Download as pdf
Download as pdf
You are on page 1of 9
1100 Robust Stability Under Structured and Unstructured Perturbations HERVE CHAPELLAT, MOHAMMED DAHLEH, wwe, rs, ax ‘SHANKAR P. BHATTACHARYYA, rsttow inte Astract—In his paper, we dea wth the problem of robes sabty {or tner time-invariant singh-nputsingle-otput contol systems sub lect to both sirectred (paramere) and wtruteed(J.) perturba: ios. We fist preset» generalization ofthe small gn thorem which taster function ceticiens vary kn peseibed interval) eee, ad er, they sed ght on the geometry Perturbations of mite pe ae preset. ‘the sabiley domain when I. Inrncpecrion Ti sy of stably rotates in the presence of uncer tainy’ isan important aspect of contol system analysis ‘Modeling uncertaites affect the system in many different vu degradation of the system performance and destabilization are the two most important adverse effects, There aze many kinds of luncertainties that ean be present in given control system, Much ofthe rescarch that has Been done so far centered om uncertain- tis ofthe unstructured type. This (ype of uncertain is usually ‘modeled as norm bounded perturbations in an appropriate normed! algebra [2]. For this type at uncerainy the stady ofthe stability ‘of the closed-loop perturbed system i aided by the small Eun rd gives conditions under which & ‘System of inerconnected components i stable. Inthe algebra of| stable transfer functions (FH), the small gain theorem can be ‘wed to supply necessary and’ suliclent conditions for robust stability under stable (47) perturbations. On the other han, posing the stability robustess problem in the algebra of Zsa be operators, the small gain theorem leads to. necessary and sulfclent conditions for robust suabilty oaly i the cls of Perron enlarged fo cade tne ating. uncer BL In most practical sivations, atleast two types of uncertainties are present. First, unmodeled dynamics which represent high frequency uncertainties, and second, parametric uncertainty rep. resenting lack of precise knowledge ofthe actual sytem parame aw et re ey 0 rt ee a a La ee Pe eaerarg ieee cet a hy ABne Ss aes mame ters, Recent developments inthe analysis of systems with para- metric uncertainty have been inspired by Khatitonov's theorem 14]. Nonconservative analysis techniques for the stability of systems with uncertainties of mixed type are very important in ‘order to quantify the amounts of perturbation that can be toler” ated bythe closed-loop system. In this paper, we give necessary land suficient conditions for robust stably when the system is subjected fo unstructured uncertainty, modeled as norm bounded ‘H,, perturbation, and structared Uncertainty, modeled as param ‘ter variation in the coefficients ofthe plant. Ths formulation has the advantage of handing both types of pertration in & ‘mile fashion and thus, overcomes the drawback inherent inthe fobusiness theorems that include only one type of perturbation, Not thatthe robust stability problem for contol systems subject to perturbations of mixed-type is a problem of current interest in {he control literature (e.g. see [S]-I8)). The important differ: fence between the result of these references ant the present paper isthe use we have made ere of certain extremal plans of extremal segments inthe plant parameter space. ‘We start by considering an’ open-loop linear time-invariant dynamic system with transfer function coeficints varying. it prescribed ranges. For such families of systems, we obtain precise and nonconservative bounds on the level (Hao) of Unstructured perturbations that can be tolerated, WE also trest the converse problem: for a prescribed level of unsrucured Percurbations, determine the largest stuctred /y-stability mar Bin in parameter space ‘We next consider & feedback contol system containing. an interval plant which is also subject to unstactred unmodeled ‘dynamics, After connecting & controller inthe feedback loop. {he uncertain parameters enter