scieee Science in your language
[en] (orig)

Control design of mixed sensitivity problem for educational model of helicopter

Abstract

The paper deals with the design of H-∞ robust controller, particularly with mixed sensitivity problem for elevation control. It briefly introduces basic mathematical background concerning robust control approach, which is then applied for typical example of MIMO system, that is a helicopter model. The obtained results are verified on real educational physical model CE 150 by Humusoft, ltd.

Read accessible full text

Control design of mixed sensitivity problem for educational model of helicopter

Author: Ožana, Štěpán
Publisher: Vysoká škola báňská - Technická univerzita Ostrava
Year: 2014
DOI: 10.15598/aeee.v12i5.1150
Source: https://dspace.vsb.cz/bitstreams/22149895-ebfa-4c98-89ef-cf4d51f73d0e/download
CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
Con ol Design o Mixed Sensi i i y P oblem o
Educa ional Model o Helicop e
S epan OZANA , Pe VOJCINAK , Ma in PIES , Rado an HAJOVSKY
Depa men o Cybe ne ics and Biomedical Enginee ing, Facul y o Elec ical Enginee ing and Compu e
Science, VSB–Technical Uni e si y o Os a a, 17. lis opadu 15, 708 33 Os a a-Po uba, Czech Republic
[email p o ec ed], pe . o[email p o ec ed], [email p o ec ed], ado an.hajo[email p o ec ed]
Abs ac . The pape deals wi h he design o H-∞ o-
bus con olle , pa icula ly wi h mixed sensi i i y p ob-
lem o ele a ion con ol. I b ie ly in oduces basic
ma hema ical backg ound conce ning obus con ol ap-
p oach, which is hen applied o ypical example o
MIMO sys em, ha is a helicop e model. The ob ained
esul s a e e i ied on eal educa ional physical model
CE 150 by Humuso , l d.
Keywo ds
Algo i hms and so wa e, simula ion o dy-
namic sys ems, obus con ol o nonlinea sys-
ems.
1. In oduc ion
The objec i e o he obus con ol is o design a
dynamic con ol sys em ope a ing in a eal en i on-
men . The changes o he su ounding condi ions can
be caused by he ollowing ac o s acco ding o [15]:
•componen aging,
• empe a u e e ec ,
•e ec o he wo king en i onmen .
The con ol sys em mus no only be esis an o
he a o emen ioned ac o s bu i also mus elimina e
inaccu acy o he model, i.e. obus ness is he ele an
abili y o he con ol sys em o accep changes. The
equi ed ou pu alue will be eached e en when he
changes in he p ope ies o he con olled sys em a e
limi ed and cons an dis u bance signals a e ope a ing.
F om he ma hema ical poin o iew, he obus con-
olle is no only sui able o one pa icula sys em
bu o a se o sys ems [15].
In o he wo ds, obus ness plays a signi ican ole
in he design o con ol sys ems as eal sys ems a e
p one o ex e nal dis u bances and measu emen noise.
Mo eo e , he e a e o en di e ences be ween he p o-
posed ma hema ical models and ac ual eal sys ems.
A ypical example is he design o a con olle ha
will s abilize he sys em e en i i is o iginally uns a-
ble and accep a pa icula le el o pe o mance a he
p esence o dis u bance signals, noises, ha d- o-model
p ocess dynamic cha ac e is ics o p ocess pa ame e
a iables. Such asks a e bes sol ed by a eedback
con ol mechanism as hey b ing along a whole ange
o p oblems acco ding o [4]:
•high p ice (e.g. use o senso s),
•sys em complexi y (e.g. possibili ies o implemen-
a ion and eliabili y),
•sys em s abili y (e.g. equi emen o in e nal s a-
bili y and s abilizing con olle s).
The need and signi icance o obus ness as a pa
o con ol sys em designs ha e been de eloping since
1980s. Robus ness in he s anda d SISO con ol is
p o ided by a sui able gain ma gin and phase ma -
gin. When he i s design echniques o mul i- a iable
sys ems de eloped in 1960s, emphasis was laid on he
achie emen o good pe o mance, no obus ness. The
me hods ha use mul i a iables we e based on he
linea –quad a ic c i e ion and Gaussian dis u bances.
I was demons a ed ha hey can be success ully used
in a whole ange o a ia ion applica ions whe e i is
possible o se up p ecise ma hema ical models, in-
cluding he desc ip ions o ex e nal dis u bance signals
o noises. Howe e , he applica ion o hese me h-
ods, called LQG me hods (linea -quad a ic Gaussian
con ol), in o he indus ial b anches e iden ly showed
hei bad p ope ies om he poin o obus ness which
led o he e o o de elop a heo y ha would explic-
i ly deal wi h he issue o obus ness in he con ol
eedback design. The pionee ing wo k on he de el-
opmen o he heo y, oday known as he heo y o
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 488
CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
op imal con ol in H-∞, was in oduced a he begin-
ning o 1980s by Geo g Zames and B uce A. F ancis.
