an.-,a,'er depends

on tile

tilt- tory.tel need

m_t be any more accurate

persistent

the model

of accuracy

question

s.'a:; dealt

the

transients,

input

the

tile raodel

For short

of

B.

that

This

section

is a central

inputs

nust

than

that

be specified

tho

input

the

results

obtained

tile

present

study,

data.

form of excitation.

create

the

than

possibility

with

a mu'h

higher

These

results

are

col:sis-

by {:ben and Igada (Reference

2-]).

RECO._I}._ DAI"I O:S

As a rest,

(1)

It

of

To deter'aine

stresses

priatt,

of discretization

level

finer

than

and

would be used

t_.'o recora=endation:;

forces

in aerospace
mtst

for

ernergt.:

systems,

be used,

determininR

even

if

elgenvalues

an approthis

is much

and

eigenvecters.

(2)

ghile

it

is virtually

uniquely

i_.passible

from the

result

check

on the

a valuable

to "idcatifv"

of raoaal testing,
analytical

provide

an accurate,

discrete

studies

of stability

,rod contr_,l.

tional

virtue

that

mimv physic/ll
spacecraft

such
tilt'

then

!aver

friction

data

_,'hich

,_n

bec,-a,:s

tnOtlt'It'_'.'

part

of
Oil

to

an_]

bast*

all

am addi-

t_f fuel

in a spitzning

to model

accurately

pIay

an;,.Ivttca[

il central

syslt.._
model.

role.

In
pr,,vide

Thi:: :....,dal t,_'sting

.d, il:i'_it,,

[,'St.

since

can t-astl)

;:._&-lit_g ".echnique

t}l*- ' ,'P,t',I;*1

in

sloshing

are model,-d
O:

has

use

There, are

phv:;I,'at

and p.irce!_ of the

b,!Si.'_

for

to

fullest.

dissipati_:n

t,f the

tk," .';tr,,ct,,r,,
'*._la'

to the

difficult

provide

and can be used

.*lodal testiug

such as the

rather

a t-a_.d, .*_.,tsda[teats

:_,,r:e paris
.LI'*"

are

structure

tests

of tl|e-systcm

be exploited

phenorat.na,

tlmt

boundary

should

such

r:ethod

model

the

in '-'hich

.:nd _,,'-e parts

REF ERI-2qCES

2-1.

Caughey,

T.

Damped

l.inear

Sept.

1965,

In Da=ped
June

-4

O'Kelly,
Caltcch

2-3.

Chen

and

tion

of

Oune

1975,

J.

and

D\'namic
pp.

J.,

.I.A.-vI.
Caughey,

Systems"

J.

"Classical
Hech.
T.
Appl.

K.

:;o,--mal
Vol.

39,

Hodes
No.

"Classical

Hech.

Vol.

in

3,

Normal
27

No.

Modes
2,

269-271.

H. E.
1961.

C.,

E.

also

Dynamic

pp.

H.

Systems"

583-588;

Linear

3.

O'Ke!ly,

Dynamic
pp.

1960

2-2.

K.,

Normal

_ada,

Hodes

B.K.

Structures"

in D_----ved Svstera.

Cr.teria

for

AppI.

Yech.

.I.

671-477.

;'--i

Englneerts

Analysis

- Test

Vol..&2

No.. o

Thesis

Correla-

APP_;DIX

OF pOOR QUAUTY

.-"
Consider

the

probIem

_2u + 2g ?u

?2u + if(x,

_t 2 .

_x2

_t

u(O, t)

t)

0 < x<

= u(1, t) = 0

(a-l)

u(x, o) = _(x,o) = o
Tile eigenvalues

and etgenfunetions

of Equation

2
_i = (1_)2

(A-!)

t c [I,

are

_,)

(^-2)
X(i)(x)= /2 sin
If

i_X

we write

u(x,

t) =E

:=I

Zi(t)x(i)(x)

(A-3)

then

o-

Zi + 2mZi + _iZi

= qi (t)

(,\-4)

_.'he re

u.(0)
= 6.(0)
= o
I
!
and

qi(t)

f(::,
I

t)

:':

(::)

(::_:

(A-5)

C_u:N_L

oF FOC;_
qUALrTY
Di.qcretization of Equa.tfon (A-k)

If

we use

matrix finite

central

difference

ele=ent spatia!

ui +

spatial

discretization

dlscretization,

+ (N)-

or constant

mask

Equation (A-l) bec_.es

ui - Ui+l - ui-i

where

ui(t)

= u(ih,

t)

_i(t)

h = ._, t c

= f(lh,

t)

(A-7)

, (_-I

The eigenvalues and eigenvectors of Equation (A-6) are

r2

i_

'_i = 4N2 sin2--2N

(a-S)
_j
.(1)

,. sin

,N

Let

ii)
_ =

_(i)

_(1)

_(2)

!(N-l)

'_i)
l
;-i/
C,\-9)

-1

OF pOc.a_UALrrY,
m

Let

u ='I'

"1
Z =

Z2i__Ii

(A-10)

where
.

