n91-, 1206 - nasa · means is needed to preveul the experimem from driftin,,_ inlo ils own...
TRANSCRIPT
N91-, 1206
MI(3{OGRAVITY VIBRATION ISOI,ATION: An Ot'TIMAI, (',ONTI_.OI, LAW
h_r Like ON I',-D1M F,NSIONAI, ('.ASE
R.ichard I). Ilampton
University of Virginia
Thornton Ilall, Mc('ormick Road
Charlotl,esville
VA 22901
Carlos M. (; rodsinsky
NASA Lewis R,csearch Center
Mail Stop 23-3
CI(,vcland
OII 44135
Paul E. Allaire, David W. Lewis, Carl R. Knospc
University of Virginia
l'hornton llall, Mc(',ormick Road
( ',harlol, l,csville
VA 229111
413
https://ntrs.nasa.gov/search.jsp?R=19910011893 2020-05-09T11:19:04+00:00Z
:\ B S'T 1,_A( _"1"
('ert,aiu exp,,rinlout, s contemplated for space platfort_s tt;_t_t I., _ ]>_,;,I,t_,,t l'tL.n,
tile accelerations t,f the platform, in this paper au optimaJ active (:out, rol i
deveh)pcd for microgravity vibration isolation, using constant state feedback gainl_
(i(tentical l,o those ()l)l,_dx_(,d from the l,inear Quadrati(. Regulator [LQR] ai)l)V( ach,
along with constant fee(lforward (preview) gains.
The (tuadrati( cost function ['()r this ('ontrol algorith_ (,lfectiv(qv _.i..,]kI.,.
oxl,erm_l a('('eleralitms of the platform _list, url)anwes I)y a fact,or l)toporliu_at t,J
(l/w) 1 I.t)w rre(luOttcy a('(:el(;ratAonts (h,ss t,hau 50 Hz tu'e .tlt, miate(t bv _,t, _+t+,t
titan t.wo ordors of ma_t_itu(le. The ('ox)tro[ relies on the at)solut(, l)o._it,i()tL a_+i
veh)citv [o(,dl)a(.k o[" th(, ex,,rim(,1_t and the absolute po,,qtion and v.l(Jcilv
fcvd-l()rv,,,azd ()f _1_(' 14.tfc)run. atL(l gonerally (leriv(,_ the _t_abilitv rot)u>:tl_,_:.
chaxa(t(,rist i('s ,_l,n:_nt('o<t I)v th(' I,QI{ ,l>l)loa.(h t()Ol)t_i]ua]it.,,;.
+lh,, [t_t+tho, i _s ,h,tivo_l is extotl(lal)l(+, to the ca,_(' it_ whi('h t_u_lv tho r_4ati',(,
l>osiliOtlS all(l \ol{)<'il it's attd lho al)soltlt_' a('coh'rations of th(, ('x[)('[illi_'lll all(i s[);l(,'
l>lat, lt)rnl are a.vailal)lo.
414
1. INTI_ODUCTION
A space pla.Lform experiences local, low frequency acceleratiot_s (0.01-:It) l tz)
(lue to equipment motions and vibrations, and to crew activity it]. Certain
experiments, such as the growth of isotropic crystals, require an (mvironment in
which tile accelerations amount to only a few micm-g's [2]. Such an environment is
not presently available on manned space platforms.
Since the experiment and space I)latform centers of _favit y do not coincide, a
means is needed to preveul the experimem from driftin,,_ inlo ils own ¢>rbital molion
and into the q>ace l)lal ['Of'Ill wall. Additionally, sonm <',:perimenl,s require umlglical_
to f_rovide l>ower, experiment c_mtrol, coolant flow, communications linl,:z_o, of
otl, r services. I_nforl_umtely. such t_easures also mean that unwanted l>l_torl_
accelerations will I,e trallsmitted to the experiments, l'his necessi_ales exp<imen[
iscflation. I'assiw' is_lators, however, cannot coml)ensale for umbilical sl.iffness, nov
can they achieve low _,_ouch corner Ir<,quencies even if umbilicals ar_' absenl.
Active isolation is therefore essenlial.
The pmhlen_, lhel|, is tO design m, aclive iso[alioll svste[ll tO mini_,,dze these
imdesired accetera, ion t ransmissi_ms, while aclth,vin_ a<leqxlale slabilily Imtr_ins and
system rot)ustness. Spatial and <o_,l rol energy linfitalions n_ust also t,,
a ('('()in ii1o(1_tt e(-[.
415
2. MATIIEMATI('.AL MODEL
The ._eneralI)roblemhas thrc_,translational and three rotatiCm_tldegr<,esof
fre('(lon_. For simplicity, however, this analysis will (onsi(l('r oI_l_ 1.1_(,
one_tinwnsioual probh'm. The general problem could be treated il_ an analogous
maturer, l,et the experiment be modeled as a mass m, with position ×(t).._\,_sume
that the space station has position d(t), and that umbilicals with stiffness k and
d_ttnl)ing c conm,ct the experiment and space station. Suppose further that a
iI_a_neli(" actuator appli('s a. ('Olltrol force proportional lo the appli(,(t (:utrecht i()).
wilh l)rOl)Ort.ionldil v (:onsl_lJtl <_<.Such a K)_odel is shown ill l:igure I.
The syste))) ('(lu_)ti())_ ()f)))()tion is
lltX ÷ C(.X--<'f)+ k(x--(1) 4 _)i = 0 ,!
l)ivisiotl I)v I11 at|(I t_'at'ral_g('lll('llt yi('l(ls
._ = k (x-(t)- (' (._-<'1)- ++i t"-l)-S i_ ))-i
[I_ state.' space _oI _,tt ion thi_ becon_es
W h(q(?.
: .\ x +)2u + f (3
x'tf't f 't
A ,: I< (I 0
, 12= c_-- [_
,116
u=i ,f=_ {°tkd+C_
The objective is to minimize the acceleration x(t).
417
:_. OPTIMAL ('()NTI{()I_ PROBLEM
'File optimal control problem is that of determinin_ lho
)l(t) --- i whi('h minimizes a. :uitable performance index
'OI/l IO[ ('11 ['I'OI] l
.I = J(x, u, 1,) (,1)
l'oz the sysWz)) des(zib('(l by Eq)). (3) subject to the state variable c(>z)di)ioa,_
--()
[ i Ill X(t ) _-<(_) (51j)
:\n()ther rcas(>nal)h, _ssutlq)t.ion is tl,at t_(t) is Ixml_(l(,(I. a,_(l it will b(, fi./l_,i
tuath(,mali('ally a(Ivai)la._(,()its (ai_(l ()ulv)tfiuiulall.v i(,._i rictiv(,) ).() _LS:lLIllCt,hal l_! I i-
I i I_l 1-(1)= L) • i( i
:\ (tm_(lrati_ l)('rfor_mmcc indt:x
.I=.1/ ix I'WIx+ w:lu- ](ll,-0
:118
has I)('(,)[ chosen, _ls oue that, lends i(,._(,If well to the radar iom(I ai)pr()ach t.o Ol)i il_)_
lh( ,,i)i)(,[' lizlfi( ,.)f (,he (lefi][il(,('(mtrol:. since an amdyl.ical s()luIio_ is desired. '" '
iutegral has been :(,h,('ted so a._ t t) yiel(I a time-invariau( c()ntroller, ller(, W ! i: ,_
square 2x2 (onsta)_t w('i_htin_ matrix wlfile w:l is tt w('i!_,hl.iug coll,'..;131tl.
Although, \V1 couhl bea full 2x2 matrix, for this probleln a ,liagonal l'ort_t
has been employed for the sake of simplicity.