The H-∞app oach i s speci ied he model o sys em
unce ain y, i.e. addi i e pe u ba ion and/o ou pu
dis u bances. In mos cases, i is enough o ind a sui -
able con olle so ha he closed loop achie es some
obus s abili y. The pe o mance is also a pa o he
op imiza ion loss (objec i e) unc ion. The elegan o -
mula ions o he solu ion a e based on he solu ions o
Ricca i equa ions, e.g. in MATLAB [4].
I we design a con olle o a pa icula in e al o
pa ame e s, hen he con ol ci cui is obus ly s able.
Ano he impo an pa ame e o obus con olle s is
hei pe o mance, mee ing he equi emen s o pa-
ame e s acco ding o [15]:
•con ol,
•dis u bance,
•speed o esponse (se ling ime).
The p oblem o he design o he obus con olle
is based on he eedback ci cui (closed loop) which
enables wo king wi h he sensi i i y and elimina ion o
he dis u bance. On one side, he eedback o a non-
s able sys em s abilizes i ; on he o he side, i may
des abilize a s able sys em [15]. Conside he s anda d
con ol diag am acco ding o Fig. 1.
Fig. 1: Classical con ol scheme wi h he de ini ion o ci cui
signals.
Desc ip ion o he signals in he con ol ci cui :
•W(s): Laplace ans o m o he he e e ence sig-
nal,
•E(s): Laplace ans o m o he con ol e o sig-
nal,
•U(s): Laplace ans o m o he manipula ed alue
signal,
•V1(s): Laplace ans o m o he dis u bance: low-
equency known and unknown dis u bances; he
sys em mus elimina e hem,
•V2(s): Laplace ans o m o he dis u bance: sen-
so s o measu emen , high- equency cha ac e
wi h insigni ican e ec .
We use he ules o block algeb a o de ine he ma he-
ma ical ela ions wi hin he con ol ci cui , i.e. acco d-
ing o [15], [9]:
• o he open-loop ans e unc ion:
L(s) = K(s)G(s),(1)
• o he ans e unc ion o he con ol e o - he
sensi i i y unc ion:
GE(s) = E(s)
W(s)=1
1 + L(s)=S(s),(2)
• o he closed-loop ans e unc ion - he comple-
men a y sensi i i y unc ion:
GW(s) = Y(s)
W(s)=L(s)
1 + L(s)=T(s),(3)
•limi ing condi ion applies o he sum o he sen-
si i i y unc ion and complemen a y sensi i i y
unc ion:
S(s) + T(s) = 1
1 + L(s)+L(s)
1 + L(s)= 1.(4)
Conside he equi emen s o he sensi i i y unc-
ion and complemen a y unc ion acco ding o [15]:
•a 1( ) = 2( )=0, he e ec o he con ol ac-
ion p edomina es and so he sensi i i y unc ion
ˆ
GE(s) = S(s)mus be small and he complemen-
a y unc ion GW(s) = T(s)will be la ge,
• he en i e con ol ci cui ca ies ou he elimina-
ion o he low- equency noise 1( ), and so again,
he e ec o he con ol ac ion p edomina es: he
sensi i i y unc ion GE(s) = S(s)mus be small
and he complemen a y unc ion GW(s) = T(s)
will be la ge,
• he en i e con ol ci cui mus no a ec he elimi-
na ion o he high- equency noise 2( )and so we
elimina e he e ec s o he con ol ac ion and con-
ol e o , i.e. he sensi i i y unc ion GE(s) =
S(s)mus be small and he complemen a y sensi-
i i y unc ion GW(s) = T(s)will also be small.
As he a o emen ioned opposing equi emen s can-
no be me by one con olle , i is necessa y o ind a
comp omise be ween he sizes o he sensi i i y unc-
ion and he complemen a y sensi i i y unc ion [15],
see Fig. 2.
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 489
CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
Fig. 2: Loga i hmic ampli ude equency cha ac e is ics o he
sensi i i y unc ion (on he le ) and complemen a y
sensi i i y unc ion (on he igh ).
