) Pd a where P, = Path gain of the A forward path; 4 = 1 ~(sum of loop gains of all individual lg + (sum of gain products of all possible combinations of two non-touching loops) - (sumofs* Products ofall possible combinations of three non-touching loops) +.... ; 4, = the valuedt J that part of the graph not touching the k'* forward path. Comparing eqn. (12.30) with eqn. (12.81) we observe that a signal flow gra? eqn. (12.29) may consist of In chapter 2, i by Mason’s gain formu and signal flow graphs are rds (128 () three feedback loops (touching each other) with gains — fs, — ays? anday*i G2) four forward paths which touch the loops and have agains by, by/, bys? andb : A signal flow graph configuration which satisfies the above requirements is si" “Fig. 12.11, From this figure, we have i Yar, + byw } | Scanned with CamScanner“ \pAIABLE ANALYSIS AND DESIGN TAO + BG 95 + bye +9 + (b= a,b.)u on(12.89) OX, + q+ (by - anh 9X + (by ~ agbgu as [e|o2t 2 Vm] for -ayby « 2 |=)~a2 0 affsa|s]ay —ayby| lal (12,340) af as 0 alles] [oy ~ aay x Y= 0 Op] dye (12.346) Xs, ‘The initial conditions x (0), x,(0),x,(0) can be obtained in terms of given initial conditions (405 50), u(O), (0), (0) from eqn. (12.33). The result is given below HO) [1 0 oso) po, ‘2 (0) |= =by 0 0 Tfuoy Es Fer 1 OH 5@)|+)-2, -b 0 |ito) (12.38) x00) lar a aflx@) [ost a2 9 llato The results given in eqn. (12.98) and (12.4) can easly be generalized for an n®* order aieential equation . IP = OY tt ay Hay = DUM + et Dy gti + Dy tt (12.36) i Fig. 12.11 An alternate phase variable formation for the transfer function of eqn, (12.31) may be ‘tained as follows: . ‘The transfer function G(s) of Fig. 12.12 may be divided into two parts as shown in Ye 12.18. Thus ¥(9) X19) YO) 2.37) 60)= FOS (12.87) Scanned with CamScannerCONT ere 1 X\9)_ here UG) a +ays +428 + Ay and LO 2 5ph e byt + by + by X,0) 1 x Pad oad sas 40,54 a, Stas+as+a, s+! + a7 a [yeal =| Fig. 12.13. An alternate repro seman, Fig. 12.12. A third-order system, of sytem of Fig. se ‘The transfer function of. eqn. (12.37) is without zeros and therefore its Tepresentation may be obtained from eqn. (12.28) Hm] fo 1 0 Ya] fo l=] 0 0 1 |/x2|+]0|u » i3| [+03 -a, -a)/x)} [1 Rs, From the transfer function of ‘eqn. (12.38), we get 9 = Dok, +B, % + ye, + Byt = fe gt, — apt, — pty + W) + Dyty + pty + bya, = y~ ashy, + ,—agbyiy +0, ~ abo) + bye Pha ra output equation (12.395) can be written in vector-matrix form as ‘te [x] y=[lby—ayby) (8 agi |r {24 Ls 7 ‘ow graph corresponding to eqns. (12.40a and c) is drawn in Fig iy Scanned with CamScannerjpiASL€ ANALYSIS AND DESIGN vA (6 oe from the —— eee (12.83), (12.94) and (12.40) (and their generalized forma) we co that the phase eal le formulation enn be obtained by inspection from the transfer ion and vice-versa. The reader will also note the similarity of phase variable representation | yatrated PY Figs. 12.10 and 12.11 with the direct programming method of simulating 181 for functions. 1a" g disadvantage of phase variable formation is that the phase variables, in general, are physical variables of the system and therefore are not available for measurement. and ad purposes. We have seen that if G(s) has no zeros, the phase variables are given by its a gt and derivatives of output (eqns. (12.27)), Unfortunately, it becomes difficult to take wf qor higher derivatives of output. If G(s) have zeros, then the phase variables bear little Xf gnblance to real physical quantities in the system as is seen from eqn. (12.33). Thus, though se yariables are simple to realize mathematically, they are not a practical set a state ‘ables from measurement and control point of view also, phase variables do not offer any ageamtase As we shall see shortly, canonical variables are most