[w 0 ]WI = la
0 Wl b
r]'ll{' perfornliuR:e index cottseqllently reduces to
I[ ''x" 2 . 2 ".I = q ) [WlaX [ 4 WlbX 2 + w:lll-]dl
-. {)
so lhal _'a.('h ::tare is weie, hu,d imlepend_,lll.ly.
If sinusoidal motion o1" lhe experit_ont is considered, so lhal.
x(I,) --: I_ si. _¢
,)
alld i.i(l) = _.'-x(t i, lhe cost ['utwtion can I)e exlm,sse_l iJi 1_,:]3_,,,cd ltle ;l.cc_,]_,vat}_ii
;|lid ('01111_] /-I,'q
I {>':; wla -+- w lbl 1'_2'_:+ + ,.,,':lu23 ,_l_.I--:3 ( I --qi, ¢9,
It is apl>ar<,nt thal ll+is l.,_,rlorljlalw_, il++]ex co+_,,:,_,l+J_,lll],+,.,,+,i_}lt:-;;t<_._,]+,r;tti,.+t+ ,_1 ,,.,
fre,+t_w'twi+"sI_ul<'h mot<, 1ha, at, higher freqttett,'+h,s.
ORIGINAL PAGE IS
OF POOR QL_ALITY
_19
.1. SOI,('TI()N
Fin(liilg the optimal control to minimize Eqn. (4) is a variatioli_tl l)rol)iem of
l,agrange, for whi('h the iuitial steps of the solution art, well-kll()wli i('._,., l']lbert
[4]). The va.riatiomd aDl)roach is out, liiled below, fi)llowillg which t lw ,()l_l)li(_,I lolls
a(hh'd I)y the nonhom()g('neous t.erm [(t) will I)e addressed. (',lrre,_) Ol)i.i))_al
controls texts either assume that _f(t) - [) (e.g., [4], t). 262) or ro<tuire that it l_av(_ _
restri(:te(! range space (e.g., [6], p. 238). The solutiou that follows l)rovi(h:,s aH
amtlytieal ()l)lh_al with¢)ul imposing such restrictions.
The a.i'_lliil(,iit ()t" the (:osl tum:l.ion .l t'r()ul !((11_. (4) is alt_lll,'l_r_'(I 1)v ll_,,
l.a_ralt_¢' mllltil)li('r A tilu(,:_ the svsteili e(liJation ()f motiou E(I u. (;11 wt_('r¢,
'\ 1A= A.,I(J
lhe result .J cart I)(, ,,xt)r('ss['_l a._
= li _[II)
wl_er_' lhe llamiltol_ia, u il is
I!
= _ "' , ATtt _ (x r \v 1 × + w:l_') _- (_- ..\x- L,,_-{)
It is (l('sire(l i.o ot)t_dtl an ol)limal s()luti()u u = u
Th,' Iirsl. variali()ti (,f J(x, li. _) is
which lliiililliiz('s .J.
1'2
t:J = ['_' foil _)It 011J
420
which is set -qual _o zero to minimize, J. l]owever, imoarating by parts,
j'0( J_ ",.
ooi) l l _) dt = _ ,__1_. dtilx 0
so that, the al) _ve expression for b.l be(x)mes
/ij }.c,x,[ 0II= ,laH _ "_'r)_._+ _,}dt- 0• 0
(13)
l'h)l h bx _md bu a.r(' a.r/)itrary ,,,'a,ria, ti(ms, s() LJ = 0 only if
81! "Ti l4a)
1 tl)_
lhe conditions ,zi,.en I)y It(In. (5) Sl ill apply.
SOlving '[.tins. (l'ta) and (I,11)) yields
_=wt x-A__ (15a!
* 1 TIt ...... I) ..\
v,.3 - _(1,51)
oy Poor (_u/_rw42I
Temporarily elimin;Lting u t)roduces the result
._ x f(1(;)
W hOl'("
A
-A T
If I"qu. {1(;) is uo_ s,)lv_'_! I,_r ,__in l,,rl_l._ ,fix _n_l 1:. I&llL. ( 151)ii will II_,l, tll_i.-.l_ ;L_
exl)re,_sion for the oI)lilLlal ('onl,r()l u .
..\s n()l,e(I I_el'orp. Ol)limal (()lltrol I.OXIS _ouerally tl('a.[ the homoa_,neous
pml)lem (where f(l) _ ()), but lhry do uot provide _tll _tualylical s_lution _o l tl_.
nonhomogt!neous ,_vstelll _lesclilwd by (5) and (16). ,S_dukvadze has lreat,,d t)l_.,
nonht)mog.n_'OlLS l_I'ol_h,_ [.l,5j, t)ul. his _lifl'i('ult ll'(';ll, lIl('tll ,_'¢'IIIS l_vgeI5* _,, h_w
remaiut'd either uu_co_l)rel=ende_l or _nutl_'r-al)l)reciat_'tl. This lne ho¢l is _'_:!_,._i,_i!'.
well-suiuxl Io low-fre_lUe]_c.ydist.url)_mct,rejection,and h_,_b_'_'uapl>li_'_ll_t.h_w,_
t,l_(? I)['(,,,,;CI_Ii)rol,l('n_.
The homo_,or_eottssolutionto Eq.. (15),where f = LI,is
I ,l ,l (17_
422
ORIGINAL PAGE IS
OF POOR QUALITY
The four eigenvalues of A may be found to be, in ascending order of real parts.
-/71 + (/_l 2- 4_2 )1/2(18a)
,)-;71 - (;fl" - 432 )1/2
'2
ilL'
(tsh)
IL3 = - iL1(iSc)
#1 = -H2(IS(l)
where 171 and ;}'2 are defined as follows:
and
2k31 = __m
(' "2 a w 1 b
in'2 m w 3
,)
39 = ;Jl'-,1 ,) + --_" in-w 3 lii"
19a)
l!)b)
The eigenvect, ors of A corresponding to the respective eigenvalues itk n]ay be chosen
to be
l
Irk
") 2 ._l(.,t2 + it.k )7.1 'l + z _Pk = __ + ---Irk _31tk '?3
"t1 + ('72 + Irk)It k
(20a)
423
where 71 , 7.2, 73, and 74 are defined below:
(20b)
(2oc)
,}
7,3 =: 9-Ill W3
(20d)
'I Ia(20e)
l:sing Eqns. (18) through (20) with (17) the solution to the homogeneous sv,,,,,,iK: is
i_1w. llgt -It 1t -ltgt ]
cle El + c'2e_ P.> + c3e-lt, l t -i_.2 t
h { "2 "'2 P:'2 +cle PZl2]
(21)
t Ekt I ' k = 1..... .1 alltl where c I ..... c I at,' /trt)it:arv c_)ustauts.with 12k = L Pk., J
.,\ppiica, lioll ill' the vitriltl.iOll ill' t);t.I'a, IIII'I_TS IlJellu)d wi:l_ Iel'lliilLa] _rJ¢lit]ol:>
(Eqns. 5b.c) leads to the gOllOl'it.i solution of t,he iio11- tllllOg('llO011S S}'SI,{'_I/I, wit h tx_o
constants of integration yet undetermined.
If the two constants of integral ion are eliminated by solving for _ in l erms :>f
x and [, the general solutions for k I and '\2 become:
424
-1_ 1t --ttgtAl (lXi + (,>x,> + (3 e + "4 (22a)
-ll. 1t -1,.2 t"\2 = _sXl + (6x2 + s_7e + so8e (22b)
ORIGINAL PAGE ISOF POOR QUALITY
in which tile _i's are functions of the eigenvalues and eigenve(:tors ()f :\, and of tile
disturbance .f(t).
