master's thesis defense improved system identification for ... · = solving for the x vector...

31
Improved System Identification for Aeroservoelastic Predictions 10:30 am 2 July 2003 MAE Conference Room 218 Engineering North School of Mechanical and Aerospace Engineering Oklahoma State University Abstract Master's Thesis Defense Charles Robert O'Neill Modern high performance aerospace vehicles are particularly susceptible to destructive fluid-structure interactions. Accurate and timely aerodynamic predictions are needed for efficient vehicle design and evaluation. System identification offers an efficient and powerful prediction methodology by substituting a trained mathematical system model for the actual aerodynamic system. Coupling the system model with structural and control systems allows for fast and intuitive vehicle analysis. The challenge becomes determining a system model that accurately represents the dominant fluid-flow physics. This thesis investigated linear aerodynamic system identification for aeroservoelastic predictions based on Computational Fluid Dynamics (CFD) flow predictions. Time and Location Presented by Presentation Preview [] [] = = + = 1 0 1 ) ( ) ( ) ( nb i i na i i i k x B i k y A k y System Model Excitation Signals Aeroelastic Sensitivity Studies 1.141 1.072 0.499 0.678 0.900 0.960 |z|=1 Density Sweep Eigenvalues Mach Fresnel Chirp

Upload: others

Post on 22-May-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Improved System Identification for Aeroservoelastic

Predictions

10:30 am2 July 2003

MAE Conference Room218 Engineering North

School of Mechanical and Aerospace EngineeringOklahoma State University

Abstract

Master's Thesis Defense

Charles Robert O'Neill

Modern high performance aerospace vehicles are particularly susceptible to destructive fluid-structure interactions. Accurate and timely aerodynamic predictions are needed for efficient vehicle design and evaluation. System identification offers an efficient and powerful prediction methodology by substituting a trained mathematical system model for the actual aerodynamic system. Coupling the system model with structural and control systems allows for fast and intuitive vehicle analysis. The challenge becomes determining a system model that accurately represents the dominant fluid-flow physics. This thesis investigated linear aerodynamic system identification for aeroservoelastic predictions based on Computational Fluid Dynamics (CFD) flow predictions.

Time and Location

Presented by

Presentation Preview

[ ] [ ]∑∑−

==

−+−=1

01)()()(

nb

ii

na

ii ikxBikyAky

System Model

Excitation Signals

Aeroelastic Sensitivity Studies

1.141 1.0720.499

0.6780.900

0.960

|z|=1

Density Sweep Eigenvalues

Mach

Fresnel Chirp

Page 2: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Imp

rove

d S

yste

m Id

enti

ficat

ion

fo

r Aer

ose

rvo

elas

tic

Pred

icti

on

s

Sch

oo

l of M

ech

anic

al a

nd

Aer

osp

ace

Eng

inee

rin

gO

klah

om

a St

ate

Un

iver

sity

Mas

ter's

Th

esis

Def

ense

Ch

arle

s Ro

ber

t O

'Nei

llPr

esen

ted

by

2 Ju

ly 2

003

Page 3: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Aer

ose

rvo

elas

tici

ty Structure

Aerodynamics

Controls

Aeroelasticity

Aeroservoelasticity

ΣΣ

Input

Output

Aer

ose

rvo

elas

tici

ty is

a c

om

bin

atio

n o

f str

uct

ura

l, ae

rod

ynam

ics

and

co

ntr

ols

sys

tem

s.

Page 4: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Stru

ctu

ral S

yste

m

[]

[]

[]

Fq

Kq

Cq

M=

++

&&&

6 m

od

e p

late

exa

mp

le

[]

[]

[]

[]

)(

)(

)(

)(

)(

)1(

kF

Dk

xC

kq

kF

Hk

xG

kx

ss

s

ss

ss

+=

+=

+

The

dis

cret

e ti

me

stru

ctu

ral m

od

el is

th

e fo

llow

ing

:

The

gen

eric

st

ruct

ura

l g

ove

rnin

g

equ

atio

n

con

tain

s m

ass,

d

amp

ing

, st

iffn

ess

and

an

ex

tern

al

forc

ing

fu

nct

ion

.Gov

ern

ing

Eq

uat

ion

The

stru

ctu

ral

syst

em p

rop

erti

es a

re d

eter

min

ed

thro

ug

h d

eco

mp

osi

ng

th

e ov

eral

l st

ruct

ure

in

to

mo

des

hap

es a

nd

freq

uen

cies

.