2. Robus Con olle Design
2.1. Robus Design Me hods
The me hods o he s a e space in he ime domain
allowed o a oid he p oblems wi h ans e unc ion
ma ices and also p o ided means o he analysis and
design o MIMO sys ems wi h mo e inpu s and ou -
pu s. App oxima ely a he same ime when he me h-
ods o op imal con ol we e being de eloped, esea ch
ocused on he ex ensions o means o MIMO sys em
s anda d con ol was conduc ed. The obus design is
based on he inding o such a con olle so ha he
esul ing sys em in he closed loop is also obus . Ro-
bus ness became he main s andpoin in he ield o
con ol, he e o e speci ica ions and me hods ollowed
sho ly, i.e. acco ding o [15], [5]:
•H∞me hod,
•H2me hod,
•LTR me hod (loop ans e eco e y),
•µ: syn hesis,
•QFT me hod (quan i a i e eedback heo y),
•Kha i ono heo em o he examina ion o obus
s abili y,
•speci ica ion o he small-gain heo em,
•speci ica ion o s uc u ed singula alues.
The ollowing p esen a ion will only ocus on he H∞
me hod.
2.2. H∞Me hod
A con ol sys em is obus i i s ays s able and mee s
pa icula beha io al c i e ia a he p esence o possi-
ble unce ain ies. The H∞op imiza ion me hod, de-
eloped since 1980s, has p o ed o be a e y e icien
and po en design me hod o obus con ol in he ield
o linea , ime-in a ian con ol sys ems [15], [11], [3].
H∞con olle s ha e hei own e minology, no a ion
and concep ion. This me hod leads o a se o sui -
able s able ans e unc ions ha a e physically iable.
Simila ly o LQR and LQG con olle s, we expec op-
imiza ion o he objec i e unc ion ha will compa e
di e en ans e unc ions and selec he mos sui able
one om he se . The equi emen s o he closed loop
a e he ollowing, i.e. acco ding o [15]:
•Physical iabili y: The o de o he ans e unc-
ion denomina o mus be highe o equal o he
o de o he ans e nume a o .
•S abili y: The ans e poles mus lie in he le
hal plane o he Gaussian plane o in he a ea o
he Laplace ans o m con e gence (p o ided ha
he con ol s aigh line and imagina y axis a e
iden ical).
The basic p e equisi e o he H∞me hod is he
knowledge o he ans e unc ion o he gi en sys em,
e alua ing ∞-no m acco ding Fig. 5 acco ding o [15],
[1]:
kGk∞=supω{|G(jω)|} .(5)
The no m can be g aphically ep esen ed as he max-
imum o he Bode diag am p o ided ha he ans e
unc ion is de ini e and has no imagina y poles, while
i s objec i e is o minimize i in he ∞-no m. I de-
c eases he apex o he Bode diag am, which inc eases
he obus s abili y ma gin [15].
2.3. Mixed Sensi i i y P oblem
Usually, p ac ical indus ial applica ions do no only
use one objec i e unc ion bu a combina ion o se e al
unc ions like ha , e.g. accomplishmen o he good
pe o mance o acking he e e ence signal a limi ed
ene gy o he e e ence signal. Then we sol e he mixed
sensi i i y ask, o he ’S o e KS’ p oblem de ined by
a gene al ela ion ( o he SISO sys em) acco ding o
[8], [12]:
minKs 



S(s)
K(s)S(s)


∞
=
= minKs 



[1 + L(s)]−1
K(s) [1 + L(s)]−1


∞
.(6)
The Eq. 6 can also be exp essed by he equi emen s
o he design conce ning he addi i e pe u ba ion, e.g.
nominal beha io , good pe o mance o acking he
e e ence signal o he elimina ion o dis u bance sig-
nals and obus s abili y [4].
Figu e 3 shows he s anda d block diag am o he
H∞con igu a ion using he linea ac al ans o ma-
ion (LFT) wi h speci ica ion o he indi idual ex e nal
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 490
CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
Fig. 3: S anda d block diag am a he H∞con igu a ion.
inpu s, ex e nal ou pu s, inpu s in o he con olle and
i s ou pu s, see [1] (p. 438). The con ol ci cui con-
ains a obus con olle wi h ans e unc ion K(s)
and a pe u bed (also ex ended o gene alized) sys em
wi h ans e unc ion P(s) ha has wo inpu s and
wo ou pu s acco ding o [15]:
•~w ( )inpu e e ence signal ec o ; ex e nal inpu
signals,
•~ ( )ou pu manipula ed signal ec o ; ou pu
con ol signals om he con olle .
The main di e ence be ween he ec o s is ha he
con olle does no a ec he inpu s. The inpu e e -
ence signal ec o ~w ( )includes an ex e nal noise, noise
om he senso s and acking ( e e ence) signals. To
he con a y, he ou pu s om he sys em a e di ided
in o wo g oups acco ding o [15]:
•~y ( )ou pu signal ec o ; measu ed ou pu s,
•~z ( )con olled ou pu s; minimized o penalized
ou pu s.