The Solution Form
Using the fact that
* ]. '_
u (t)= _.3 AIb__ [cf. Equ. (151))] (_:3)
the optimal control is found to be
. q:}o-/.q tat t -l_2t f elt2tu (t) = qlXl + ,l.2x.2 + , 'Itl f.,(I)dt + q,1e f.)(t)dt
where
(2-1a)
--m kql = --7 (m - ltllt2) (24b)
--[11 C
02 =--_(_+1_ I +It. 2) (21()
m t 2 (: k (21(1),/:)= 'a (_)(/'t + _/q + _)
I 9 k_ rn,_ )(.,.). (,q4 a _ _ +_lL',+_) (_,h,)
(It should be. notod that the [eedba('k _a.ins ql and tl.2 are lhose wili('it w<_uid
result from applying standard LQi{ theory to the hoJnogene()us >3s_ez_ _'(tua,'ion
= Ax + bu). In gqns. (24) ltl, it,2 are the eigenvalues of :\ with ne.g_ttive real
parts, [see Eqns. (18a,b)] and
0_" POOR" Q_UALWY
425
f,)(t) = k ,1 + c' (_Ill Ill
(24f)
By tel)eared application of tile method of integration I)y parts, tile control may be
re--expressed in terms of an infinite sum:
1,,r+l + q4 r_-O /;.)r$i ]LL(t) = qlXl + q2x2 + q3 "r=O
(25)
I{(,writiu_ f.) in terms (ff d iul(I (1. the (:ontr()l function I)e('omes
4¢
+ E (-t) i-I (' ' " )i k ,___qq3 ;/1 ,1 (t(i)(i=1 _( +- )+(-1 _ __7_T+It I It.2 1t ] _.) [-_r{)_
+ [(--l)n-I (: , q:_ q,l )l (1((---_ + n "' n)(t) + hi_h,'r ol<h'r _'rm,_
ltl #2(_,(i)
11
This lila.y I)e wril, len in a tllor(' al)l)('alill_ I'orill as
u (t)=c x(t)+("p v .{:(l,) + ('dO d(t) _- (:(11 (t(t) + Ifigh(,r orcl_,r t,erm:
(27
in whi(:h the constant coelTiciem, s ('1)' ('v' ('(lO' and C(l[ Jn_v I)e del'ine(l t'ro_L t:_(ln._
(24) and (26). harly, if the int'init(, stuns converge rapidly ,,hough, The Ol)tilnal
('ontrol can be apl)roximaJ(,d hy
426
.,g
u (t) = Cp x(t) + cv _:(t) + cdO d(t) + ('dl c'l(t ) (28)
For v(_ry low frequency disturbances tile higher or(l('r t(,rms itl 1";(t11. (25) at-(,
negligibly small, and the control (Eqn. (2¢)) closely approximates the opt, imal. If.
in fat:t, the second- and higher-order derivatives of d{t) are i(lenti(',lly zero, th('
apl)roximation is exact. It can t)_ shown that for the critit:ally damp(,(l closed loop
s3sl._m the eigeuvalues are real alld equal, and that th(, ('owvet',4(,t_c( _ _s _xt()t(, _al)id
than for the overtlamp(,d system. Further, as the clos(,(t--lt_op :vst_,m t,i,_oHv_.tlu(,_
t)4'c(_l_e mor_' uegativ(: l,h(, C()IlVCr_OI|('e S])(_()(l ff,()CS lip _l.,q w(,l[.
427
5. CONTROL EVAIAJATION
Physical Realiz.a.bility of the Control
The control, Eqn. (25), is physically realizable, if the states and sufficien,
derivatives of d(t) are accessible (or estimable by an observer), and if the higheL
order terms are negligible. It is not necessary that the eigenvalues be real. alt hough
the proof of this re(iuires a more general linear-algebra or sta c-lra.,tsition-t,_atrix
approach.
If values a,_, a:_signed to the systent parameters, associated controller _ains
can I)e evaluated. Suppose t,hatm = 100 Ibm, k = 0.3 lbf/ft, c = 0 Ibf-sec/fl. at_ct
+_= t0 lift/Amp. With w 3 arbitrarily set at 1 and Wlb vari 'd. associated z te,-,._,
vahtes of Wla can be found below which the eigenvalues it I ,,nd 1_.2 will +ilwavs b_:
real. Such values atre tabulat,ed in 'l'al)lo 1. Stated otherwise., tho lal,tlal¢,¢l ',alt_¢,s
of the weights Wla and Wlb are those integer values (for the _ak,, of sittq_licity) lot
which t,he closed loop system is closest, to I)eing critically {l.tltq}e<t ,,it hour bein,g
underdamped. Corresl)onding controller feedback and t'eed-(orward _uins (l¢)r t+h_,
first five derivatives) are also included.
The states x(l,) and x(l,) and tit,, _h,rivat.ives ¢ltq_)(l). ,l'l)ll/ ,tll_[ +[(21(t) t_,
clearly availal)h, for a)_ ('ill'lh---])i|se<[ s_.sl(qll. [l(jw(W(,l. ill .-,l,_l(('. lilt, ()llIv /t[,5()l[lVl,
measurements which can be dir('ctly a.va.ilal)h+ are, x(1.) and d(t), itot_t wl,icI_ xiw).
('l(t) and x(t), d(t) are obtainal)le only by successive integration(s). |{earran_,cm,,tlt
of (28) into
or
u (l) = (cp + cdo)x(t) + (c v
u (t) = (Cp + Cd0)d(t) + (c v
+ c.ij )x(t) - c +c>{..<(t, -,I(_ ]-,.,:++,r_cti - d(t)]
+ ,. ilj,'t(t,) + cp[x(t)-dC, )] + Cv[i(t)- a(,)]
-.-, ,
30 )428
obviates the need for one accelerometer, but one accelerotucter plus two iutegtations
remain necessary for either the platform or the experiment. Since [x(t)--(t(t )] (or
one of its integrals) h_ not been weighte(l in the performa/_ce index J, expt,tmmnt
drift will be a problem that must be corrected either I)y another control loop or by a
change of system states. The latter could be accomplished by incoq)orating an
accelerometer attached to the experiment into the. state e(tuatiot_. .\ltetnatively.
oue could append a u integrator to the plant, include the ('urr(,l_¢ i(t) +ts a t l,it<l statue.
and optinfize the contr()l di/dt. But fur the sake o1 simplicit,v (i.+... f(,w,,r .,lm(,s) 1t_(_
f(>tZll_'r )la.s })e(_ll i-ts,'.;IJlil('<l {wil, holt| (levehqml(,/_t) in 1his paper.
The higher order terms of the control [E(lns. (25) and 2_;)] can b(' m,%i('('t(.,+i,
for h)w fre(luencies, if the eigenvalues it l and It2 at(' of sufficient _nod_+l_:. lhr,s,,
eigenvalues, in turn, are under the (:ontro[ of the ([e: i_ner, d(_t(utltim,(l I)v t_i: , tlui((,
of woigllts Wlit, Wll), and w3. It is al)l)ar(,llt frolll ,'.ql]. (25) 111;1t tl Ii) (,_.,(,!_ti;tllv
le(hl+(_s to two alteltm, titLg ])()wer :eries.l,'or a sintlst)idal <lisl_trl);_ti((' (ff It('(lli(_:v _
the :eri(,s form of the t'<mtrol ('(mver ges for +#//_+i+ < I (i = 1.2). It (au be silowti
,X. )r ;+', 2 r ,,that each alternating power series converges like *.L' (-1 ,-) . With +'lowr=f) 'It
fte(llletlcy distt+rban(:es (i.e., small re.lative t() systen+ clos(,d loop ,,]a+,nv:_l:+_,s) +t
(t)l+Irol ['orttl('([ I)v series t, rtltt('a.ti(m vmv ('[os(,lv al)l)roxitttalos t t_(, ()l)t i_.al.
lPor examph,, sut)l)OS[ + t]la! I]1(! _(>rt_mlized ft'('qu_,_(i('s 1.+'//_il i_)r ;t ._it _l_-(+i(I,tl
disturl)att(:e are less than 1/5, and that only the feedfot'ward (t)tlttol t(,tttt._ { !0(ll:
and (dl(](t) are inclu(h,d with th(, fe(,(ll)at'k l.erlllS. |_Vell SO. t+h(' f_+,'(tl'()t_at(t +, rli(,t:
of the trtmcated cot_trol, at any tim(, t. will I)e a ('urrent that is still witlti_t .l , [i.<.