Page 5: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Co

ntr

ols

Sys

tem

Det

erm

inin

g a

use

ful

con

tro

ls m

od

el

is c

on

cep

tual

ly s

trai

gh

tfo

rwar

d. T

he

exac

t im

ple

men

tati

on

met

ho

do

log

y ca

n v

ary.

A

gen

eral

ized

co

ntr

ol

law

fo

llow

s:

Structure

Aerodynamics

Controls

Aeroelasticity

Aeroservoelasticity

ΣΣ

Input

Output

[] x

Kr

The

con

tro

l sys

tem

pro

per

ties

are

det

erm

ined

th

rou

gh

th

e d

esig

ner

's

spec

ific

sy

stem

p

erfo

rman

ce

crit

eria

. A

co

ntr

ol

syst

em

mo

difi

es t

he

over

all s

yste

m re

spo

nse

.

Page 6: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Ob

ject

ive

The

ob

ject

ive

is t

o d

evel

op

an

im

pro

ved

sys

tem

id

enti

ficat

ion

met

ho

do

log

y fo

r ae

rose

rvo

elas

tic

pre

dic

tio

ns.

Su

bst

itu

tin

g

the

tr

ain

ed

ae

rod

ynam

ic

syst

em

mo

del

fo

r th

e ac

tual

ae

rod

ynam

ic s

yste

m w

ill a

llow

fo

r an

eff

icie

nt

and

po

wer

ful p

red

icti

on

met

ho

do

log

y.

The

Pro

ble

mTr

adit

ion

al f

ree

resp

on

se a

nal

ysis

of

aero

spac

e ve

hic

les

wit

h

Co

mp

uta

tio

nal

Fl

uid

D

ynam

ics

req

uir

es s

ign

ifica

nt

com

pu

tin

g p

ow

er a

nd

do

es

no

t al

low

fo

r in

tuit

ive

or

qu

ick

anal

ysis

an

d

des

ign

.

Page 7: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Dec

om

po

se in

to 4

are

as

Aer

od

ynam

icSy

stem

Mo

del

Trai

nin

gM

eth

od

Exci

tati

on

Sig

nal

Perf

orm

ance

Cri

teri

a

Ho

w to

rep

rese

nt

the

aero

dyn

amic

sys

tem

Ho

w to

det

erm

ine

the

aero

dyn

amic

sys

tem

par

amet

ers

fro

m ra

w d

ata

Ho

w to

exc

ite

the

do

min

ate

aero

dyn

amic

s

Ho

w to

eva

luat

e an

d s

elec

t a

syst

em m

od

el

Page 8: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Un

stea

dy

Aer

od

ynam

ics

No

gen

eral

clo

sed

fo

rm s

olu

tio

n e

xist

s fo

r u

nst

ead

y ae

rod

ynam

ics.

Two

cl

assi

cal

un

stea

dy

aero

dyn

amic

exp

ress

ion

s re

sult

fro

m a

ssu

min

g a

n

inco

mp

ress

ible

, inv

isci

d f

low

wit

h s

imp

le m

oti

on

s an

d s

imp

le g

eom

etry

. Th

e W

agn

er, T

heo

do

rsen

an

d w

ave-

equ

atio

n e

xpre

ssio

ns

allo

w i

nsi

gh

ts

into

rep

rese

nti

ng

un

stea

dy

aero

dyn

amic

01

23

45

67

89

10

0

0.2

0.4

0.6

0.81

Ch

ord

s T

rave

led

Unsteady Lift Ratio

Ma

ch 0

.3

Ma

ch 0

.8

Ste

ad

y S

tate

Wa

gn

er

Wag

ner

Ste

p In

pu

t

Re

al C

(k)

00

.10

.20

.30

.40

.50

.60

.70

.80

.91

0.20.10

0.1

0.1

0.2

0.3

0.5

1

24

∞0

Th

eo

do

rse

n F

un

ctio

n,

C(k

)

Re

du

ced

Fre

qu

en

cy

Uck

2ω=

Ima

gin

ary

C(k

)

Theo

do

rsen

Har

mo

nic

Mo

tio

nL

=πρb2

(h+

Uα−

baα)+

2πbρ

U(h

+U

α−

baα)C

(k)