suitable from analysis point view. In spite of these disadvantages, phase variables provide a powerful method of state arable formulation. A link between the transfer function design approach and time-domain gsim approach is established through phase variables. We shall discuss this link later in sis chapter. gutespace Representation Using Canonical Variables ‘ichle or normal-form representation of a system, the matrix, A turns out to be form of state model plays an important role in control theory. Let us general transfer function (12.24), reproduced below: Y(s) bos” + bys") + Ti) “9 = ass Cuaprer 12 +By38+ by $a,48+a, Assume that the denominator is known in factored form and that the poles of the transfer fiction, located att Ay, Ayy un» Ay, ate all distinct. The transfer function can then be expanded isto partial fractions as (see Appendix I) Ys) ¢ Tig = 7) =b0+ z ‘here ¢; are the residue of the poles at s = 4;. The block diagram model for the transfer function is shown in Fig. 12.15. Defining the ‘utput of each integrator to be a state variable, we can write the state equations as 3-4, ay Aa + uy Dy seresey Me (12.41) The output y(¢) is given by Y= Cy, + py He + Cyt, + Byte +-(12.41b) Tig ettations (12.41) describe the eanonical state model of the transfer fuction of eqn. (12.40). * state mode! can be expressed in the vector-matrix form as Scanned with CamScannerSM 815446 6 0 O14 g ‘onical side model described above, the system mtr a diagonal m ¢ poles of Ts) as its diagonal elements. It is also observed clements of column vector # of the canonical state model are all unity and the elemensd'a pa erter © are the residues of the system poles. An alternate canonical state model sz below. ea ot 2) ly ae nd [en A y= Li. U2) soy ee * The unique decoupled nature of canonical or normal-form representation is &°= from eqns. (12.48). By decoupled we refer to the fact that in meant form, the n firs differential equations are completely independent of each other This decoupling feats | we shall see later in this chapter, greatly ‘helps in the analysis of the system. Scanned with CamScannersE ANALYSIS AND DESIGN ants ye disadvantage of the canoniea fy * variables, are not real physical xample: rm is equally important, ‘Tho eanonicn! variablen, has variables of the aystom., ‘Thin in illustrated through owing @ : i? cansder the differential equation FOV + 1 + Oy =i + Bu 4 170+ Bu jssuming 2070 initial condition, we get Y(s)_ 5) = +88" 417848 (244) Us) s+ 6s" +118 +6 = 5948s? +178 +8 (s+ Xs+2K643) (12.45) ‘The block diagram simulation for T(s) of ‘eqn. (12.45) is shown in Fig. 12.16. From .g244), we can write the following: # sY(s) 1 4.3 (12.46a) se ae (12.460) cic Mameharrs ree bares 2 S°Y(8) 9 sya s (12.466) Tie) 4-8 a Gan ees and (42.46) may be expressed as a sel 21 | sU(s) - Us) gl UG) 8 9\|°t? s°¥(s)— s°U(s) - 2U(s) + 6U(s) 44 8 ¥(s)— 8°U(6) = 218) + U8) v@) 3 X19) Ue) 2 xo 3 9|| 7@ X4(s) ve) 1020 Ifa Ou 1-4 f:) (12.47) —2i+6u) [-1 8 Oh xs, i ical vari the ical variables are not physical variables of t 7 tute for measurement and control. (Note that the initial “ terms of specific initial conditions (0), 5(0), From eqns. (12.47) we fine 7 xem and therefore are not avail | "tions x, (0), x4(0), x9(0), may be obtained > 50, w(0), (a), (0) from eans. (12-47)- eS Scanned with CamScanneri | Fig, 12.16. Block diagram for T(s) of ean. (12.46a), ignificant developments in the applicatj Example 12,3: There have been significant de s ction technology to various elds, One such appleation is he Itligent Whey Th. is hhas to move from place to place avoiding obstacles. The simplified bloc lagamot and the wheelchair is depicted in Fig. 12.17, Write down the canonical state vargy "i ‘i le hy the complete system, Also draw its block