The ask is hen de ined so ha he in e nally s a-
bilizing con olle K(s)is sea ched o in he con ol
ci cui o he obus con ol o he gi en gene alized
sys em P(s) ha minimizes o penalizes he con olled
ou pu ec o ~z ( ). In o he wo ds, we minimize he
maximal no m o he ans e unc ion be ween ~w ( )
and ~z ( )by he gi en ela ion acco ding o [4]:
~z =P11 (s) + P12 (s)K(s)I−1
P K P21 (s)~w, (7)
whe e:
IP K =I−P22 (s)K(s),(8)
we ge a linea ac al ans o ma ion a e he adjus -
men :
~z =Fl[P(s), K (s)] ~w. (9)
Then, he H∞op imiza ion p oblem can be ex-
p essed by a ela ion, i.e. acco ding o [4]:
min
Ks
kFl[P(s), K (s)]k∞.(10)
Fig. 4: S anda d block diag am o he mixed sensi i i y p ob-
lem: con olle and pe u bed sys em con aining nomi-
nal sys em, con ol e o and manipula ed alue weigh -
ing il e s.
Figu e 4 shows he s anda d block diag am o he
mixed sensi i i y p oblem and i is basically a mo e
de ailed illus a ion o Fig. 3, whe e i is easy o deduce
he ollowing ela ions, i.e.:
• o ex e nal inpu signals:
~w ( ) = ( ),(11)
• o ou pu con ol signals om he con olle :
~u ( ) = u( ),(12)
• o measu ed ou pu s:
~y ( ) = e( ),(13)
• o minimized o penalized ou pu s:
~z = [z1( ), z2( )]T=
= [W1(s)e( ), W2(s)u( )]T.(14)
The ollowing applies o he gene alized sys em con-
aining weigh ing il e s acco ding o [4]:
P(s) = P11 (s)P12 (s)
P21 (s)P22 (s),(15)
while:
P11 (s) = W1(s) [I, 0]T= [W1(s),0]T,(16)
P12 (s)=[−W1(s)G(s), W2(s)I]T,(17)
P21 (s) = [I]T=I, (18)
P22 (s) = [−G(s)]T=−G(s).(19)
The weigh ing il e s W1(s)and W2(s)a e com-
monly used in p ac ice; in such case, he Eg. 6 can
be o mally adjus ed in o he o m desc ibing a objec-
i e unc ion, i.e. ( o he SISO sys em) acco ding o
[4]:
min
Ks 



W1(s)S(s)
W2(s)K(s)S(s)


∞
.(20)
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 491
CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
3. Con olle Design o
Helicop e Ele a ion
On he basis o he a o emen ioned heo e ical indings
om he ield o he H∞ obus con ol, we can design
a obus con olle acco ding o he selec ed concep ion
when we deal wi h he ques ion o he au onomy o he
con ol and he p oblem o mixed sensi i i y and also
conside he inpu signals. De i ed ma hema ical mod-
els in ele a ion and azimu h a e c ucial o he design
o con olle s, when we design he con olle o bo h
he ma hema ical model and eal model. Howe e , his
pape only ocuses on ele a ion, due o he ex en o
he ask. Finally, bo h esponses a e compa ed o one
ano he .
The ma hema ical desc ip ion o helicop e model is
aken om o icial manual, see [6]. The design o ele a-
ion con olle pa icula ly comes ou om Figs. 2.1,
2.2 s a ed on pages 8, 15 o his manual.
3.1. Au onomy Requi emen
The main p oblem o his MIMO sys em namely in-
cludes he elimina ion o he ele a ion-azimu h cou-
pling. All in all, we na u ally wan o elimina e bo h
ela ions bu we know ha he azimu h-ele a ion ela-
ion is no as signi ican . The concep ion is hus based
on he assump ion o designing wo independen con-
olle s, i.e. an ele a ion con olle and an azimu h
con olle .
The de ini ion o au onomy says ha he e e ence
signal ~w ( )mus only a ec jus one co esponding
ou pu signal ~y ( ), [7].
Au onomy also equi es ha he ans e ma ix o
he open loop is diagonal, he elemen s o he ma ix
a e only loca ed on he main diagonal. A p io i, diago-
nali y is equi ed in he con ol ma ix. Wi h espec o
he explici ly w i en sign, he ans e unc ion o he
co ec ion e m is gi en by he gene al ela ion based
on he heo y o ma ix de e minan s (Laplace comple-
men ) acco ding o [2]:
Ri,j (s)=(−1)j+1Rj,j (s)G∗
j,i (s)
G∗
j,j (s)
,(21)
whe e iis ma ix line index, jis ma ix column in-
dex, (−1)j+1 is Laplace (algeb aic) complemen , G∗
j,i
is ma ix de e minan G∗
j,i (s)and G∗
j,jis ma ix de-
e minan G∗
j,j (s).