Wav

e Eq

uat

ion

No

Mo

tio

nSu

bso

nic

Sup

erso

nic

D2

Dt2

=a2 0∇2

Page 9: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Aer

od

ynam

ic M

od

elin

g R

equ

irem

ents

Co

nsi

sten

t B

ou

nd

ary

Co

nd

itio

ns

Inp

ut-

Ou

tpu

t D

ynam

ics

Star

tin

g a

nd

En

din

g C

on

dit

ion

s

Acc

ura

cy n

ear s

yste

m s

tab

ility

po

int

Cap

ture

s d

iffer

ent

tim

e sc

ales

Mo

tio

n li

mit

atio

ns

to re

mai

n in

lin

ear r

egio

n

Page 10: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

[](

)[]

∑∑

− ==

−+

−=

1 01

)(

)(

nb ii

na ii

ik

qB

ik

fA

kf

Inte

rnal

Resp

onse

Inpu

t Res

pons

e

Aer

od

ynam

ic R

epre

sen

tati

on

s

f(n

)+

...+

f+

f+

f+

f(d

elays)

=x

(n)+

...+

x+

x+

x+

x(d

elays)

+C

Usi

ng

th

e p

revi

ou

s u

nst

ead

y ae

rod

ynam

ic

resu

lts

and

m

od

elin

g

req

uir

emen

ts, t

he

follo

win

g g

ener

ic a

ero

dyn

amic

fun

ctio

n is

pro

po

sed

.

No

w, a

sp

ecifi

c sy

stem

mo

del

nee

ds

to b

e se

lect

ed. T

he

bes

t ca

nd

idat

es

wer

e th

e in

dic

ial r

esp

on

se m

od

el a

nd

th

e A

RMA

mo

del

.

Ind

icia

l Res

po

nse

ARM

A M

od

elTh

e A

uto

Reg

ress

ive

Mov

ing

Ave

rag

e (A

RMA

) mo

del

co

nsi

sts

of a

n in

pu

t-o

utp

ut

exp

ress

ion

w

ith

in

tern

al

and

in

pu

t re

spo

nse

s.

Ind

ivid

ual

co

effic

ien

ts A

i an

d B

i det

erm

ine

the

syst

em p

aram

eter

s. Th

e A

RMA

mo

del

d

irec

tly

corr

esp

on

ds

to a

dis

cret

e-ti

me

OD

E re

pre

sen

tati

on

.

The

ind

icia

l res

po

nse

co

nsi

sts

of

a st

ep r

esp

on

se. T

he

syst

em r

esp

on

se is

d

eter

min

ed t

hro

ug

h c

onv

olu

tio

n. T

he

syst

em p

aram

eter

s ar

e d

eter

min

ed

by

the

step

re

spo

nse

fu

nct

ion

g

(t).

Det

erm

inin

g

g(t

) b

eco

mes

co

mp

licat

ed w

hen

th

e b

ou

nd

ary

con

dit

ion

s ar

e co

up

led

.

y(t

)=

∫ t 0

g(t−

τ)u

(t)d

τ

Page 11: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Trai

nin

g M

eth

od

olo

gy

The

trai

nin

g m

eth

od

red

uce

s th

e ra

w a

ero

dyn

amic

tim

e h

isto

ries

to

usa

ble

aer

od

ynam

ic s

yste

m m

od

els.

Re

ad

xn.d

at

De

tre

nd

SV

D o

ne

ach

mo

de

Pa

cka

ge

into

.m

dl

FA

XN

FA

OF

FS

ET

S

A, B

This

is a

lin

ear e

qu

atio

n o

f th

e fo

rm:

Exp

ress

ing

th

e in

pu

ts a

nd

ou

tpu

ts in

th

e A

RMA

sys

tem

form

yie

lds:

[]

bx

Ar

r=

Solv

ing

fo

r th

e x

vect

or

yiel

ds

the

AR

MA

mo

del

co

effi

cen

ts.

Sin

gu

lar

valu

e d

eco

mp

osi

tio

n (

SVD

) is

use

d t

o s

olv

e th

is l

inea

r le

ast

squ

ares

alg

ebra

ic s

yste

m. S

VD

is

pre

ferr

ed b

ecau

se o

f it

s ro

bu

stn

ess

and

its

abili

ity

to s

olv

e ov

erd

eter

min

ed s

yste

ms.}

Out

put

Cur

rent

In

puts

Cur

rent

and

Past

nr1

Out

puts

Pr

evio

us

)(

)1(

)1(

)(

)(

)(

)1(

M

44

44

44

44

44

48

444

44

44

44

44

76

ML

M

LL

L4

44

48

444

47

6

MLt

y

BAn

bt

-x

nb

t -

xt

xt

xn

at

-y

t -

y

iin

r1

- -

[[=

[[

[

[

Page 12: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Para

llel T

rain

ing

Para

llel

trai

nin

g i

s p

oss

ible

by

mak

ing

no

tici

ng

th

e st

ruct

ure

of

the

trai

nin

g d

ata

mat

rix.