diagram (in state variable form), ins Us) [2346547 | u Yo, (s+1) (s+ 1)(s+2) Fig. 12.17 Solution. The transfer function of the complete system is Y(s)__ 2s? +6s+7 Ue) G+ GHD Notice that the system has repented roots. Decomposing the above transfer funcins, the method of partial fractions yields Ys) |= $_,_3 Ui) (+) (+O? +R) a We shall define the state variables term by term. Because of repeated root ats=-1é state variable of the second term will feed into the first term, X,(s) Grp 79) > Heh ty > eon tH fi U(s) GRD ua hie ipentgeu a UE) P G42) ~%0) > unigt 2x > ip=-2eyeu : Output =-m 484 + ar, i ‘The state variable model in matrix form is written below from the above ¢4P* Scanned with CamScannernt NAN AND DEAN at Sorin Nock \ VA | Olfay) po | we | hm a) wl) | ago tate vatiable block dlaygrann ty drawn ty Biye 118 j Of} p@ds] peh® t t *K Li} a | | 4 | 4 | | | [2 votiew the dotted block in matrix A (oun. (0)). Thorefore, x and xy, re nol so roponted root abs = = 1. "Phin block (dotted) is novwn a fonda decoupled form as far ns possible, to Modal Fig. 12.40 Observatio: indecouplod (avin choca Ak his is tho sinnptoat Auuatton of Transtor Funetton from Sta Tg ansivoved the quextion of obtaining tho ato model for a given transfer function Paar Necor igetetainye), wo next considor tho problom of dotormining the transfer taton from given atato model of single-inpulsinglo output aystons, ‘he Tink bobween transfor funtion and phavo variable formulation hag already been daonshgted For the continuous caso tho transfor function trom wate model using phase ats me ye nbuatned by inspiration from the yonorul rovulls (12.28), (12.54) or (12.40). Be satin cae rod a to draw’ algal flow graph for the given phas variublo model and then isn ho transfor Punto vane Mason's gain Formuly Diverter exw wil follow Inter go 600), Fora gonoral stato model Ba) 12,486) xe Ax + Bu yx Ox4d ‘betransfor funetion may be obtained 08 follows: ‘aking tho Laplneo transform of cams. (22.50), wo have aX(o) (0) = Ax) + BU) (a) = OX(s) + dU(s) Scanned with CamScannerCONTROL Syste, Solving for Ys), wo obtain Y(x) = Clal = conditions Ay xt) + Cll = Ay! BU(s) + dU(s) wwe got the aystem transfer function ay Assuming zero ini Ns) out AB ad Ns) = Ty _ Cnt AB a ee to he made here is that while the sty, ‘Ny nionunique, the transfer function of the system is wits ere faneig (12.51) must work out to be the same irrespective 0 Mode} na as mn ae ‘of eqn, (12.49), v a the characteristic equation } Ist-Al = An important observation (hat ne IME SYSTEMS | STATE VARIABLES AND, LINEAR DISCRETE: State va atble methods for the analysis a ‘ carlier sections of this chapter can be extended for the analys systems. . In Chapter 11, we discussed analysis and design of discrete-time systems using yng function, There ved that while dealing with diserete time systems, we often enays” two different situations. The first situation involves systems that are completely diseee respeet (o Lime in the sense thal they receive and send out discrete signals only, Itis hrs find examples of physieal systems of this nature, other than digital computers. Most afi physical systems are of continuous-time type. In socio-economic and some other nonphjss systems, «1 natural diserete unit of time exists and sueh systems ean effectively be mole using difference equations. obse ‘The second sit is that the components of the system are continuous time dens: but the ils at certain points of the system are discrete with respect to time because afte sample and hold operations. Both the cases will be studied in this section. ‘The general form of state model for a multivariable discrete-time system is x(k + YT) = fix(kT), why] soll YARD) = gl(xCkT), wk) nll 2388 where x(}7) state vector, u(T) = input vector and y(k7) = output vector. From these equations we sce that given the initial state x(0) and values of inputs 0 WT), U2T), .., wT); we can uniquely deduce the evolution of state x(7), x21) x{(é + DT) and the evolution of the output y(0), y(7), ..., y(T). Also note that eqn. (125 a set of first-order difference equations which represents dynamics of discrete-lime and eqn.