The ela ion Eq. (21) a he has a heo e ical cha -
ac e . Di ec ly de i ed condi ions o au onomy pay
o a a small amoun o egula ed signals (in ou case
he e a e wo signals) a he han ha ing o exac ly
emembe i s con en and a oid making a mis ake in
he algeb aic complemen . The si ua ion is depic ed in
Fig. 5, di ec ly modi ied o he helicop e model.
mech.pa
(ele a ion)
G (s)
1M (s)
1(emp.)
U1(mo )
mech.pa
(azimu h)
R (s)
21
G (s)
11
G (s)
21
G (s)
22
uM
uT+++++
G (s)
M (s)
(emp.)
U1( eak)
G (s)
2M (s)
2(emp.)
U2(mo )
Fig. 5: Block diag am o he elimina ion o he ele a ion-
azimu h coupling; he azimu h-ele a ion coupling is ne-
glec ed.
I he manipula ed alue exp essed by ol age UM
induces an undesi able esponse in he ou pu o he az-
imu h mechanical pa desc ibed by he Laplace ans-
o m G21 (s)UM(s), hen i can be comple ely com-
pensa ed by he co ec ion e m R21 (s)unde he con-
di ion acco ding o [2]:
G21 (s)UM(s) + G22 (s)R21 (s)UM(s)=0
⇒R21 (s) = −G21(s)
G22(s)
.(22)
The ans e unc ion o he second co ec ion e m
can be ei he w i en di ec ly wi h he use o he p in-
ciple o cyclical subs i u ion o indexes o a condi ional
equa ion can be se up again Eq. (23), i.e. acco ding
o [2]:
G12 (s)UT(s) + G11 (s)R12 (s)UT(s)=0
⇒R12 (s) = −G12(s)
ˆ
G11(s)
.(23)
The ela ion Eq. (23) desc ibes he azimu h-
ele a ion coupling ha , howe e , is no signi ican , and
hus can be w i en acco ding o [2]:
R12 (s) = G12 (s) = 0.(24)
The diag am in Fig. 5 can also be in e p e ed in he
ollowing way: he in e nal physical coupling be ween
he ele a ion and azimu h in he o m o ans e unc-
ion G21 (s)canno be elimina ed wi hou he basic
(cons uc i e) in e e ence in o he sys em. Howe e ,
we can qui e easily implemen an ex e nal connec ion
be ween he inpu s o he helicop e se by he co -
ec ion e m R21 (s) ha will ensu e he same as he
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 492

CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
un easible elimina ion o he c oss-coupling inside he
MIMO sys em ans e ma ix, decomposi ion o he
model in o wo models ha seemingly do no in luence
one ano he . Thei con ol can hen be ensu ed by wo
con ol loops independen o one ano he [2].
I we subs i u e in he ela ion Eq. (23), we will e-
cei e ans e unc ion o he co ec ion e m, Eq. (25):
R21 (s) = −T 2s2+T 1s+ 1
B(s)
a UMs+b
a2UTs+b2
A(s),(25)
whe e A(s) = T2s2+ 12,B(s) = T1s2+ 12. The
ela ion (Eq. 25) clea ly shows ha he ans e unc-
ion in his o m does no mee he condi ion o phys-
ical iabili y as he nume a o o de is highe han
he denomina o o de . Thus we de ine wo ine ias
wi h condi ions ξ1T 1and ξ2T 2. The phys-
ically easible co ec ion ans e unc ion elimina ing
he ele a ion-azimu h coupling is gi en by he ela ion
Eq. (26):
R21 (s) = −T 2s2+T 1s+ 1
ξ2s2+ξ1s+ 1
a 0.55s+b
a20.2s+b2
A(s)
B(s).(26)
3.2. Con olle Concep ion
The concep ion o he helicop e model obus con-
ol design was pa ially explained in he equi emen s
o he au onomy o he ele a ion and azimu h con-
ol. The second pa conce ns he H∞ obus con ol,
namely he modi ica ion o he mixed sensi i i y p ob-
lem (MSP) whe e we also penalize he g oup o ex e nal
ou pu signals in addi ion o he con ol e o ~e ( )and
manipula ed alue ~u ( ):
•~
d1( ) = ( ) = ~w ( ) e e ence o ex e nal con ol
signal,
•~
d2( )low- equency signal (dis u bance),
•~
d3( )high- equency dis u bance signal (noise).
We use weigh ing ans e unc ions o also called
weigh ing il e s o penalize signals incoming and ou -
going om he ex ended sys em in he ele a ion o az-
imu h. The whole block diag am o he con ol ci cui
wi h ex ended sys em is shown in Fig. 6.