Th

e co

lum

ns

rep

rese

nt

the

mo

del

ord

er, t

he

row

s re

pre

sen

t th

e in

div

idu

al t

imes

tep

s.

Cat

enat

ing

ad

dit

ion

al r

ow

s o

nto

th

e m

atri

x is

per

mit

ted

. Th

e o

nly

cav

eat

is t

o e

nsu

re c

on

sist

ency

of

pre

vio

us

forc

es a

nd

dis

pla

cem

ents

occ

uri

ng

b

efo

re t

ime

zero

.

Ad

van

tag

es:

Elim

inat

es lo

ng

-lag

co

nta

min

atio

nEl

imin

ates

mo

dal

dis

per

sio

nD

istr

ibu

ted

co

mp

uti

ng

Exci

tati

on

tai

lori

ng

Le

ss s

ensi

tive

to e

rro

rs t

han

s

eria

l tra

inin

g

Dis

adva

nta

ges

:Ex

tra

bo

okw

ork

Mu

st jo

in a

ll si

gn

als

for m

od

el e

valu

atio

ns

y(0

)x(1

)x(2

)y(1

)x(2

)x(3

)y(2

)x(3

)x(4

)

·

A1

B1

B2

=

y(1

)y(2

)y(3

)

Page 13: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Exci

tati

on

Sig

nal

sFo

r th

e ae

rod

ynam

ic id

enti

ficat

ion

pro

cess

to

cap

ture

a

use

ful

mo

del

, th

e d

om

inan

t ae

rod

ynam

ics

mu

st b

e ex

cite

d. T

he

exci

tati

on

is

dir

ectl

y re

late

d t

o t

he

inp

ut

sig

nal

. Th

e fo

llow

ing

cr

iter

ia

wer

e id

enti

fied

as

im

po

rtan

t to

su

cces

sfu

l in

pu

t si

gn

al d

esig

n. T

he

crit

eria

ar

e b

ased

on

ph

ysic

s, p

erfo

rman

ce a

nd

sys

tem

mo

del

ch

arac

teri

stic

s.

Cri

teri

aTh

e ex

cita

tio

n m

ust

be

con

sist

ent

wit

h t

he

un

stea

dy

aer

od

ynam

ic re

pre

sen

tati

on

.

Stat

ic o

ffse

t fo

rces

mu

st b

e ca

lcu

late

d

Exci

te t

he

do

min

ant

un

stea

dy

aero

dyn

amic

s w

hile

bei

ng

kep

t in

th

e "l

inea

r" a

ero

dyn

amic

ran

ge

Inp

ut

and

ou

tpu

t d

ynam

ics

mu

st b

e ex

cite

d

Exci

te t

he

syst

em w

ith

in a

use

ful f

req

uen

cy ra

ng

e

Page 14: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Mu

ltis

tep

01

02

03

04

05

06

07

08

09

01

00

10

4

10

2

10

0

10

2

10

4

Dis

pla

cem

en

t

PSD

01

02

03

04

05

06

07

08

09

01

00

10

2

10

0

10

2

10

4

Fre

qu

en

cy (

%N

yqu

ist)

Ve

loci

ty

PSD

20

24

68

10

10.50

0.51

Ve

loci

ty

20

24

68

10

1

0

0.51

Dis

pla

cem

en

t

0.5

Ad

van

tag

es:

Easy

to im

ple

men

tSi

mp

le fu

nct

ion

al fo

rmB

road

PSD

Co

mm

on

flig

ht-

test

sig

nal

Dis

adva

nta

ges

:In

con

sist

ent

Bo

un

dar

y C

on

dit

ion

sN

on

intu

itiv

e ex

cita

tio

n le

ng

thH

ole

s in

PSD

Page 15: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Ch

irp

00

.10

.20

.30

.40

.50

.60

.70

.80

.91

1

0.