(12,605) is an algebraic equation. For an nth-order linear time-invariant system, eqns, (12.50) reduce to the fall" | x(k + 1) = Ax(h) + Buh) ; x(kT) xt) ae y(h) = Cx(k) + Dut) (28 form: Scanned with CamScannerans ANALYSIS AND DESIGN ie a a ernst Furtion from Discrete-Time State Model ve She etrans oth sides of eqn. (12.492) (assuming sealar input), A 2X(z) ~ zx(0) = AX(z) + BUG) sing for Xie), we get X(2) = (@I~ A)" zx(0) + (eI - AY BUG) aking the z-transform of eqn. (12.496), we get (for scalar output, D = @) Ye) = CX(z) + dUte) golving for ¥(2), we obtain ¥lz) = Cll - A) zx(0) + Cl - Ay! BUG) + dU@) assuming zero initial conditions, we get the system transfer function as ol (12.52) Tie) ia Cat- ayes a= Caticl Ae od (12.53) getting the denominator equal to zero, we have the characteristic equation l2I-Al=0 (12.54) ‘The roots of the characteristic equation are the eigenvalues of matrix A. é Pee 2 (EB dso ZATION i) Te dbserved in corlier sections that the state model of a system is not unique. Some of the vite models presented employed physical variables, phase variables and canonical variables. Fam application point of view, the physical variables for system representation are most vrfles the resulting state variables are real physical variables which can be easily measured vase for control purposes. However, the corresponding state model in this ease generally convenient for investigation of system properties and evaluation of time responS®, The vmonical state model wherein A is in diagonal form is most suitable for this purpose It is thecfore, useful to study techniques for transforming a general state model into = canonical tne These techniques are often referred to as diagonalization techniques. Consider an nth-order-multi-input-multi-output state model x =Ax+Bu (12.554) y=Cx+Du (12.550) “Assume that the matrix A in this model is nondiagonal. Let us define a new state vector such that x=Mv where M is a (n x n) rionsingular constant matrix. Under this transformation, the original State model modifies to 2M AMv + M7 Bu «.(12.562) y=CMv+Du (12.566) Scanned with CamScannerCONTROL SYsTEg «| that MeEAM isa dingonalized mays, wel, Under tis condition, the may Hequations (12.66) may be writen Mig! such Ifthe matrix M can be solected : nonical state model given by (12.56) is a ea Z the diagonatizing matrix or the modal “t = M7 AM= nal matrix; B= ae where A= Mot AM = 8 diago » diagonalizing matrix is faci tated by the use of The determination of the digi of eigenvalues and eigenvectors is bregy My, With this purpose in mind, the concept 0} vt below. Eigenvalues and Eigenvectors Let us now give attention to the equation Ax=y and view it as transformation of n x 1 veetor x to x it vector y a a ae Operator 4 question which we will like to answer here is: whether oe oe vc x sn that may operator A transforms it to a vector Ax (A is a constant) /.e., to a vector having the ,, direction in state space as the vector x. Such a veetor x is a solution of the equation Ax = ix olla Now our question is about the existence of solution of ean. (12.58), which may be wie i as (i= Ax =0 llD55 ‘The set of homogeneous eqns. (12.59) have a nontrivial solution if and only if Cae (026 ‘This equation may be expressed in expanded form as QA) =H aR + ag? +. 