linearly in the coffiiems of the characteristic polynomial and a polytopic family of characteristic polynomials is obtained, For this case, we give a neceseary and Sufficient condition for robust stability of the closed-loop stem, This condition involves checking stability and calclating. norms along a fixed number (independent ofthe order Of the Plano of exteme segments which we call Kharitonov segments And which were fist intedced in (1. ts worth noting at this point thatthe results ofthis paper are intimately related to other recent results on the passivity of interval systems [9], (10) Im the sty of stability robustness under sector-bounded nonlinear perturbations, conditions tat involve strict positive realness of certain plants arse. The genet= alization of these results to cover both parametric and sector. bounded perturbations led to the study of passivity Of interval stems. The results of this paper ean be viewed as a pallet development to tat sty where the aforementioned Kharitonov Segments also play a crucial role ‘Our result) are based “on methods originating from Khartonov's theorem [4] and its extension, the Box theorem [1], and we slat by defining the appropiate nation. 1, Norariox AN Prsunavany Restrs We first review some basic facts about Kharitonov polynomi- als and present some notation that we vel we throughout this Paper. Let (5) bea real polynomial. We will denote by 3°™%(3) (0018-9286 90/1000-1100801.00 © 1990 IEEE the even part of &(5), and by 8°%(9) its ead part, tht is (5) = (y+ has? tis Fs) (54s + ish 4 -) . a) We aso write B°(u) = B"( jo) = By . (yx SH) Bw) oa Now, conser a fail ¥ of eal intra polyromies Hs) = igh dee 4b sth Bt were Boe [xe ols BEL Here bue [Kas Pal Definition 2.1: The four Kharitonov polynamials sss sated with are given by the following KY(s) = xo tet ras +98 best barat te + KB ls) 0}, and HAC.) wil represent te spice of functions (6) thit are bounded and analytic in Cs ‘With the standard, norm We where Fiji) isthe boundary function associated with f(s), [Before we delve ino the main results, we present prelim nary lemma that willbe used inthe following sections. This fist Fesul isan obvious characterization of proper rational functions G3) which are in HL(C.) and which saaly [ele <1 ‘Lemma 2.1: Let”a(3) = m(3)/d(s) be proper (eal oF complex) rational function in H(C..), with deg (d(3))~ @, then fella <1 ifand only if ‘AU inl < [dy BA: dls) + e?n(s is Hurwitz forall in (0,2) Proof of Lemma 2.1: Condition Al is obviously neces sary eeause when «goes to ini. the ratio | n(Js) do) | fens to the limit |ng/dg|. The necessity of condition BI is ‘stublshed in the following Wa). Proceeding by contradiction, ‘sume that condition BI is vielsted. Consider the family of polynomials sup | Fda) | = {dls) + a(s):de [0.1]} ‘This family has « constant degree (see Al) and contains one stable element, namely (3). Thus, i condition BI is violated, the continuity of the roots of a polynomial With respect 10 iS oefcients implies that here must exist *€(0, 1] ah wig R for which { jng) + ¥n{ J, 1102 Ik follows that n(Ju,)/d{ ji) as magnitude equa o 1/2 = 1, and this would contradict the fact that g(s) has infty norm less than unity. Therefore, 9 contains only Hurwitz polynom als and, in particular, d{s) + (5) is Hurwitz Using this Tact, ‘we now conser the Family 9 = {d(s) + e%n(s):#€[0,22)} ‘Agtin, this family has 2 constant degree (3 Al) and contains ‘one stable element, namely ds) + m(s). Ths, if condition BI Js violated, de same continuity argument implies that there must exist 0° [ 0,23) and we. for which iw.) + eM nl) Ie follows that ne) ja,) has magnitude equal to nity snd this again contradic the fact that (3) as Init orm less than unity. ‘Conversely, assuming that conditions Al and BI are tre, et us suppose by contradiction that lle. > 1. Since | g(Ju)| is ‘contingous fonction of wo and since is limit 38.0 goes t0 infinity, i | mg/d | <1, then there must exist at Test One &, in R