The e a e many di e en ways o he ex ension o
he nominal sys em. Howe e , he mo e ex e nal in-
pu s and penalized (e o ) ou pu s he e a e, he mo e
di icul i is o selec he weigh ing il e s. The weigh -
ing il e s a e gene ally s able ans e unc ions (no
necessa ily p ope a ional unc ions) o a pa icula
o de . Thus, he mo e we add, he highe he o de
he esul ing sys em will ha e. Such an in e connec ed
sys em can be hen used o exp ess he s a e desc ip-
ion, o he ans e unc ion o he H∞op imal, o
E(s) U(s)
K(s)
W (s)
e
W (s)
u
G (s)
0
z ( )
1
z ( )
2
W (s)
d
W (s)
cmd
W
noise
d ( )
1
d ( )
2
d ( )
3
Y(s)
d( ) +y( )
2
d d( ) + ( )+y( )
2 3
-
+++
+
+
(s)
d( )
1
d( )
2
d( )
3
e( )
u( )
Fig. 6: Block diag am o he connec ion o he H∞ obus con-
olle , nominal sys em and weigh ing il e s o he he-
licop e model.
subop imal obus con olle K(s)wi h one deg ee o
eedom (1DOF con igu a ion).
The meaning o he indi idual blocks is as ollows
acco ding o [10]:
•Wcmd (s): his weigh ing ans e unc ion akes
ca e o he e e ence acking. A no malized signal
appea s a he inpu and he signal a he ou pu
is in ele an physical uni s,
•Wd(s): his weigh ing ans e unc ion adjus s
he equency and ampli ude cha ac e is ics o ex-
e nal low- equency dis u bance signals a ec ing
he nominal sys em,
•Wnoise (s): his weigh ing ans e unc ion ep-
esen s he models o noises o senso s in he e-
quency domain. I ies o de ec a pa icula piece
o in o ma ion in he con ol de i ed om labo-
a o y expe imen s o p oduc ion measu emen s.
Na u ally, he noise shows a high- equency cha -
ac e ,
•We(s): his weigh ing ans e unc ion penalizes
he con ol signal om he obus con olle and
hus he con ol e o . I de e mines he in e ed
alue o he expec ed o m o he ou pu signal.
The signal ha appea s a he inpu o he il e
is in ele an physical uni s and i is no malized
a he ou pu ,
•Wu(s): his weigh ing ans e unc ion penalizes
he con ol signal om he obus con olle and
hus he manipula ed alue signal. I de e mines
he in e ed alue o he expec ed o m o he ou -
pu signal. The signal ha appea s a he inpu
o he il e is in ele an physical uni s and i is
no malized a he ou pu ,
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 493
CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
The comple e calcula ion o bo h he ans e unc-
ion o he H∞ obus con olle and he key H∞no m
is execu ed in MATLAB. The calcula ion algo i hm
is based on he co ec in e connec ion o he nomi-
nal sys em wi h he weigh ing il e s in co espondence
wi h Fig. 6. The p og am solu ion in he M- ile appea s
as ollows, i.e.:
sys emnames = ’G Wcmd Wd Wnoise We Wu’;
inpu a = ’[d1; d2; d3; u]’;
ou pu a = ’[We; Wu; Wcmd-G-Wnoise]’;
inpu _ o_Wcmd = ’[d1]’;
inpu _ o_Wd = ’[d2]’;
inpu _ o_Wnoise = ’[d3]’;
inpu _ o_G = ’[u]’;
inpu _ o_We = ’[Wcmd-G-Wnoise]’;
inpu _ o_Wu = ’[u]’;
cleanupsysic = ’yes’;
P = sysic
NCon ol = 1;
NMeasu e = 1;
= [NCon ol NMeasu e];
[K,CL,gop ] = hin syn(P,NMeasu e,NCon ol);
Fi s o all, we de ine wha sys ems we will in-
e connec , he a iable sys emnames. Then we de-
ine he inpu signal ec o (ex e nal con ol signals
and con ol signal om he con olle ), he a iable
inpu a . The ou pu is ep esen ed by he a i-
able ou pu a con aining he penaliza ion o he con-
ol e o (We), manipula ed alue (Wu) and he mea-
su ed ou pu (Wcmd-G-Wnoise). Subsequen ly, we con-
nec all he inpu s o he weigh ing il e s. By he
cleanupsysic command wi h he a ibu e alue se
o yes we con i m ha we wan o emo e he a i-
ables sys emnames,inpu a and ou pu a om he
MATLAB wo k en i onmen (Wo kspace) immedia ely
a e he c ea ion o he sys em in e connec ion.
The a iable P ep esen s he ex ended sys em o sys-
em in e connec ion (sysic, Sys em In e connec ion).