40,20 (1284) ‘The values of 4 for which eqn. (12.60a) is satisfied are called eigenvalues of matrix and eqn. (12.605) is ealled the characteristic equation corresponding to matrix A. From eqn, (12.51) it is observed that the poles of the transfer function 7(s) are giveals qs) = | sI- A I= 0. This in fact is the same equation as eqn, (12.60a) which determines tt eigenvalues of A. It is therefore concluded that the eigenvalues of the state model and tt? poles of the system transfer function are the same. All the conclusions established in Cha 6 on system stability based upon the location of transfer function poles (i.e., roots oft characteristic equation) are therefore valid for eigenvalues of state model, Thus a state mul! is stable if all its eigenvalues have negative real parts. (In Chapter 13, we shall study stilt’ in details). Now for 4= 4, satisfying eqn. (12.57a), we have from eqn. (12.59) (I-A =0 Aaa Let x= m;be the solution of this equation, The solution m, is called the eigenvetor* associated with eigenvalue 4,. _ Scanned with CamScannerjgABLE ANALYSIS AND DESIGN Y aid curio hation of eqn, (y qn. (12.6 | then there ane Gy ay aye pends on the rank o NP) independ matrix (AI ~ A), If is the lepende 2 rane eigenvalues of matric Pendent solutions ( A yenvectors). Mat hence We awe only one in foe iw distinet, then the rank r of matrix (AI - A) is a 4, This eigenveetor, may be ouieed igenvector corresponding to any particular d by taking cofactors of matrix (41 — A) along, orn sy, a0 the coreors oF matrix (AL — Ay spire Ch LA). +m, be the eigenveet i i hen we have areas where Cuapter 12 0 0 a, = MAM The matrix constructed by placing the eigenvectors (columns) together is therefore is a diagonalizing or modal matrix M of A, i-e., M=I[m, !m,!.../m,] Note that A and A have the same characteristic equation; therefore the eigenvalues are invriant under this transformation. When A is expressed in the form given below, 0 1 0 0 0 0 1 0 abe fle : (12.62) 0 0 0 1 q Gy, — Any ~ Anz ~ ay tten the modal matrix ean be shown to be a special matrix (called the Vander Monde matrix) 4 = Be dyn ody ve|# dn -(12.63) Lge a a Scanned with CamScannerCONTROL S¥Steyg Example 12.4; Consider a matrix A given below: o 1 0 Ae| 3 0: 2 -12 -7 -6 Corresponding to this matrix, the characteristic equation is 7 6 jaai[3 aA -2 120-7 A+6 = (+ (A+2)(a+3)=0 ‘Therefore the eigenvalues of matrix A are Ay =-1, 4y=-2, 42-3 1 is obtained from co-factors of thy ‘The eigenvector m, associated with 4, = -1-1 0 (al-A)=|-3 -1 -2 wT 6. ‘The result is (eigenvector has unique direction) Mati ~~) The eigenvector m, could also be obtained from a solution of the homogeneous ua, corresponding to the matrix (JE - Alx = 0 ie, = 8x, ~~ Bry = 0 12e, + Ty + 5x; Choosing x, = 1, we get x, =~ 1, x5 =~ 1, which is the same result as obtained byt method of co-factors. Solution of homogeneous equations is computationally more ef! than obtaining co-factors of row when the dimension of A is large. Similarly eigenvectors m, and m, associated with 4, =—2 and A, = —3 respective et ~Linls The modal matrix M obtained by placing the eigenvectors (columns) together is t* by Are M=[m,!m,im,)=|-1 -4 -3 lt 1 3 It can now be verified that, oi on 2 A=(MTAMJ=| 0 -2 0 0 0 -3 Scanned with CamScanner,AB6LE ANALYSIS AND DESIGN wh ff a diagonal matrix with ej Po “rectly without the need va c é ienvalues of. i " to compute “ ea diagonal elements. In fact 4 could be alized Eigenvectors a gf sussion on eigenvectors so far we have am ct. Let us now relax this Testriction, Consider nee do eigenvalues of matrix A are ~ mé ae Ast 0 3 1 ae at a5 i walues oft : ; | _ ee Are 4 = 1, Ay = 3, 2 =8. The eigenvector associated with may be obtained from the row co. factors of the matrix | -3 -1 9 G@lI-ay=|-1 1 3 al 3 Cul fo ™=!Cip/=|0 Crs) 0. Gofactors along first row give a null solution, Letus take co-fctors along the second row. Cuapten 12 To obtain ei ‘vectors associated with the repeated eigenvalue at 2 = 3, we construct the matrix eee GQI-A=| -1a - -1 1 4,-3 For 4y= 8, the rank r of 3 x 3 matrix (2,1 A) is 2. Therefore one independent eigenvector ssiated with 2 = 3 can be obtained. It is given by Ci] [asl4e-3)+2] [2 Cyy|=|(4,-3)+2 |=]2 Cy) [-144, 2 m, The vector m, for the modal matrix M-=[m,!m,!m,) "ay be generated from the independent eigenvector m, as follows: a aon da, 2A-3 3 d =|-2¢. 