for which But this implies hat nj) and dtu.) der only by 9 conpes msbie of odie'l and, everest is pale to finds [02s oa Ju) + een Jag) = 0 and this obviously contradicts condition BI ° "In fac, his fesult has an immediate generalization to the multivariable ease and even though we vill not use it in the following, we belove that his generalization is of interest in its ‘own right. This extension to the multivariable case is therefore Let H2'"°(C, ) be the space of matrn-valed fonctions Fs) that are Bounded’ and analytic in C, withthe norm LF. = 587 Gua FLs)) Fora conta complex main, | wl den he following Herons Al ee WAvIy = Oyun A). Lemma 2.2: Let Os) be 8 poe rol wae tion init a HES" Aan hoa a of gael Pe maak Os) NID" Obe ase cope det fos of ta ove rng ply ati, wih) ci econ FO ti an 0 i A2: Gop] < 15 de D0) + Uff] No i Hawt for wiry mati cau mor “re pot of ema 2.2 can be fun in he Ape I, Gexexauizariow oF rt Sia Gain Taronsa Fo "Rowust StaBiir¥ UNDER PERTURHATIONS ‘oF Mxen Tyee A. Computation of the Unstructured Margin ‘Consider the following version of the small gin theorem In this problem stable linear time-invariant system is perturbed a feedback by a stable transfer function A P with bounded He, norm, as ilusrated by Fig. 1. "The question which is usually addressed inthe robust stability Lt Bgl | ——| » a Tterature [2], [12] is that of stability ofthe closed-loop system for all sable’ P contained in an H, ball of prescribed radius. Referring to Fig. 1, we state the folowing wellknown rest el Theorem 3.1: 1 g(s) isa stable transfer function, then the closed-loop rystem femains sable for all perturbations AP sausfying |4P |. +dystidye[d7 — eyny.d7 + ems) contains an unstable polynomial. This upper bound is easily Caleulated, using, for example, the method proposed in [14] ‘which i staightorward. One way 1 CompULE qa is thed 12 Follow the following algorithm, 1) Set LBOUND = 0, UBOUND = e, 2) Let ¢ = UBOUND + LBOUND/2. 3) Update 9, (0). 4) Ifthe 16 systems in %,(«) have H, norm <8 then set LBOUND = «. etherwise set UBOUND = ') If [UBOUND ~ LBOUND| is small enough then EXIT, ‘otherwise GOTO 2. “This procedure is jus bisection algorithm which requires, a. each step, to check the Hi, norm ofthe 16 curent Kharitonov ‘stems, We llusrate th computation of the 7, structured ‘margin in the following example. ‘Example: Let the nominal system be given by Ins Trate ‘The Hee norm of (5) is equal to 1. Let ws fix the bound on the unstructured margin to be equal 1/2."To simplify the notation, tve assume thatthe perturbed system is ofthe form L+a~(14d)s Tre+ Grae a) = eal) where Ja] se, |b) se lel se Idi se ‘We sek the largest such that Heaocdlan 2+ forall (a, b,c, d) satisfying (3.3). For this simple example, én, can be computed analytically and {is readily found to be equal to 1/3. The extremal systems are 63) 42 4 373 oe ‘ me i ptgets gt gets IV, Geneeatizarioy of rite SMALL GAIN THEOREM FOR ‘Conor Srarists We now present more dettiled results forthe casein which a fixed contole sinluded in our intial system. Mathematically, ce this cae fs also important when the parameters of interest are not the original transfer function coetcints bat ener linearly ot Mine linear imo the system. Let be a family of siritly proper single-input single-out- put imerval plans, withthe addtional asumpéion that 9 (see (2.4) isa family of fixed degree polynomials. Thus, referring to (2.4) we have that 8,2 4? O08 Hy $8, <0. (a) Assume also that we have found «stabilizing controller e(s) = ind3)/d,(3) for the entite family. We therefore have a family of Stable closed-loop systems and We consider unstructured additive Perturbations as shown in Fig. 3. Here, aso, we want (0 Setermine the amount of unstructured perturbations that ean be tolerated by this family of interval plans. In order to do 80, we have to find the maximum of the H,norm of the closed: ioop safe onton eS) g(3)(5)" "over all elements 263) Before stating our result we need 1 introduce some more ‘Let be the set of four Kharitonov polynomials associated with Also, Tet or, be the subset of .F consising of the four following line segments: [A (5). 