To comple e he enume a ion o pa ame e s o he cal-
cula ion, we ha e o de ine he numbe o con ol ou -
pu s om he con ol (NCon ol) and he numbe o
measu ed ou pu s (NMeasu e). We will ob ain he cal-
cula ion o he H∞con olle (K), closed loop ans e
(CL) and maximum closed loop ans e no m (gop )
by ac i a ing he hin syn unc ion wi h he ollowing
pa ame e s: P,NMeasu e and NCon ol.
The ac i a ion o he hin syn unc ion can also be
ex ended by mo e inpu and ou pu pa ame e s; in his
ac ual case acco ding o [10]:
• wo algeb aic Ricca i equa ions a e sol ed,
•γ∈(0,+∞),
• he closed-loop ans e unc ions is calcula ed
wi h he use o he linea ac al ans o ma ion
CL =F{P(s), K (s)},
•γ0=kCLk∞=kF{P(s), K (s)}k∞.
MATLAB, namely he Robus Con ol Toolbox, con-
ains o he unc ions ha can be used o sol e he issue
o he design o a con inuous o disc e e H∞ obus con-
olle . Fo comple eness, we only gi e he p o o ype o
he unc ion ocused on he s anda d mixed sensi i i y
p oblem:
[K, CL, gop , INFO] = mixsyn(G, W1, W2, W3)
The p oblem o he mixsyn unc ion is he numbe
o he weigh ing il e s and hei cha ac e as hey only
penalize he con ol e o (W1), manipula ed alue (W2)
and measu ed ou pu (W3). Wi h ega d o he selec ed
design concep ion, i would no be possible o penalize
inpu s wi h his unc ion.
3.3. Ele a ion Con olle o
Ma hema ical Model
The ans e unc ion o he dynamics o he ma he-
ma ical model in ele a ion is gi en by he ela ion ac-
co ding o he ela ion Eq. (27):
Gψ(s) = Ψ (s)
UM(s)=
=7.3315s+ 1.1883
3s4+23s3+116s2+519s+ 1000.(27)
The ampli ude and phase equency cha ac e is ics
a e shown in Fig. 7 which also clea ly show ha i con-
ains he highes alue unde he ollowing condi ions:
• equency: ωMAX = 4.9448  ad ·s−1,
• ans e unc ion module:
|Gψ(jω)|MAX =
=−16.05 [dB]∼
=0.1576 [−].(28)
The sys em is o he ou h o de and con ains ou
s able poles, ou o which wo a e complex conjuga e
and one is a double pole:
p1=−0.2105 + j4.9448,(29)
p2=p1=−0.2105 −j4.9448,(30)
p3=p4=−4.(31)
The maximum no m o he gi en sys em is (in acco -
dance wi h he maximum alue o he ans e unc ion
module):
kGψ(s)k∞= 0.1577 [−].(32)
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 494
CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
-200
-150
-100
-50
0
10
-2 10
-1 10
010
110
210
3
-270
-180
-90
0
90
F equency ( ad/s)
Magni ude (dB)
Phase (deg)
Fig. 7: Ampli ude equency cha ac e is ic and phase equency
cha ac e is ic o he gi en ans e unc ion o he ma h-
ema ical model in ele a ion.
The o ms o he weigh ing il e s o he gi en sys-
em a e as ollows, i.e.:
•weigh ing ans e unc ion o e e ence signal:
Wcmd (s) = 1
0.25s+ 1,(33)
•weigh ing ans e unc ion o low- equency dis-
u bance signal:
Wd(s) = 0.5
0.1s+ 1,(34)
•weigh ing ans e unc ion o high- equency dis-
u bance signal (noise):
Wnoise (s) = 0.01s+ 1
s+ 1 ,(35)
•weigh ing ans e unc ion o con ol e o signal:
We(s) = Ke
1
Mes+ωbe
s+ωbeεe
= 0.001 s+ 0.5
s+ 0.0005,(36)
•weigh ing ans e unc ion o manipula ed alue
signal:
ˆ
Wu(s) = Ku
s+ωbu
Mu
εus+ωbu
= 10−7s+ 1
0.01s+ 2.(37)
The weigh ing il e s o he con ol e o and manip-
ula ed signal ha e a p esc ibed ans e unc ion o m
acco ding o [14], he ans e always con ains he same
nume a o and denomina o o de as o ensu e he s a-
bili y o he in e ed ans e unc ions. The il e o
he con ol e o is low-pass and o he manipula ed
alue i is high-pass. The g aphic dependences o he
sensi i i y unc ions and in e ed ans e unc ions o
-60
-50
-40
-30
-20
-10
0
10
-5 10
-4 10
-3 10
-2 10
-1 10
010
1
-90
-45
0
F equency ( ad/s)
-150
-140
-130
-120
-110
-100
10
-2 10
-1 10
010
110
210
310
4
0
45
90
F equency ( ad/s)
Magni ude (dB)
Phase (deg)
Magni ude (dB)
Phase (deg)
Fig. 8: Ampli ude and equency cha ac e is ics o he low-
pass weigh ing il e We(s)(uppe pa ) and high-
pass weigh ing il e Wu(s)(lowe pa ): ma hema ical
model in ele a ion.