50

0.51

00

.10

.20

.30

.40

.50

.60

.70

.80

.91

21

012

00

.10

.20

.30

.40

.50

.60

.70

.80

.91

42

024

Dis

pla

cem

en

t

Ve

loci

ty/w

Acc

ele

ratio

n/w

2

01

02

03

04

05

06

07

08

09

01

00

10

4

10

2

10

0

10

2

Dis

pla

cem

en

t

PSD

01

02

03

04

05

06

07

08

09

01

00

10

2

10

0

10

2

10

4

Fre

qu

en

cy (

%N

yqu

ist)

Ve

loci

ty

PSD

Lin

ear F

req

uen

cy S

wee

p

d(t

)=

sin(

ωt2

)

v(t

)=

2ωtco

s(ωt2

)

Ad

van

tag

es:

Co

nsi

sten

t w

ith

fun

ctio

nal

req

uir

emen

tsFl

at P

SDIn

tuit

ive

exci

tati

on

len

gth

Dis

adva

nta

ges

:Po

or l

ow

freq

uen

cy p

erfo

rman

ceO

nly

cap

ture

s an

alyt

ic re

spo

nse

Can

ove

rexc

ite

the

syst

em

Page 16: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

DC

Ch

irp

Lin

ear F

req

uen

cy S

wee

p w

ith

a D

C O

ffse

t0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

10

0.51

00

.10

.20

.30

.40

.50

.60

.70

.80

.91

1

0.

50

0.51

00

.10

.20

.30

.40

.50

.60

.70

.80

.91

21

012

Dis

pla

cem

en

t

Ve

loci

ty/w

Acc

ele

ratio

n/w

2

01

02

03

04

05

06

07

08

09

01

00

10

10

10

5

10

0

10

5D

isp

lace

me

nt

01

02

03

04

05

06

07

08

09

01

00

10

2

10

0

10

2

10

4

10

6

Fre

qu

en

cy (

%N

yqu

ist)

Ve

loci

ty

d(t

)=

−1 2

cos(

ωt2

)+

1 2

v(t

)=

ωtsi

n(ωt2

)

Ad

van

tag

es:

Sam

e ad

van

tag

es a

s th

e ch

irp

Imp

rove

d lo

w fr

equ

ency

exc

itat

ion

Dis

adva

nta

ges

:Pe

ak fa

cto

r is

twic

e th

e ch

irp

's

Page 17: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Fres

nel

Ch

irp

The

Fres

nel

ch

irp

is t

he

inte

gra

lo

f th

e si

ne

fun

ctio

n. T

he

Fres

nel

form

is c

om

mo

nly

see

n in

op

tics

.

00

.10

.20

.30

.40

.50

.60

.70

.80

.91

1

0.

50

0.51

00

.10

.20

.30

.40

.50

.60

.70

.80

.91

0

0.2

0.4

0.6

0.8

Dis

pla

cem

en

t

Ve

loci

ty

01

02

03

04

05

06

07

08

09

01

00

10

4

10

2

10

0

10

2

10

4

Dis

pla

cem

en

t

PSD

01

02

03

04

05

06

07

08

09

01

00

10

4

10

2

10

0

10

2

Fre

qu

en

cy (

%N

yqu

ist)

Ve

loci

ty

PSD

v(t

)=

sin(

ωt2

)

d(t

)=

S(t

)=

∫si

n(ωt2

)

Ad

van

tag

es:

Flat

PSD

for v

elo

city

DC

off

set

com

po

nen

t

Dis

adva

nta

ges

:N

o c

lose

d fo

rm s

olu

tio

nD

isp

lace

men

t PS

D h

as h

ole

sO

ffse

t Fr

esn

el is

imp

ract

ical

Page 18: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Sch

roed

er S

wee

p

05

10

15

1

0.

50

0.51

Dis

pla

cem

en

t

05

10

15

30

20

10

0

10

20

30

Ve

loci

ty

01

02

03

04

05

06

07

08

09

01

00

10

6

10

4

10

2

10

0

10

2D

isp

lace

me

nt

PSD

01

02

03

04

05

06

07

08

09

01

00

10

2

10

0

10

2

10

4

Fre

qu

en

cy (

%N

yqu

ist)

Ve

loci

ty

PSD

Ad

van

tag

es:

Flat

PSD

Op

tim

al p

eak

fact

or

Des

irab

le lo

w fr

equ

ency

resp

on

seH

arm

on

ic s

ign

al a

llow

s an

arb

itra

ry

e

xcit

atio

n le

ng

th

Dis

adva

nta

ges

:D

oes

no

t st

art

fro

m re

stSe

nsi

tivi

ties

to e

xcit

atio

n m

eth

od

The

Sch

roed

er fo

rm is

bas

ed o

na

sum

of c

osi

ne

term

s w

ith

asp

ecifi

ed p

has

ing.