3 (9)}: [4 (8), 3 (5)]5 [x5 (9). 4 (99) [K269). KA (9)] We call these the four Khartonoy segments associat with By tine segment (K',(8). K3,(s)l we mean of course all ‘convex combinations of the form (1 = KA (3) +3 (5) ote that 7, ds nt cpt he convex combination of every posible pair of two elements in 2, and, therefore, ‘numbering of te four Kharitonov polynomials which we adopt in our notation is important (see Definition 2.1). We can similarly define the subsets 1, and 27, of 9. Finally let, be the subst of 9 defined by ats) : doom [ey € HM) 6% aad ls) o ox (n(s) €74, and d(s) €2*,) (One possible clement of is, for example Ki (3) #09) = TVR) +E) where 2 is some real number in [0,1]. Notice that, cone {hins at most 32 one-parameter families of rational fctions ‘which we cal absively "segments." “The folowing theorem shows that in order to compute the exact level of unstructured perturbations that canbe tolerated by the whole closed-loop family. iti both necessary and sufciont ‘0 compute the maximum H_ norm along those segments, Theorem 4.1: Given an ssterval family of strictly proper plans and a stabilizing conllere(3) for, the eased loop in Fig. 3 remains stable for all stable perurbations AP such that [SPI 0, max Hle(s)(1 + e(s}e(s)) ‘be 0 be piven Tre cus fe tine ‘ata contol Gt stables 9 and hat sues sep Hef s)(1 + a(sde(8))" "I and only if such controler exists for 4, ‘We belive that the previous theorem sets the stage for further investigation into the syathesis problem by precisely determining what the role of the controller is in the robust Stability of & Family of interval systems, YV. Concubine Resares In this paper, we have established precisely the amount of ‘unstructured peturbations that can be tolerated by a farily of ‘Single-inpat single-output control ystems rubject to prescribed parameter variations, when these parameters are transfer func- tion coeficens or depend afin linearly on those. Te rests ‘useful as an unstructured stability margin calculation. We showed how the structured st prescribed level of unstrictured perturbations can be found ‘These results qualify as bona fide stability margins for the system bocause they are nonconservative, "The results presented here are of direct use instability analy. sis. Is our hope tha they may also aid inthe development of « Symthesis procedure: for srbcured perturbation, possibly by {apping into the synthesis theory svallable for unstrctured per turbations [16]. These issues should be the subject of father research Arrexorx Prot or Lena 2.2 Sufficiency: By contraction if |= 1 then A2 implies that there exists ge R such that |G" Jo GUo)f = 1 But itis well known iat [G(Ja,)I = [G4 Ja,)GUJu) and therefore, for that same’ wy, we have 1OCJoy)] = 1. Ths, there existe ve? such that 1G(ie,)oll = Lol “This implies tat [[e]mticero- "i nd therefore there exis unitary matrix U sch that u[d]vie.)O-desd_ ‘As a result (Dt i) eee "Necessity: Iti clear that condition A2 is necessary and therefore ve assume that it hols inthe follwing. In order to ‘establish BD, the fist step is to prove that o (9 fs) is Hurwitz forall Xe 0,1]. Consider the family of polynomials {ris =a4(009 #94] oc9} 10.1 1 tow ening A ines aly nr te cin wre D(s) = DycH(s) + Des). N(s) Nott (3) + Ni) faa) where Hs) = diag (s* Pls ky being the colums degree of the ih column of D(s), and '‘Dy(s), Nts) comin the lowerdegree terms (or further explaniion of the previous andard notation see, for example, Kalla [11) THis easy to see that Gle=) = Nj Dj Using (A.1) we also eet Bore) = (2+ Aff] ou) nto) + 0,69) #92] [Now the fact that (sis column reduced implies that det(Dy,) sm, T pov im ch sw ht de (u-r9[J]se) #0, brat defeat Spy coin rae 81 ax[oceaé]ne) =o 7 ins inte da(r nfo) = 407 [f]orm)) 0 sr wo nt str VP ah a [gletr- 3 and therefore [slow 0 that [|G*(20)G(2=)] = GC)? = 1, which isa contradic: ‘With this claim, we know that the family of polynomials (5) contains one stable element, namely Py), and has xed segree. Using now the contiuity of the roe of a polynomial with respect toi coefiints, we see that this family contains ‘an unstable polynomial if and only ii also contains a polyno ‘ial witha Foot‘on the imaginary axis, However, if PJs) (O for some A, &( 0,1 and some w,€ R, then = 16s = 5s = lls det (D(je,) + d5[ 4] MUea)} = 0 and the same argument as the aforementioned leads 10 the ‘contain that Or 4e,)G( iw) = G(s)? = 1 ‘To complet the proof, we need only apply the same reason- ing asin the first step of the prof wo the family of polynomials {rata = (0 u[f] nto) -7 wins main om) i canbe proved fist that this family has constant degree and the first part of the prof shows that it contains ne stable polynomial (corresponding to U'~ 1). Now note thatthe set of fnitary matrios i pathwise connected. This id tothe fat that any unitary matrix U can be expressed as U~ ©, where Fis some Hermiian matrix and therefore itis the mag? of 2 convex se under a continuous mapping. This implies, 1m tun, ‘hat the family of polynomials Py(s) is pathwise connected. and thus allows us tose the continuity propery of Ue roots of 2 polynomial with respect to its coeficients. 3 Revenences 0 ta MD eh ae Erna cant SN wa Eee erry Sgt ren Av 01 SOT a MEE scarce 11 PE cremate ity tng it Vpeores cesatn ancora 1107 Bere tar cs ee onl ystems with nce framers and ve ars ene at ae ea es Be ae EE eae, a oe en feel datatm kG pe ta Ee ee ih san cs aah eet Rapes 1 eave toa fc ped te Wor Reba 1 GREE SSM, meen nr Toma" Conir, 1989, to be polished. i 112) C°K, Dor an Mt. Yihasugt, Fda Sens: Ipa-Ou aay Wee Rees, Oe gente of sabi een” Ad Nauk Recach SSR Ser Fes Mats Wal Ty pp. 33-3. i 1) HS Rona aL Re Sy me tis) 26 ities, VY ng Li Roa aan of a ate nee Seat Sands emai en ee Ma ve anata njormanan Snes New Vk Spins Nig Mey Chapt a on Mancile,Fae Natori Sepineure ces” Telconeancaons TENS). Para France, i 198. and feed tie MS. epee eee etsoetng 0) Texas A&M Univer, College Stason in 587. Pro 158610 1987 he waa Reserh Assn ln be Depart of Bc Egicting. Teas AM Universi. He caren completing the PAD. cere ceric enpocing at ane tery: where he aso an Assn? Veco ln the Deparment of Heel Engineering Ms caren each cress tei ear yen try ea onl hea ad ial oes Nr Gp reeds Dips Gra Rech Ava on Monammed Dabeh (S'45-M°87) was brn in onde i196 He seca te BS dec in flee earings Tease A&M Unt Sip. Calege Suton in 1083 andthe MLA an PRD depts apie and computa mate te from Practon Universi Prison Nl {lone and 198 rape si 7 hea th be Den Hy geeral are of irs sea ol ‘her hs cif afexs a erst ioe abst coo abate sot rind puramete yene Shankar P._mhauacharyya (S'67-M7 Suf86- 18) wan bore a Rangoon, Bema on Je 2, 1986. He rte the Bacslnof Tok notogy Honours) dee i eel eines fromthe lien Ii of Toco, Bont. st Grate Program i aging (COPE) at the Feder Unversity. Rio de Tae, Br, “hee he esti th Grate Props Astamatie Corral and was ‘Gatun of te Beprnent of Esta! Eapnering fo 1H 98, 1108 acing 1974 and 1975 be was with he Dyas and Control Division of| {be Sys Dymus Labor at NASA's Maral Spe ght Coser 1 a Navorl Academy of Secs Reseach Fellow. In 980 he fed ‘Tenas AM Univer. Coleg Suton, where b's pesealy Profesor of ‘hci Bagmering, erected Flight Laing Award 198) ‘tas ben + UNDI coma oh Goverment of nd in 950 He as eld visting ectrngnsignments we Univer of Flrece, aly, te Fotis Unverdade Catz, Rn dene, Bra te Ian tite Inn TRANSACTIONS ON AUTOMATIC CONTROL, VOL 8, NO. 10, OCTOBER 190 Mexico, Meio, sl te Ini insite of See, Baga, na He 1s conte 1 seer rt ona otro ery. Hs peseat recarch sais with te sig frost ono ems His meres face laying Incanto sre rr Bhuachanga tered arse Aust Ete fr the IBEE Teac ‘ne om AuTowntc Conrno. in 195 an 1986 ad a eer Of ‘Boar of Govern ofthe IEEE Cana Syste Society 1986

You might also like