150
100
50
0
-50
-100
-150
-200
10 10 10 10 10
-4 -2 024
F equency( ad/s)
S
T
1/W
u
e
1/W
Singula Values (dB)
Fig. 9: Ampli ude equency cha ac e is ic o he sensi i -
i y unc ion S(s), complemen a y sensi i i y unc-
ion T(s), in e ed ans e unc ions 1/We(s)and
1/Wu(s): ma hema ical model in ele a ion.
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 495
CONTROL ENGINEERING VOLUME: 12 |NUMBER: 5 |2014 |DECEMBER
he ma hema ical model in ele a ion a e as s a ed in
Fig. 8.
One o he gene al equi emen s o he ampli ude
equency cha ac e is ics o he sensi i i y unc ion
S(s)and he in e ed ans e unc ion 1/We(s)de-
ines obus beha io :
∀ω∈R:|S(jω)|≤|1/We(jω)|=
= 1/|We(jω)|⇔kWe(s)S(s)k∞≤1.(38)
Figu e 9 implies ha he ela ion Eq. (38) is ully
me . Simila ly, i is also possible o de ine he ampli-
ude equency cha ac e is ics o he in e ed ans e
unc ion 1/Wu(s)and he p oduc o he con olle
ans e and he sensi i i y unc ion K(s)S(s):
∀ω∈R:|K(jω)S(jω)| ≤ |1/Wu(jω)| ⇔
⇔ kWu(s)K(s)S(s)k∞≤1.(39)
150
100
50
0
-50
-100
10 10 10 10 10
-4 -2 024
F equency( ad/s)
KS
S
1/W
u
e
1/W
Magni ude (dB)
Phase (deg)
Fig. 10: Ampli ude equency cha ac e is ics o he sensi-
i i y unc ion S(s), p oduc o ans e unc ions
K(s)S(s), in e ed ans e unc ions 1/We(s)and
1/Wu(s): ma hema ical model in ele a ion.
Figu e 10 implies ha he ela ion Eq. (39) is ully
me . Fo comple eness, we gi e he alues o he key
H∞no ms, i.e.:
•op imal H∞no m:
γ= 5.752010−4,(40)
•closed loop H∞no m:
kF{P(s), K (s)}k∞= 4.783710−4< γ, (41)
•sensi i i y unc ion H∞no m:
kS(s)k∞= 1.5357,(42)
•complemen a y sensi i i y unc ion H∞no m:
kT(s)k∞= 0.9997.(43)
The o de o he designed H∞con olle o he
ma hema ical ele a ion model co esponds wi h he o-
al o he o de s o he indi idual elemen s o he ex-
ended sys em, he nominal sys em is o he ou h o -
de a he mos and all i e weigh ing sys ems a e in
he i s o de a he mos . The a o emen ioned implies
ha he con olle will be a sys em o he nin h o de
a he mos . I s equency cha ac e is ics a e shown in
Fig. 11.
150
100
50
0
10 10 10 10 10
-4 -2 024
F equency( ad/s)
180
90
0
-90
-180
Magni ude (dB)
Phase (deg)
Fig. 11: Ampli ude equency cha ac e is ic and phase e-
quency cha ac e is ic o he H∞con olle o ma he-
ma ical model in ele a ion.
Acco ding o Fig. 6, a model o he H∞con olle
and he ex ended sys em was c ea ed in Simulink. The
con igu a ion o he g oup o ex e nal inpu signals is
as ollows:
• e e ence:
d1( = 0) = ( = 0) = −0.25 [−]
o ∈(0; 20i,(44)
d1( ) = ( = 20) = −0.05 [−]
o ∈(20; 100i,(45)
•LF dis u bance: no included in he model,
•HF dis u bance: band-limi ed whi e noise wi h
powe o 0.00001 [W] .
The esponse o he modeled ele a ion sys em o e -
e ence ~
d1( )is as shown in Fig. 12. Thanks o he
balanced a io o he sensi i i y unc ion and he com-
plemen a y sensi i i y unc ion, he elimina ion o he
noise and dis u bance will be e ec i e.
3.4. Ele a ion Con olle o Real
Model
We will use he ans e unc ion om he ma hema -
ical model o he design o he con olle o a eal
c
2014 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 496