Th

e fo

rm is

anal

ytic

in t

ime

bu

t n

ot

smo

oth

in fr

equ

ency

.

d(t

)=

N ∑ k=

1

√1 2N

cos(

2πkt

T−

πk

2

N)

v(t

)=

N ∑ k=

1

−2πk

T

√1 2N

sin(

2πkt

T−

πk

2

N)

Page 19: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

No

ise

An

oft

en p

rop

ose

d s

olu

tio

n is

to u

sea

no

ise

inp

ut

sig

nal

. 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

11

0.

50

0.51

01

02

03

04

05

06

07

08

09

01

00

10

3

10

2

10

1

10

0

10

1

PSD

10

6

10

4

10

2

10

0

10

2

PSD

Ad

van

tag

es:

Flat

PSD

No

lim

it o

n e

xcit

atio

n le

ng

thEx

cite

s en

tire

flo

w d

ynam

ics

Dis

adva

nta

ges

:Fl

at P

SD o

nly

occ

urs

in t

he

limit

No

n in

tuit

ive

exci

tati

on

len

gth

No

n d

eter

min

isti

c ca

use

s B

C p

rob

lem

sEq

ual

po

wer

mea

ns

"sh

arp

" ch

ang

es

PSD

for 5

00 a

nd

500

0 d

ata

po

ints

Mo

tio

n

Page 20: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Art

ifici

al N

ois

e0

0.5

11

.52

2.5

33

.5

05

10

No

isy

1

00

.51

1.5

22

.53

3.5

05

10

1

Cle

an

01

02

03

04

05

06

07

08

09

01

00

10

30

10

2

5

10

2

0

10

1

5

10

1

0

10

5

10

0

10

5

10

10

10

15

Tran

sfer

Fu

nct

ion

Ou

tpu

t

Inp

ut

Freq

uen

cy (%

Nyq

uis

t)

Pow

er S

pec

tral

Den

sity

01

02

03

04

05

06

07

08

09

01

00

10

14

10

1

2

10

1

0

10

8

10

6

10

4

10

2

10

0

Pow

er S

pec

tral

Den

sity

Tran

sfer

Fu

nct

ion

Ou

tpu

t

Inp

ut

Freq

uen

cy (%

Nyq

uis

t)

No

isy

Sig

nal

Cle

anSi

gn

al

Co

ntr

ary

to in

tuit

ion

, ad

din

g n

ois

e to

an

exis

tin

g s

ign

al is

oft

en d

esir

ed fo

r bet

ter

hig

h fr

equ

ency

mo

del

ing.

Page 21: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Mo

tio

n S

pec

ifica

tio

n

Stri

ct S

pec

ifica

tio

n

Stat

e Sp

ace

Spec

ifica

tio

nA

dva

nta

ges

:Si

mu

ltan

eou

s B

C u

pd

ates

Inp

ut

sig

nal

s ar

e fil

tere

d t

hro

ug

h a

"

stru

ctu

re"

wit

h m

ass.

Dis

adva

nta

ges

:Re

qu

ires

sel

ecti

ng

a "

stru

ctu

ral m

ass"

Un

der

sira

ble

sto

pp

ing

co

nd

itio

ns

lead

to m

oti

on

s th

at a

re n

on

linea

r.

Ad

van

tag

es:

"Per

fect

" si

gn

als

Easy

imp

len

tati

on

Dis

adva

nta

ges

:N

ot

con

sist

ent

in a

dis

cret

e ti

me

sen

se

For

stri

ct

mo

tio

n

spec

ific

atio

n,

the

mo

tio

n

bo

un

dar

y co

nd

itio

ns

at

each

ti

mes

tep

ar

e d

eter

min

ed

fro

m

a n

um

eric

al o

r an

alyt

ical

exp

ress

ion

.

For

stat

e sp

ace

mo

tio

n s

pec

ifica

tio

n, t

he

bo

un

dar

y co

nd

itio

ns

are

up

dat

ed

sim

ult

aneo

usl

y th

ou

gh

th

e m

oti

on

sta

te

vect

or.

This

exa

ctly

du

plic

ates

th

e ac

tual

d

iscr

ete

tim

e C

FD m

oti

on

form

.

Page 22: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Mo

del

Eva

luat

ion

An

eva

luat

ion

met

ho

do

log

y is

nee

ded

to

co

mp

are

and

sel

ect

the

"bes

t"

aero

dyn

amic

sys

tem

mo

del

. Vis

ual

insp

ecti

on

is d

oo

med

; th

e ev

alu

atio

n

mu

st b

e b

ased

on

a n

um

eric

al m

easu

rem

ent.