1 =|1 m eee 1 ag =3 1. eat day |e Scanned with CamScannerCONTROL SYSTENg jatrix Mis then Riven, eigenvector. The modal ™ 0238 m-|8 2 is 21. forms A to the Jordan matris, i ‘The vector m, is generalized ‘The modal matrix M now tran: [1 0.0 wt AM=| 0 0 Jordan block «, to have a state model whose A matrix has eigenvalue 4, such that rank of n x n matrix (4, TAD is less than (n =), fe. there aes than one independent eigenvectors corresponding to 4. However, we shall assume jg. discussion that for the state models we deal with, for each 2, of multiplicity 9, there ei, model matrix M such that the Jordan matrix J = Mt AM will have aq x q Jordan i corresponding to eigenvalue 2. It is possible in some rare case Linear Transformation of State Vector (Discrets-time Case) ate modal is equally usefully in the dice, n the continuous-time case the canonical em ar ‘The transformation x(k) = Mu MI = moda! matrix transforms eqns. (12.49) to the following form vie + 1) = Av) + Butt) ; A= M+ AM; B = MB (1288) y(h) = Evlk) + Duk) ; 6 = CM wll! ‘The techniques of obtaining modal matrix have already been discussed in this sets If matrix A has repeated eigenvalues, it can be transformed to Jordan canonical form. [EEG] sovution oF state EQUATIONS 3/0" In the preceding two sections, we discussed various state models used to represent a syst In the present section, we shall develop methods for solution of the state equations from whi the system transient response can then be obtained. Let us first review the classical method of soluti: ideri order se Lei ion by considering a firs oy Be Teal Scanned with CamScannerMALE ALIN AND DESIGN priv equation bruv tho solution XCD) oo seater tat ' ay eo 12.06) {ol ww now consider tho stato equation HD 9 Ax) 3210) =x, 2.67) | enropresonts homogencous® linear fi ; py analogy with tho seatar yatom (unforeed system) with constant coefficients. u AK, We assume A solution of the for x) sa 4 | CO Ry AYE LE sa AY 1a, are veetor coofficionts, we By substituting the assumed solution into eqn, (12.67), we got Ay 4 aye + Bay ‘The comparison of veetor coef = Alay + 8,64 yf? + sn) nits of equal powers off, yield a= An, May \ Alm Cuapter 12 Inthe assumed solution, equating x(t = jy we find that Ay = Xp ‘The solution x(¢) is thus found to be All x0 = (Keane gyal Laisa Bach of the terms inside the rackets is an n xn matrix. Because of the similarity of the tality inside the brackets with a scalar exponential of eqn. (12,66), we call it a matrix ‘qonential, which may be written as 1 Laiyi eM aT HAL APE + wn qa te (12.68) ‘The solution x(t) ean now be written a5 x(t) = ex, sl 12.69) "The stale equation of a linear aystom is foment trol fe ied to the system) i is ve ano vector (je, no control forces are applied to the sys Aisa constant matrix and wie 8 rom, On the ather hand, if Ais a constant matrix and Venmeas pelea pom represents ‘a nonhomogencous linear aystom (ie., control forees are ero vector, the ‘plied to the system). Scanned with CamScannerCONTROL, S¥5Teyg s the initial state xy at = 0,16 diye, F . (12,69) it is observed thal , at on This laa in state is carried out by te mare aw a property, e is known as state transition matrix at van Let us now determine the solution of the nonhomogeneous stat ng, (0) = Ax(d) + Bu(e); x(0) = x9 In, Rewrite this equation in this form “th (0) - Ax() = Bute) d ite e~M!