(1/5) 2 of the feedforward portion (>t' the actual ot)titltal. It' Ih(, ttortttalizt,(I
t're(luett(:ies are below 1/10, this <.I.l)[)roxilll+tl,}(Jql (.Tror xvi]l bo l('ss th_-tlt 1'/,,. "labl_+ 1
shows that the gains c(l i t)f higher order derivativ(,,., (lIi)(t) [see E(tn. (2(;) fur
algO)raic representations] are, in fa(-l, quite, small.
+_t,+, .:.+,,+g D_,2- ,,,+-,
429
Ill some eircumsl, ances there may be design ccmstrainls which prevent the
designer from selecting weights that will lead to sufficiently rapid convergence.
ttowever, since convergence occurs rapidly even for eigenvalues of relatively small
modulus (I,.,/Iti] < I/3), in a great many cases the (lesi_oner will have much
latitude in his choice of weights. For "low" frequency disturbances, in these cases, t_
control which includes only one or two f_dfi)rward u_rms will be "close" _o the
optimal. I he:e frequencies will be well-attenuated.
tligher frequency disturbances will also be well-attenuated, provided the
input-to--output transfer function(s) are at least strictly proper iu the l,_tplaco
Transform variaJ)h, s. This will not__._I)e the case for the, l)r_'sent t)rol)h,l_ it _ltor_' th_,r_
c >j are included in II,_'¢'¢_lllml t'lactic,_il_, ,lli._three feedforward _a.ins (cd0, Cdl, el2
means that, only i_rot)ortional and first--derivative feedforward [Eqn. (251 wilh r =
0,1 or Eqn. (26) with n = 2] should be added to the feedback control terms..,ks wilt
be seen shortly, however, adding even the proportiom_l foe_llorward term(s) can
dramatically iml)rove the (tist,ul't)al|Ce re.i('(:tiol| over l,haJ aI'[or_i_'d I)v LQI_ fe_,dback
alone.
Transfer Function and Block Diagram
Neglecting the higher order terms.
otn put accelerations or displacements is
the transfer funct[oll bet, woen [llI)ll_ amt
,_ ('
s-X(s) _ _ (77 - _'_11 ss21)'s)t = (77ts-il7.-2 --+ (7_c
+ (k_ _.¢10
+ c )s +- (17 +
(31)
and a block diagram of the controlled system can be drawn as in Figure '2.
•13o
Control Stability, Stability Robustness, and General Robustness
Since the control feedback gains are the same as those obtained by solution of
the standard Linear Quadratic Regulator (LQR) problem, the closed loop system is
stable and enjoys the stability robustness characteristics guaranteed by (,he (LQR)
approach to optimality, viz., a minimum of 60 ° phase margin, infinite positive gain
margin, and 6 dB negative gain margin [6]. Additionally, nurnericaI checks indicate
that it enjoys substantial insensitivity, or general robustness to _|ncertainties in k, ('.
and m, as indicated by Table 2 and Figures 3 through 10. By comparing the Bode
plots of Figures 3, 5, 7, and 9 (correspon(ling to controls using both LQR F/B and
proportional F/F) with those of Figures 4, 6, 8, and 10, respectively ((:orrespondil_,z
to controls using LQI{ F/I'I only), one ca. see that adding fee(l-f()rwar, l
substantially improves (listm'l)ance reje(:tion at low frequencies, t"or ('xa_@e .
('Oral)arisen of Figures :l with Figuret indicates thal th(, optimal ('onlrol iil(_thod
described above ca_ lead to acceleration reductions of ,grca,ter tha, l_ four orders of
magnitude for all frequencies. This reduction is more tha, n two orders of magnitude
below that afforded by LQI{ feedback al(me at the lower frequencies, i.e., those most
heavily weighted in the l)orfornmuce i.dex.
The order (_t' I,he r(,dut:t.ion is ('V('lltlla.llv Iimit(M [_v c()lllro[ (<)_t. ()t COIlrSO.
I)robably in terms either of a,t',tuator-lelated limitations (such as heat-rcllit)va[ or
force-generation reqtlirenlents) or of power limitations (especially in a space-statioll
environment). The control also leads to displacement reductions of lhe same
magnitude, limited in this case by actuator-stroke or spatial limitations. Providina
a unit transtnissibility for very low frequencies and weighting (xml)and/or f(x-d)
in the performance index ,I wou[d [)e steps toward ad(hessin_ these, latter
limitations.
431
Computational Aspects
A significant amount of algebra was required to solve the two-state problem
of this paper, and the labor involved increases dramatically with each additional
state. Itowever, such symbolic manipulators as MACSYMA may be used to ease
the workload if a symbolic solution is desired. Further, well-known numerical
methods exist (i.e., Potter's method [7] or Laub's lnethod [8]) for solving the
solution to the homogeneous system. These can readily provide the feedback gains
in numerical form, even for problems with many states. It _rlight be anticipated.
then, that a nunmrical m(,thod also exists for finding the desir( 1 feed-forward gains.
Such is tile case, as will Iw shown in a later paper.
432
6. CONCLUS[ONS
This paper has applied an existing method for obtaining an optimal control
to the microgravity platform isolation problem, for which the disturbances to be
rejected are low-frequency accelerations. The system was assumed Io be
representable in the form _ = A x + _b u + f, with quadratic cost function
l, oo Tj = 1.z (x WlX + w3u2)dt and diagonal weighting matrix W 1. The cosultant,0
control law was found to be simple, stable, robust., and 1)hysically realizable.
Further it was shown to have excellent acceleration- and (lisplacemenl.-altet_uation
(ha.racteristics, and to be frequency-weighted toward the low en(l of the a(celeratio_,
S[)('C(i'Hln.
The method is extendable to the case for which only relaJive positions and
velocities, aml al)so]ute accelerations, are available; and can be al)pli('(l so ,s to
weight relative (lispla(:(,n_,nts in the i)erforrnan('e index.
The approa(:h as presented is al_,ebraically
manipulators can 1)(: used to ease the algebraic ,abors.
produces feedback gains identica,! to lhose obtained
intensive, but symbolic
Further. since the method
by the LQR al)proach t()
()ptimality, numerical computation of those gains is easily ac('omplishod, even for
largo s,:s_ems. Th(, f('('(l-forward gain:_ can I)e found numerically with COml)arabl,'
e0.5c.
7. ACKNOWLEDGEMENTS
The authors wouht like to recognize NASA for partial fimdin_ of thi,_ _ork,
Eric Maslen at the University of Virginia for his many helpful control-r(_ ated
suggestions at problem points an(l regarding flmlre work, and Dr. Gerald Brown of
NASA Lewis for his insights into the _nicrogravity isolation problem in an ort)ital
environment.
433
8. REFERENCES
l. llamacher, tl., Jilg, R., and Merbold, U., "Analysis of Microgravity
Measurements Performed During DI," 6th European Symposhun ol]
Materials SciencesUnder Microgravity Conditions, Bordeaux, December
2-5, 1986.