Eval

uat

ion

bas

ed o

n M

od

el E

rro

rs

χ2

=||A

·x−

b||2

nst

p

02

04

06

08

01

00

12

01

40

16

01

80

20

01

06

10

4

10

2

10

0

10

2

Nu

mb

er

of

Mo

de

l Co

eff

icie

nts

c2

02

04

06

08

01

00

12

01

03

10

2

10

1

Nu

mb

er

of

Mo

de

l Co

eff

icie

nts

Fo

rce

RM

S

Ind

icat

es t

he

mat

rix

solu

tio

n q

ual

ity

Ind

icat

es t

he

mo

del

pre

dic

tio

n q

ual

ity

RM

S=

√ √ √ √1 N

N ∑ k=

0

(y(k

)−

y(k

))2

Page 23: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Mo

del

Eva

luat

ion

Eval

uat

ion

bas

ed o

n A

ero

elas

tic

Stab

ility

Usi

ng

th

e m

od

el f

or

cou

ple

d a

ero

stru

ctu

ral

pre

dic

tio

ns

exer

cise

s th

e m

od

el's

pre

dic

tio

n c

har

acte

rist

ics

in a

rea

listi

c an

d i

ntu

itiv

e m

ann

er.

This

eva

luat

atio

n i

s b

ased

on

sys

tem

eig

enva

lue

pre

dic

tio

ns.

Th

e co

up

led

aer

oel

asti

c p

lan

t m

atri

x is

:

05

10

15

20

25

30

0.4

0.4

5

0.5

0.5

5

0.6

0.6

5

0.7

0.7

5

0.8

12

3456

7891

01112131415161718192

nb

Ord

er

Dynamic Pressure

na

Ord

er

Eig

enva

lue

Form

Stab

ility

Bo

un

dar

y Fo

rm

[ Gs+

q ∞H

sD

aC

sq ∞

HsC

a

HaC

sG

a

]

Page 24: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

AG

ARD

445

.6

Two

Mod

e St

ruct

ural

Mod

el

Firs

t Ben

ding

Mod

e9.

6 H

ertz

Firs

t Tor

siona

l Mod

e38

.2 H

ertz

45 d

egre

e sw

eep

Asp

ect R

atio

260

% ta

per r

atio

NAC

A 6

5A00

4 Ro

ot

Tip

V

Page 25: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

AG

ARD

445

.6

05

10

15

20

25

30

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0

1

23 4 56

7 8 91

0 1112131415 1617181920

nb

Ord

er

Dynamic Pressure

na

Ord

er

05

10

15

20

25

30

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.91

0

12

3456

7891

011121314151617181920

nb

Ord

er

Dynamic Pressure

na

Ord

er

05

10

15

20

25

30

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0

12

3456

7891

011121314151617181920

nb

Ord

er

Dynamic Pressure

na

Ord

er

05

10

15

20

25

30

0

0.2

0.4

0.6

0.81

1.2

1.4

1.6

1.8

01

2

3 456 7 8 910111213 14 15161718 1920

nb

Ord

er

Dynamic Pressure

na

Ord

er

01

02

03

04

05

06

00

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0

1 234567891011121314151617181920

nb

Ord

er

Dynamic Pressure

na

Ord

er

05

10

15

20

25

30

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0

1

2

3456

78 910

nb

Ord

er

Dynamic Pressure

na

Ord

er

Ch

irp

DC

Ch

irp

DC

Ch

irp

Larg

e A

mp

litu

de

Mu

ltis

tep

Fres

nel

wit

h

Stat

e Sp

ace

Spec

ifica

tio

nSc

hro

eder

Page 26: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

AG

ARD

445

.6

0.4

0.5

0.6

0.7

0.8

0.9

11

.11

.20

0.2

0.4

0.6

0.81

1.2

1.4

1.6

1.8

Ma

ch N

um

be

r

Dynamic Pressure [psi ]

Stab

ility

Bo

un

dar

y

Page 27: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

AG

ARD

445

.6 S

ensi

tivi

ty S

tud

ies

Eig

enva

lues

in z

-plan

e at

Mac

h 0.

499

Uni

t Circ

le, |

z|=

1

Mod

e 2

Mod

e 1

Incr

easin

gD

ensit

y

Incr

easin

gD

ensit

y

|z|=

1

1.14

11.

072

0.49

9

0.67

80.