"'"

Zi + 2_ki + "qlZt = (II(t)

q_'(t)=

(A-If)

= ?(t)
q:;-I

:;-I

,(J)_r.
(t)
j=t
Co=parisons

of Equations

(A-8)

and

-j
.(i)

'i

_ *i

o,:
-* sin

(A-9) shows that

X(i)(j..)

[-:
_.

(A-i2)
i::
sin
Z':
__-

(a-13)

-/

OF PoORQUt.,.LITY
Thus,

in

projections

this
of

case,

the

the

first

eigenveetors

of

the

N eigenfunc:l_ns

approx _ir_ting
of

the

system

continuous

For the sake of illustration, let us use N = 4.

are

system.

Equation (A-13) then

gives

8
From Equation

(A-14)

cI = 0.0255

2 - 0.0997

(A-15)

c3 -0.2158
Tnis shows clearly

how the errors

is _.Titten in matrix

I-_crease with mode order.

If Equation

(A-6)

form,

Mu + 2.'.q4u
_ Ku = g(t)

K=

1:i
i]
21

(A-16)

,;,,:} = -f . :- (:_,

I-(!,

(A-17)

'_- -'._i_
17-"_
_"_.i_f
,..L_
_ _.i -"-ii,A.i_..
_,.7__-"L_J-

_ ..
-_-_'_

_.,_......
_:_,_ _'_.'_:_'r_--:
_-_ "_._'_'_,_,_'_'t_'_'_"_

OF POOh QUAL[? ._

Let us now use Theorem I to construct a 3rd-or_er

same eig_values

of the continuous

system.

syst_

having

the

Let us select

slni!

(I)=

sin _ 4

(A-19)

1
Eerice

-2

(A-20)

-2
Using

Equation

(2-4!). ve have
I

HO

', .

=_I

DO=_Z
(.,,-2')

Ko --=

I-_,
."_
.

20
-8;2

-S'
I_3J

OF pOOR QU_UT
9

.-

In this case
difference.

_ have
Thus,

chosen

-.

a" so that

H i_ the

[11.103-6.9785

same as that

obtained

by finite

1.2337]

and

K_

Comparison

of Equations

the same M and D matrices

of Equations

_---

(!)

(A-23)

and

K0 is a fu!l

(A-17)
(they

(A-27)

matrix,

and

(A-21)

(A-23)

show that

were so constructed);
show ccnsiderable

while

both

however,

=_dels

have

the K _atriees

differences:

K is a tridiagonal

Jacobi

=atrix.

(2) The =agnituiesof the elementsof the two K matrices:::emarkedly

different.

Thus, we see td_t if modal testing is used to identify a 3r-order model

of the continuous sys:em, the matrices of Equation (A-17) _:ouldresult.

If the

stiffness r-atrixK0 is compared with K, the mahrix obtained from finite difference or finite element discretization, we see that they're not even close.
Hence, we see that a stiffness matrix obtained from mtxlaltestin_ cannot be
used to check that obtained from systematic analytic reducticn techniques
such as finite difference or finite element.

A-_

_:

OF FO0_ QUALITY
"_-

To

illuntrate

aliasing,

suppose

that

g(t)

in

Equation

(A-16)

is

given

by

g_(t )'= ,'_

whichwill

-_

excite

Using

only

Equations

qi(t)

tile

first

(3-25),

= _'2

=2

aode

(3-26),

sin

i_

sin_-

_;-. :

cos -'t

of

oscillation

and

(3-27),

+ t_ sin

[,

_- cos

(A-2_)

Equation

(A-16).

we have

--+
i-

of

+ sin

cos

cos

,=t

(A-25)

a't

(A-26)

Thus

qt(t)

- 0 unless

i = 8k

(A-27)

k = O, 1, 2 "'"

qSk__lit)

= (!)

4 cos -t

(A-2S)

Thus

ql(t),
If
'"

": J -: 4:,

0 " :z. - ,q-.

str,,n_;lv

the

cxcitt.d.

only
I st

q7(t),

qg(t)

mc_'e wi!l

be

the

first

arid

7 th mode

of

the

etc.
strougly

con._ipuous

# 0
e,:cited.
structure

If,

i_L-:-ever,

ca:_ be

For structures with well separated

restricted
a serious

to the ban@width
problem.

have rather closely

a more serious

and excitation

of the first N modes, aliasing

Some structures,

_uch as shell-like

spaced eigenvalues

problem.

elgenvalues

_ile

does not present

structures,

tend to

and, in this case, aliasing becomes

this discussion

was restricted

for sim_pllclty

to the case N = 4, the same features show up for all values of N.

LANGLEY RESEARCH CENTER

3 1176 01327 5772

DO NOT REMOVE SLIP FROM MATERIAL

Deleteyour name from this slipwhen returningmaterial

to the library.
NAME

NASA Langley (Rev. Dec. 1991)

DATE

MS

RIAD N-75

_
I
jl

i'" I