2. Alexander, J. Irwan D., "Experiment Sensitivity: Determination of
Requirements for Vibration Isolation," Vibration Isolation Technology
(,enter, September 28-29, 19,_.Workshop, NASA Lewis Research ' ,"
3. Elbert, Theodore F. Estimation and Control of Systems. New York: Vau
Nostrand lh,inhohl Coral)any luc., 198.1. Chapter 6: "Optimal ('¢)lktml of
Dynanfic Systt,lns."
Salukvadze, M. E., "Analytic Design of Regulators (Constant.
Disturbances)," translated in Automation and Remote Control. Vol. 22, No.
10, pp. 1147-1155, October 1!)61. Originally published in AvtomaTika i
Teh,nmkhanika, Vol. 22, No. 10, 1)1). 1279-1287. February 1961.
Salukvadze, M. E., "The Analytical i)esign of an Optimal Control in the
Case of Constantly Acting Distm'l)ances," translated in ._\ummati()n and
Romote Control, Vol. 23, No. 6, Pl). 657-4;67, .June 1%2. (')ri_ir_;_llv
published in Avtomatika i Teh'makhanika, Vol. 23, No. 6, Pl). 721-7:{1. July
1!)62.
Anderson, B. I). O. and .Moore, .I.B. l,inear Optima! ('ontrol. Englewood
Cliffs, New Jersey: Prentice t|;LII, 1971. pp. 70-71.
Potter, J. E., "Matrix Quadratic Solutions," SIAM .lourtlal of Al)t>liod
Mathematics, Vol. 14, No. 3, May 1966, pp. 496-501.
Laub, A. J., "A Schur method for solving algebraic Ri(:cati ,_quations," IEEE
Transaction_ on Automatic Control, Vol. AC-24, pp. !)13-92t.
°
.
.
.
,
434
Plate
Shaker
k
C
---gsD---Actuator
////////////111/////11//I
xtt)
Figure 1. System Model
435
Tabll, 1. Optimal F/F and F/B Gains for Sele('tod StateVariable and Control Weightings.
Syst em Parameters:m = 100 Ibm k = 0.3 lbf/ftc = 0.000622 lbf-soc/ft ({ = 0.17,)(_ : 10 Ibf/aznp
t W, iKhl s ___
Wla ] wIh l 13,;:1;,,23 ;I
I41 Ill
li.1 5 I
92 fi _ I
126 7
1t15
211!1 '1
5,_t 15
IIl.l.I 2t1
11il7 2"_
2:12!1 lit I
3171 .15 I
11 13 Iq_ I
_i:I_5 i111
1ti5Xl KII
25!11 I 11111
I"/11Gauls
CP
I. 11845
3, 1:1'24
4. 7659
6. :17:12
7. !17111
'1 51117
11 PI'II)
12 _11:1
I 1 121i!t
Ili 11:121
21 u71o
Q. IJ_'l
111 I_1!1
I'_ 12!17
"_h 2_11i
t_ I I:llil
¢v
I. 116117
2.1-113
2._210
:1. 1514
I. 1552
3.7:1711
:1 !19.1!1
I. 2;I;,11
I. Ili7,1
3. 1729
li. ,120!I
[ I11i_11
7 TI.II
_, tblll
I 'lh ]:lhll III. q'_)(i
I ] 12_. 7:172 12,111Z5
I ! .....__ _ llil).!l:lX!t I | I 1417
CdO _ ] I"d 1
O. 02941 -oI
0.0297 I -41.00q11
0.0"2!1_ I 41.1x)q)(1
t). 0'2!1!1 I I1 11114)11
11.112!1!t 11.O111H1
0.112!1!1 I i1.11111/t)
11.112!.19 t1.tl41011
t1.tl2!1!1 ] I1.11111111
11.112!19 It.1tllt111
I1.11'2!1_1 11.I141tttt
tl I1:11111 11 1)11111
n
tt.1t:11111 I I1.11011 [
o .11.1191) il I10(11
l) 11.1111) II I1111J [
11 11;111tl 111111111
1111:11141 t1 tllltl I
11 I111t11 l) 1111111
Ii I):IIXI tl .1111tll
II I1 Itlt1 II 1t0411
l"/t +' li,,,L _ tl _1
1',12 (:d:)
--0.04)70 -41 04_7
_1 ttt1,111 -.11.11t.119
41. o11211 o 1)o10
-.11 Ill) 1:'i _1,111_7
_) I11112 _9 ()005
-0.t101t) 41tltt04
-41 111111X q) .11110:|
_) t1411t7 _1. I)002
-o. 0006 _). Ol)1)2
-41 0o06 -0. uO02
-o. III111,1 -4) 111)(II
-41, I|(111,I+ -1) It11111
t). 1111112 tl. 11111111
-_1 t It 111'_ -41 Ill)Ori
-II i)l)|l_ 41 II(tt)ll
4i itllll I 41 il 1t II )
1t Ill)lll 11 1191t11
111111111 _ll !1141| I
41 1111111 _t clllltll
Cd.l (',15
-0. UO,19 -0 1111:1:2
4) 0009 _1. 0110.1
q), Ol X/'| 41. l)(J() [
_1 04x12 _) 1to111
_1).11ollt 41.t)l)llq
-,11,1)11111 _I iilia)q)
_11, t)IJl) [ -4t t111ttll
-Q.tlttIII -lJ Ill tlt_ )
_J I)1t1111 _! 11111!11
-Q. ti111!1t _1 Illitll)
ql I/I11!11 41 Ill)l!ll
_11 ()()lltl it _luil9
4) lll)ltl) 41 I)lll!lt
_11 11111111 -It lillillt
_j 11111111 41 LIIitlll
41 11111111 !) I)tltllt
41 t1111111 _t ()114t11
411111t111 41 11111111
--II, Iit1111l -1) Illllllt
436C?.:7_:r,_'+,_ PAGE IS
s21)(s)
X(s) ,[_/_/ _ [
Plant
C,)ntrol l_,r
Figure 2. Blot'k Diagram
437
Table 2. Closed loop transfer functions for system with designparameter values of k = 0.3, c = 0.000622, and m = 100; butwith actual parameter values as shown. 61, G3, (;5. _Lnd (;7include both LIIR F/B and proportional F/F; G2, I;4, (;6. and G,_include LQR F/B alone. Weigtlting parameters used were
Wla = 258, Wtb = 10, w3 = 1 (see Table t).
System Parameters
_lbf_ Ib_tseck _tiT-_ c ( )
I). 3
m (ibm)
o.ooo622 ioo(<=().12)
O. 3 O. 00o622 100
O. 45 O. 000622 I00
O..15 O. 000622 100
0. 00622 tO0
O. 3 0.00622 100
O. t5 O. 00622 91)
(). 45 0.00622 90
Closed Loop l"ransfer function
s'-'D(s)
al(s) - 0.0000622s + ().0001.)
O. 31056s"-4. 4675s _ 16. Ot)2 t
G2(s) =
63(s) =
0.0000622s + 0.(_3()0
0.31056s2+4. 4675._+ lb. 0 ?2
0.0000622s + 0.0151
0.31056s2+4.4675s+lG.OTT l
0.1)t)00622s + ().()450(;4 ( s ) - .)
(}.31(}56s-+4.4675_+lb.(}TT t
t;5(s) -0.000622s + 0.00()!
.)0.31056s "+4. 4680s*16. 062 t
G6(s) -0.000622s + 0. 0300
0.31056s2+4.4680s+16.o62 t
(;7 ( s ) - .)0.000622s + 0.0151
(;8 ( s ) -
0.2795(is'+4.4(_5()_+16.(,TT I
(_.000622s + 0.0t50.)