900

0.96

0

Den

sity

Swee

pM

odel

Sens

itivi

ty S

tudy

0.4

0.5

0.6

0.7

0.8

0.9

11

.11

.20

0.2

0.4

0.6

0.81

1.2

1.4

1.6

1.82

Ma

ch N

um

be

r

Sta

bili

ty B

ou

nd

ary

[p

si]

+10

%

-10%

As-

Is

Str

uct

ura

l Fre

qu

en

cy

Stuc

tura

l Fre

quen

cy S

tudy

02

46

81

01

21

41

60

.7

0.7

5

0.8

0.8

5

0.9

0.9

51

1.0

5

1.1

Da

mp

ing

Ra

tio %

Sta

bili

ty B

ou

nd

ary

[p

si]

Ma

ch 0

.49

9

Mach

1.0

72

Stuc

tura

l Dam

ping

Stu

dy

Page 28: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Win

g/F

lap

Co

ntr

ol

U

θδ

De

L

Sch

emat

ic

CFD

Gri

d

05

10

15

20

25

30

35

0

0.51

1.52

Dis

pla

cem

en

t

1 2

05

10

15

20

25

30

35

20

10

0

10

20

Ve

loci

ty

1 2

05

10

15

20

25

30

35

0.0

1

0.0

050

0.0

05

0.0

1F

orc

es

12

05

10

15

20

25

30

35

0

0.51

1.52

Dis

pla

cem

en

t

12

05

10

15

20

25

30

35

20

10

0

10

20

Ve

loci

ty

12

05

10

15

20

25

30

35

2024x

10

4F

orc

es

1 2

Trai

nin

g S

ign

als

Page 29: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Win

g/F

lap

Co

ntr

ol

Op

en L

oo

p

00

.01

0.0

20

.03

0.0

40

.05

0.0

60

.07

0.0

80

.09

0.1

40

20

0

20

40

Dis

pla

cem

en

t

12

00

.01

0.0

20

.03

0.0

40

.05

0.0

60

.07

0.0

80

.09

0.1

2101x

10

4V

elo

city

12

00

.01

0.0

20

.03

0.0

40

.05

0.0

60

.07

0.0

80

.09

0.1

500

50

Fo

rce

s

1 2

00

.10

.20

.30

.40

.50

.60

.70

.85

05D

isp

lace

me

nt

12

00

.10

.20

.30

.40

.50

.60

.70

.81

00

0

50

00

50

0

10

00

Ve

loci

ty

1 2

00

.10

.20

.30

.40

.50

.60

.70

.82024

Fo

rce

s

12

Clo

sed

Lo

op

k={3

.5, 0

, -0.

01, -

0.04

}

Clo

sed

Lo

op

k={3

.5, 0

, -0.

01, -

0.05

}

Clo

sed

Lo

op

Eig

enva

lues

k={3

.5, 0

, -0.

01, -

0.05

}

Eig

enva

lues

nea

r th

eu

nit

cir

cle

are

cau

sin

ga

no

n-p

hys

ical

inst

abili

ty.

Page 30: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve

Co

ncl

usi

on

sIm

pro

vem

ents

wer

e m

ade

in t

he

syst

em i

den

tifi

cati

on

ro

uti

ne.

A c

om

par

iso

n o

f cla

ssic

al u

nst

ead

y ae

rod

ynam

ic s

olu

tio

ns

was

per

form

ed. T

he

AR

MA

fo

rm r

emai

ns

the

pre

ferr

ed

syst

em re

pre

sen

tati

on

.

A p

aral

lel t

rain

ing

met

ho

d w

as d

evel

op

ed. P

aral

lel t

rain

ing

al

low

s fo

r d

eco

up

led

exc

itat

ion

an

d y

ield

s b

ette

r sy

stem

m

od

els.

An

exc

itat

ion

sig

nal

su

rvey

was

co

nd

uct

ed. T

he

DC

-Ch

irp

g

ave

the

mo

st

con

sist

ent

resu

lts.

H

igh

fr

equ

ency

lim

itat

ion

s o

f th

e cu

rren

t si

gn

als

wer

e id

enti

fied

an

d

eval

uat

ed.

Mo

del

q

ual

ity

"Eva

luat

ion

cr

iter

ia"

wer

e te

sted

. Th

e co

up

led

sta

bili

ty p

red

icti

on

eva

luat

ion

ap

pea

rs to

giv

e th

e b

est

"mo

del

qu

alit

y" in

dic

atio

n.

Page 31: Master's Thesis Defense Improved System Identification for ... · = Solving for the x vector yields the ARMA model coefficents. Singular value decomposition (SVD) is used to solve