0.27950s-- t. t6S(/s_lO.oTT
438
M 10 -Z _=.-a
gn
i 18-31 ....t --u
dc
le-4 _
1_-5 -2L •
le -6 _
le -7 i I i I iilj
10 -3 le -2
Gl (OFt Ctrl with F/B _ F/F, St}stem Xnovn Exact
\
\\
\\;
J
\\
\
45
0
-45
\
\
le 2 lO 3
Frequenc 9 (rad/sec)
-1 tOO le
Ph4
$
@
Figure 3
439
fl I0 -Za
gn
it I0-31
10 -4 _.
16 -5 :_
10 -6 =
101(_-3
GZ (Opt Ctrl vith F/B Onl_l, _Istem Known Exactl_l)
i u JJlLl _ , , i,,ll
10-1 leo
\
\
\
\
\
\
\
i i | ,llll
10
\\
\
\
45l:c
e
-45
-90
-135
................ 180
10 Z 19 3
Frequency (rad/sec)
Figure 4
44()
rl 10 -2a
g11
1£ 10 -3
U
d
10 -4
10 -5 =
10 -6 __-
10 -7
I0 -3
C_ (Opt Ctr) vitb "/B & F/F0 5 stem $tifrness Estimate Poor)
\
\
\
\
\\
\
\\
\
'\
\
"_ \\"\ \
//
\
/
/
/
/
\
ecL_
-45
-90
-135
-180
................. ZZ5
10 Z 10 3
Frequencg (rad/sec)
FiKure ,5
441
n t9 -Z41
gn
it 16-3
U
d
10 -q
10 -5
lO-b
10-7 , .......
10-3 10 -Z
Ctrl uith W/l OnlM, _ stem Stiffness
\
\
\
\
\
| . i l,l|* i _ 1 II,*J
10- l tO 0
£st lmmte Poor)
\\
\\
\
\ \ /\ /
\
\-
, _ ii,|*lt J t i LllJ
10 1 10 Z
Frequenc 9 (tad/see)
0
-45
-90
-135
-18o
........ -Z25
le 3
Ph&
S
e
Figure 6
442
Ctrl Nitbn9 Estimate Poor)
\
le Z le 3
Frequencg (rad/sec)
Figure 7
443
IS le -za
gM
1t le-3
U
d
le -4
le-5
1@-0
G6 Ctrl with FIB Onl_ Esti_te Poor)
\
\%
9OL_
-9O
'-180
-Z_
Phd
S
C
i i _ ,,l| , , i i,,Jl --360
10 Z 10 3
Fr_luem= 9 (rad/s_c)
Figure 8
444
M 16 -Z =&
gn
! -3t 10 __,,U
d
10-4 _
10-5 :
18-6 _
G? (Opt Ctrl with F/B a F/F; S_lstem k,c,m Estimates Poor)
\\
\
\
\.\
0L_
-45
-7I0 ........
10 -3 19-Z
L i , iJIJl!
-135
-180
, i Lllll , _ l,I*_, I , llJ_LA J L . ILiJJ --_-_5
19 -1 lO 6 101 11)2 193
Frequency (tad/see)
Pb4
S
C
Figure !)
445
N 10 -Z :41
gn
it 10-3:
U
dC
I0 -4 =
10 .5
10 .6 :
10 -?10 -3
GB (_ ,t Ctrl with F/B Only; _rstms k,c,m Estimates Poor)
\
\ \
\ I
L. ' L J,,,, | , i l,,,J i _ . ..J,i i , , sit.,
-Z le - ! lee le le z le 3
Frequency (rad/sec)
0L_
-45
-90J
t
f
_.. -135
\\
-180
........ -ZZ5
Pha
s
Figure 10
446
_ University of Virginia ROMAC
Experiment Compartment
Umbilical Connection
Scientific Experiment in Spacecraft
J447
_ University of Virginia ROMAC
DISTURBANCE LEVELS
Quasi-Steady or "DC" Accelerations
Relative Gravity Freauencv (Hz) Sourqe
1E-7 0 to 1E-3 Aerodynamic Drag
1E-8 0 to 1E-3 Light Pressure
IE-7 0 to IE-3 Gra vity Gradient
Periodic Accelerations
Relative Gravity Freauency (Hz)
2E-2 9
2E-3 5 to 20
Sourqe
Thruster Fire
(orbital)
Crew Motion
2E-4 17 Ku Band Antenna
Non-Periodic Accelerations
Relative Gravity
1E-4
1E-4
Freouency (Hz)
1
1
Source
Thruster Fire(Attitudinal)
Crew Push-Off
44,_
_ University of Virginia ROMAC
_enerolII I I
Probl em ,
In _he upcoming
s_a_ion, planned
spaceNASA
for comple _ion in _-he I?_)0 'S_
accelera s _rans m,_t: ecl1 i I I II I I " -
_rom _he space s_a ÷ i on _o an• _p • ri men _a I pIa_fo rm conga;ned
on (inside) the space $tation.
(¢ . ,,minimize : .reduce _-o -_ 10 .6,f possible
"low _Crt_ucncy": o. oo! _o _0 HZ
4 4 9
_ University of Virginia ROMAC\
Lorcntz
Actuator
MagneticSupport
Experiment MagneticMass
Support
Shaker
Concrete Base
Control Law Validation Apparatus
450
_ University of VirginiaROMAC
X X
X X X
X X
X X
X X_X
1
X X X X
K X X X X
F
K X X X X
X X X X X X XX X X
X
X
X
X
The Lorentz Equation: F = il X B
F = Force " 1 = length of wire i = current
X Represents the tail feathers of a magnetic field B vector into the page
J45]
_ University of Virginia ROMAC
\Sh.tkcr
Lorentz Actuator
DesignSection A-A
Large
Air Gap
,dllllllF_."_llll_coi,l_rCoil --.--411111E_Jr_11_'_- _IIIIIIL_ Magnet
Rod Connecdng
Iron
to Ex_nment
| "_:_=,._[_i__i_h___l_{_-_
MlftllIHI_! _Z_II1lIIilI1lliIlljIjI11liiii_IllI_lIl11IiII1l___F-
Permanent Magnet InsulatingLargeAir Gap
Connecting Magnet Iron PlalePlate
Cross-scclional View of Aclualt)r
\452
_ University of Virginia ROMAC
tJ/i
//]/
]
J/
C on'irollii,"
/]//N/I/I/ IlII/IIIlIlI/I
i"l"l
d(t)
Sys'tem Model
453
_ University of Virginia ROMAC
Objec÷ive:
((Fi nd "('he be s_"ee
minim; ze x ('(').
454
_ University of Virginia ROMAC
\
E Clua 4:ion o4: _ o_ion :IIII I I |II III
•. -. I]x ('_)= m'
\455
_ University of Virginia ROMAC
_S_a'['e Ec_ua'_ions:
ORB
x=Ax+bu+_ cmB n _
x2 C¢)
m
{ o }_o __
456
_ University of Virginia ROMAC
ProblemILl I
S_a_emen _ :I I
"Tr?
Determ;ne ,he control u(_')
wh;ch m;n;m;zes _he performance
indev --- --- -- - - - - - -- ----7
m I w g
concl;_ ;on $
I_or
t$u jec_ _o '(:he
' x o)= xI - --0II l im _x('_} = £I t: .--_oeI
I;m _c(,)= @
11
i
I!IiI
11I
11
11
457
_ University of Virginia
fS o l.,io_ M • _ho_l
ROMAC
( D i "F_e. ren_- ia I
E_ua_ions Approach) :
I. Au_men_ _he performance ;nde. x _T
Lagran_e m-l÷; pliers.
2. Take _he I s_ rattan;on _ _ o_ ±he
•au_men_e.d per_:ormanc.e Jndex and
Se_c ;± e_ual "to zero:
1
_e
_he ar b;_r ory vor;otl onS g X clncl
5.bs_;_u_ for _ in ÷ke st_e
to y ;e.ld
/_ BW;" B'r]f t-[w, +(:+}458
_ University of VirginiaROMAC
x (o)-- x.I;m x(*)= o
•_ ..+, oo
l;m T (,) : o
459
_ University of Virginia ROMAC
Solv, fo_ __(_);. _¢rms of
Such _:_a_ _:h_. rtma;n;n9 n
¢ons_cln_$ are _/;m;na_e_.
_. Use the t_u=_ ;on
"_o 1:;rid u* ;n terms o'F x.
X_ (_:} I_ a mannt
arbl'trary
Result:
: B ×=,X,, x
,,,..<-,,÷ -zz _2,f(*)a_:
! !̂]where iS "(:he .TordeIB
0 -
a._o., ca l Form of t_ Hamil+o.;+,
ma+rix I_'W, BW_"B'J,__,.
white h c.or,ia;ni only +he ne:_.+;ve
elcjenvalue.s of +he Hom;Itonlanw_cl_:r;X, correspondlnq _o the
llgellviill,le/ ot _" _he _lJoi_d-ioop
,y,,-.,, Co,,,.,,,;.,,j iA, Bl<..,,.oli,bl,J,
_h.,. X:[X,X,,X,]x,,;' +_'" .i,..,,.o+o,m-*,-ix which le.a, "to the
alDov+ .T. C, F, )
460
_ University of Virginia ROMAC
|0,
No_e:
w_ere ){ "'= -".y,(o) ( _ )"'tl X11
a._ where
I.*¢Sr-*im 5
|'n_eqra_;on e_ the ;nde_C,'n;_e
;_esral rt_;res constants
of ,_e:_ra_; on _-ha_ ea re e []
ide_;ce/ly zero.
X,.,,,^,_,._.T,x, ^ x"= 7,,, A- ew;'e'Pi11, .... 11 II
deve]op
P,,-X2sX _' [Pis _h,_ so(ution "('o "thewLII- known _(gtbra_c
R;cca't_ I_qua_ion] ;
aqu;v_len_ _ormS _or U ci'('t_:
_.u_*(.e)= - W_"B_"Px + W_'e" ,,,,="_'""e'^")er^_._,,"_"_f(,}a,
= - W._' B=P x - W_"BT ..==:Y_'_",._o(-A-)'"*'X_,_f_'') _''
w;' - -"s*=_¢ _ran.;_c;on ma_r;x approach y;elds
..u*(._)= - W_" B'P x_- W_" B" Y¢'"'_"^+..,_ ,[ e^t"_"_Tx_,__ a=
461
_ University of Virginia ROMAC
So[u÷ion,L
_u*("_ : -W,;' B"P x + W; "'B'''c''_'''z2e'^* ,1"• ^* '_z,"("_-?(,) a-t"
w;'8"P,, w;' B""""" -^"f _"""""
: -w;',"P_- w;'B"x,,"°"__(-^-')"" x'"'__'"rmO ZI :
: - w;'B"P,_- w;'B"_ (-_,-')"*'Pf("_r'=o
W-,B_'Px W3B'r -"" ,)"*' ",'_f..-):- _- -' x,,,.zo(-^- x,,
D,.opl,;,, _ 6;gh.r oraer e.r.,s (_o_ r-o):
¢o)_.'.It.(.._) tlt-lnT_ i¢,.-I nT_l(-l)'l_-I _!("1_ ,.-w_ _ rx_ +vv s o 1,zz /_ _¢, S
-. ,11--1 aTn till "1 D T V ('1)'1 A -I V ¢.1} P '_- V 3 O r _X + vv3 o ^zz _' "zz ""
-I T -m T -'r -'r T-W_ B Px 4- Wa B X,, A X.P-¢_
-- -I T -I 1' -T -1" TW_ B P __÷ W3 e X,, h X=,__
wh,re _.T= -P(Ph ÷.W,)"= (_-BW3'BTp) "T
Lx'-',,^ x';' '_ " "'"" "" ""''_= ) " _'zz i_ ^zz
These.are. several "forms for "the co.trol law.
J462
_., University of Virginia ROMAC
\
Solu't'i on:I I!
(-w;'B'P)+(w;' p)
_ Univer$ity of Virginia ROMAC
A Jica_ion _'o our
specific problem :i I
[ (.e) Cpx ¢_)÷ Cvk _,)
+ Ca a¢,) • Ca a (_)0 I
where Cp_C _, CcloDand Ca,
are con $_'a n _" _clinS.
\4fi4
_ University of Virginia ROMAC
S_ D (s_
F-
IJI
D C.s) ._
It
I(
J
Block Di,,5_m
465
University of Virginia ROMAC
\
Table I Optimal F/F and F/B Gains for Selected StateVariable and ontrol Weightings.
System Parameters:m = I00 Ibm k = 0.3 lbf/ftc = 0.000622 lbf-sec/ft (( = 0.1%)a = 10 lbf/amp
_, igh I,:',
_la[ _lh
---4-
2 I
I0 "2
23 I 3
41 4
64 5
92 6
12ti 7
165 8
2{}cI !J
25,'_ IO
5_1 15
IO3,1 20
I(i17 25
2:12q 30
;| I TI ;|5
,I 1I:1 II)
!I:1:15 60
165141 t_O
25911 I00
--v_
w:I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
t
I
I
F/IfGa iu,_
(', Cp v
I. 3845 i. 3637
3. 1324 !. 9863
4.7659 2.4413
6.3732 2.8210
7.9701 3. 1544
9.5617 3.4552
Il.1950 3.7354
12._115:| 3.9919
14 .,1'21;'_ .I. 23_0
16.032.1 ,l.1674
24. 0740 5..1729
32.12S9 6, ;120!!
I0. IKI9 7.0680
,1_.2297 7,7,1:11
56.2_1(i t_, 36,1(i
Irl .:1:¢61 _.9420
96.5:lfJfl IO. 9.r_26
12g.7372 12.6475
160.9389 14. 1407
("dO Cdl
0.0294 -0.0006
O. 0297 -0.0001
O. 0298 -0.0000
O. 0299 O. 0000
O. 0299 O. 00O0
0.02_J O. 0000
O.0299 O. 0000
It. 0299 0. 0000
O, 0299 0.O0_X)
O. O299 O.0000
O. O3OO O. 00O I
O.0;100 O. 000 1
t) 03011 0.0001
0.0300 0.0001
0,(;300 0.0001
O. 03(X) O. 000 l
O, 0300 O. 0001
0,0:100 O. Of)OI
O. 0'300 O. O(.HJ 1
I-'II"(;ai,s
Cd2 Cd3
4). 0067-0.0070
-0.003O
41.002O
-0.0015
-0.0012
-0.0019
-0.0010
4). 0007
-0.0005
4). O0O9
4). 0004
4). 0002
-0.000 i
CdI Cd5
4). 00,19 4). 0032
4). 0004
-0. 0001
-0.0001
-0.0000
-4). 0010
-0.0008
41.0007
-0. OOO6
-0. 0006
-0. OO04
4). iX)03
4). iX)02
41,0002
41,0002
4J. [KI(Jl
4}, o0o I
-4),o0o l
4). 0OO I
4). 0004
-0.0003
4). 0002
4). 0O02
4). 0002
4).0001
4), 000 I
4). 0000
4). 000O
4). o000
4}.(}(M_,J
-0. o(,_0
4),0000
4). 0000
4). Or)' )1 -0. 0000
-0.00( t -0,0000
4). 0001 41. 0000
-4). OffOO -4). 0o00
-4).0000 4).000o
-0. 0000 4). 00(_}
-0. 0000
4). Of)O0
4J. (l(_()0
-0.0_00
4). O0 }0
-o, f)ooo
-0. 0000
-4) 0000
................... i
)
-0. o<XX)
4). 0090
-0. (x_)o
-0,0000
4). OfX}o
4). 0000
-0.0000
-0. _X)O
466
c +,
University of Virginia ROMAC
Table 2. Closed loop transfer functions for system with designparameter values of k = 0.3, c = 0.000622, and m = 100; butwith actual parameter values as shown. 61, G3, 65, and 67include both LqR F/B and proportional F/F; G2, G4, 66, and G8include LqR F/B alone. Weighting parameters used were
Wta = 258, Wlb = lO, w3 = 1 (see Table i).
System Parameters
, Ibf_k t tirC-_
0.3
0.3
0.45
c (Ibftsec) m (Ibm)
0.000622 100
(/:o.1%)
0.000622 100
,e . ,
0.000622 100
0.45 0.0006'22 100
al(s) =
Closed Loop Transfer Function
s2X(s)s2D(s)
0.0000622s + 0.0001
0.31056s2+4.4675s+i6 0524
G2(s) -
I
G3(s) -
O. :3 O. 00622 100
0.3 0.00622 100
0.45 ).00622 90
0.45 ().00622
G4(s) -
0.0000622s + 0.0300
0.31056s2+4.4675s÷16 0624
0 0000622s ÷ 0.0151
0.31056s2÷4.4675s÷16 0774
0 0000622s , 0.0450
O.:31056s2,4.4675s+16 0774
0 000622s , 0.0()01Gb(s) -
0.31056s2÷4.4680s+t6 0624
0 000622s ÷ 0.0300G6(s) -
0.31056s2+4.46_0s+16 062_ i_, , , ,, I
G7(s) 0 000622s _- 0.0!51.-)
0.27950s'-,4.46SOs*i6 077
90 _8(s) -0 000622s * ().0450
.3
0.27950s'-4.46SOs÷16.077_
J
O_IG!N._L PAGE IS
OF POOR Q4JALITY
467
University of Virginia ROMAC
M i0 -z(I
gn
1t 10-3 :
U
d@
I0 -4 _=
10 -5 =
le -6 =
10 -?10-3
Gl (Opt Ctri ulth F/B 41 Ir/]F, S_stem Knmm I_xactll )w
f
\
\\
\
\
j_
i i 1 JJ|,_, J a m ll_Jl
lO -1 10 0 19 1
\
\
\
cL_
\
\
,\
\
45
-45
-99
-135
........ 186
10Z 103
Frequen_ (tad/see)
Pha
s
e
Figure 3
J468
_ University of Virginia ROMAC
M le -z
gn
it 19-3
U
d
19-5 _
19-6 _..
I0-710-3
GZ (Opt,.Ctri
i l , , i1,, i l i i i1,1
19 -Z 10 - 1
with FIB Only, _Jstem i{nown £xacth
\
\
\
\
k
\
\
\
lcc
lo I
'15
-45
-9O
-135
\\
\
\
I0Z le3
Frequency (red/see)
Pha
S
e
Figure 4
\469
_ University of Virginia ROMAC
II le -Z8
gn
lt le-3
U
d
le -4
le -5 =
le-6 _:
le -?10 -3
_3 (opt Ctrl with F/B S F/F, _ten Stiffness Estl_te Poor)
'\\,
\\
\
\
\
ecL_
-45
-90
\
_,. X -135
-18e
.................. ZZ5
10Z 10 3
Frequent V (rad/sec)
\
\
\
/
#
/
/
/
/
\
\
x
\
, _lll , l , ,lit| i , llllJA
10 - 1 190 le
Ph8
S
C
Figure 5
)470
_ University of Virginia ROMAC
M 10 -z =A
gn
1t 10-3 --U
d
le-4 _
10 -5 _
10-6' ..
10 .7-3
IO
64 (Opt Ctrl with
* i Jl ,i*Jl . • [ JILL
10 -2 10
ir/B Onl_l, _ stem Stiffness £stim_te Poor)cL_
\
\
\
\
/
\
\
.... 45
\
, \\
\\ i
\ /
\
19 1 10Z lO 3
Fr_-_luency(ead/sec)
Pha
s
f_
Figure 6
471
_ University of Virginia ROMAC
M le -za
gII
it le-3
U
d@
le-I
le-'5
le-6
G5 (Opt Ctrl with F/B a F/F,
-3, J A ,Jl,[ l J i J lli_
I0 -Z 10-1
I
I
I
./J
J
S _tem Danp
'\
\
P
jl/,
10
imj Est lut,
\
\\
Poor )
\
, , L lJ_J
le Z
Frequencg (rad/sec)
9e
e
-90
-1Be
-Z'_
16 3
Figure 7
\472
University of Virginia ROMAC
II le-z =A
gn
|t le-3 -
tl
d
le -4 _
le -5 _
to -6 =
le -?le -3
G6 (_
l i i i_., i • _ _*Jll
IB -Z 19
,t Ctrl with F/B Only. _stem Danplng Estimate Poor)
\
\
\
i _ _ [li._ i . ll,,l_ l . , ,.ll.
1 100 10 1 102 103
Frequenc g (rad/sec)
96cL_
0
-186
-Z?e
Ph4
S
FigHre 8
473
University of Virginia ROMAC
el le -Z =
gn
tt le-3
U
dC
19-I i
le -5 =
le -6 =
7le-7
I0-3
G"/ (I]
19 -Z
)t Ctrl with F/B A F/F; 5_stem k,c,m Estimates Poor)
\
\
\
\\
"\\
\
, J i _,,A|
101
0l:c
-45
L
"_ -135
\,
-180
.......-ZZ5
10 z 10 3
Frequencg (radlseC)
Fi_,ure 9
474
_ University of Virginia ROMAC
II le -z =-&
gn
1t 10-3 -
U
de
le -4 =.
le-5 :
r-
19-6
10-710-3
68 ,t Ctrl with F/B 0.11 k.c.n Estimates Poor)
"\
\\
\
\
\
cL_
\\
"\
\
\ / f
",,. /
"b
I
\
\
s i z i,[,l , i , , ill,
- 1 100 le I
-45
-9O
-135
- 180
...... -ZZ5l L i i,_J
lO 2 lO 3
Frequency (rad/sec)
Pha
s
Fi_zre tO
47,5
_ University of Virginia ROMAC
C onclu sions,[ I li
II
4
e
A n op't"imal con't"rol kin, ID..en+or
An =pproxl'ma_;on +o +hl's
has been found which roses
de_c erm ine_l
+he n°nhomosentou, LQ R problem.
OF* i m=l ¢on+rol
"FeLal belck
Th_ op't";m= Iadvan+ag_s :
and _eeal _or wclral_alns.
con+rol has +he _:+llow;n_
• I
=. The :_n,.s c=n be eal;ly ele_'ermlne_.
b. The, control ;s very robust ((;O=ph=seB e 0
malr:_lnj ;n_;ni+e posi+;ve _oln vnar_an)
• . )C. "rht con+rol is =pplicabl_ _o = wide
ran:_e of' problems.
el. Th= con+ro[ o_c_ers subs+an++=l
;mprovemen_" In dlS_:Ur_elnce rejection
ove.r _=_C _l++orell_.d by LQ R Fe¢=[loa_k
0
e. _¢ =o._roi can be e=s;Iy Impl_-mP.n_¢cJ.
476