(Lx(t) - Ax()] = S few Multiplying both sides by e~’, we can write e~™'(Lxx(¢) — Ax(6)] ae Whee Integrating both sides with respect to ¢ between the limits 0 and ¢, we goy ; eM x(t) eSBu(t) dt 0 lo ‘ eM x(t) —x(0) = fe Bula) de Now pre-multiplying both sides by e4, we have “ Ae oe) x)= x00) + fe Bu() dr ite Rem Fordson the initial state is known at ¢ = ¢, rather than ¢ = 0, eqn. (12.71) becomes X= AM x(t) + Pe Bula dr nn Properties of the State Yrai n Matrix, 1m the above discussion, the state transition matrix has been defined as Be) = ett wherein the initial time has been taken as ¢ = 0. In general for a linear time invarient sys if the initial time is t = » State transition matrix becomes W(t ~ 1.) = eAt-t) Since the state transition matrix depends upon (¢ — {9)s to is conveniently regarded: zero. Certain useful properties of the state transition mati 1.0) = e* = 2. BE) = et = (eA § LQ py or & Kt) =O 2) 3. OE, + ty) = eM +4) _ pati oat, = Wty) BE) = a¢,) O,) rix (t) are given below: Scanned with CamScannerpfikBLE ANALYSIS AND DESIGN ef jn torms of the slate transition ya.72) ean be writen as matri : tix 4X2), the solution of the forced system given by XU) = 4 k 0) x(0) + fo 1) Bu(r) dr instead the initi i (12.724) sf inste initial state is known. at oy x(t) = OU 1 Pet) tg) + [Oe - 2) Buln) dr (12.736) ner? (t= t,) = @At~0) Mt 2) = eAt-9, ssanple 12.5: Obtain the time response ofthe fellowing syste 4) _f1 Ofer EE} grou) 6 unit step occurring at =0 and x") =(1. 0} golution. We have in this case : T Fl a-li ‘Ih ‘The state transition matrix () is given by Cuapter 12 1 1 eNa TH Ate ANP gi AN Substituting values of A and collecting terms, we get Me 1+t+ 050 +... 0 tePe., Lett 050? +. 2 (ou ge “| Pa P22, The terms @,, and yy are easily recognized as series expansion of ¢. To recognize ®,, ure terms of the infinite series should be evaluated. In fact #5, is tet. ‘The computation of eM by expanding it into a power series in ¢ and then adding the tomesponding elements in the matrix terms of ‘the infinite series, is practical only for simple tueee We will shortly discuss alternate methods of computation of state transition matrix. Returning to the example under consideration, we have the state transition matrix as ; ow = le 4] ‘The time response of the system is given by 4 x(t) = 00) [x + fore Budt| Now with u=1, : fe e*(1-2). oO peu wie Scanned with CamScannerCONTROL S¥STEY fi etdt 1-et ly yBu dt = : “Vet | per [eta-nae fe Jo Therefore, ‘Then the solution x(¢) is given by j os Ie Ps Computation of State Transition Matrix Computation by Laplace transformation Let us consider an unforced system whose sta / x = Ax, where A is a constant matrix ite equation is ‘Taking the Laplace transform of this equation, we obtain sX(s) — x(0) = AX(s) where X(s) is the Laplace transform of the u vector. Equation (12.74) may be rearranged as {sl - AIX (s) = (0) [st - A}? x(0) A124, anforced response and x(0) is the initial ens, or ‘Taking the inverse Laplace transform, we get x(t) = C(st - AY] x(0) where x(¢) is the unforced response of the system. This solution obviously must be iden: with the one obtained earlier in eqn. (12.73). The comparison yields a different approxi: determine the state transition matrix which is given below: (t) = et = £1 [(81 - A] = £1 Os) where (s) = (sI - A)" is called the resolvent matrix. Let us consider now the response when the control force vector u is applied. The equation for this case is x =Ax+Bu Performing the Laplace transformation gives 8X(3) - x) = AX(s) + BU(s); x, = x(0) or (sI-A)X (s) = x, + BU(s) Therefore X(6) = (G1 - AyIx, + [(s1 - A)! BUG)] -o By inverse Laplace transformation (= CMGI AY] x, + CU(6I - A) BU(S)] ‘a = O10 Xo + COS) BUG] a Scanned with CamScanneryAnBle ANALYSIS AND DESIGN 2.6 in which (t-aya[® sl %=[5 I ape determinant of (61 ~ A) ig \ ISI-Al s(oayp \ mperefore, the resolvant matrix is given by \ ®(s) = (I Ay =) “ | (s- Ll (=, 7 c-p ° i 1 1 | (-1F =D Hence & = CT - Ay = e 0 a 8] =e trample 12.7: Consider a control system with state model x x 0 x, (0) 0 (EJ[.2 ATE] [9] -B w= site Compute the state transition matrix and therefrom find the state response, i.e, x(t) 20. Solution. The characteristic equation for this system is 5 ee iF el eS Cuapter 12 eM vaeale|3 (a+3) ® #4304220 Therefore the eigenvalues of matrix A are 4, =~ 1, Aa = = 2. Since the system matrix A ‘sin the companion form, we can choose the Vander Monde matrix. me BEL 24a ae ‘the modal matrix. Then A=M1.AM x= Axt bu Scanned with CamScannerTransforming with modal matrix, we get mee +1 AMv-+ MO bu; x = My = Av+ Mtbu The state transition matrix in the transformed domain ole 4] This is easily established by writing out of the decoupled equations v(e) = ev(0) - ‘Transforming back M"x(t) = eM x(0) ae x(t) = (Me M-!) x(0) . b(t) = Me M4 Substituting values, we get ole Mie Oe See oo U1 -2]L0 oe fl-1 - 1] 204 a 20% et yoga ‘Writing down the expression for x(¢) x) =e x(0) + [eM Pbule) dz Computing term by term 1 x0 | DoF — 20" | Abs jut) = 1 fort ae} a f Yet =20* ) [rete T 0 |= 26" + 4eF ~ Bef +207 J, tte] 7 ang = 20 +2" |~|-242 Ret - e% 1 ~ ef +26 Collecting terms, et et et eX (t) = woe tae See Simplifying the above equation, we get Scanned with CamScannerpfIABLE ANALYSIS AND DESIGN ation by Techniques Based on the yaaa transition matrix may be fon Theorem (ie For large systems, this ere, using the technique based on the Cayley-Hamilton _( gher two methods advanced ear} is far more convenient computationally as compared lp a Tier. To being with, let us state the Cayley-Hamilton : WA=1A-Al ep ipecharactoristic equation of A, then | : oe : ae (12.79) This theorem provides a simple procedure 7 jestudy of linear systems, we are mostly concern series ofthe povters of a matrix. Given anare erpeqn- (12.78) above, let 2, 2, for evaluating the function of a matrix. In \ ed with functions which can be represented \s n matrix A with the characteristic equation | «vy 4, be the eigenvalues of A. The matrix polynomial RA) = hl + yA + hy? + 4b Ath, GAM (12.80) | anbe computed by consideration of the scalar polynomial BAD = hig byt bP? at hy dh + hey Ped 2.81) 4 Ifa) is divided by the characteristic polynomial q(J), then we have — | @. QA) + may 5 i qa) qa) & | « fA) = QUA) gl) + RO) (12.824) | shere R(A) is the remainder polynomial of the following form: RA) = 0+ 0 AH 0 BA cre + Oy PH (12.82) Ifwe evaluate /\A) at the eigenvalues Ay, Ayy nn 4, then q(A) = 0 and we have fla) = RA); i= nm -(12.83) The coefficients dy jy... Oh 4» can be obtained by successively substituting 2,, apy... | 4,into eqn. (12.83), \ Substituting A for the variable 2 in eqn. (12.82), we get | lA) = QA) a(A) + ROA) Since q(A) is identically zero, it follows that fia) = RA) = OL + AT + on + 0, AO (12.84) 2 which is the desired result. The formal procedure of evaluation of the matrix polynomial lA) is given below: 1. Find the eigenvalues of matrix A. 2. Ifall the eigenvalues are distinct, solve n simultaneous equations given by eqn. (12.83) {or the coefficients gy ay over, yea IEA possesses an eigenvalue 4, of order m, then only one independent equation can be ‘Sbtained by substituting 4, into eqn. (12.83). The remaining (m — 1) linear equations, which Scanned with CamScanner‘ust be obtained in order to solve for as, ean be found by differentiating yy im it (12.82). Since “a, 4 HAD! 6, 520, t,amy m1, a |, it follows that dig(a) RA) 5920) Uy seseey MAT i A iea, Ny, 3. The coefficients a; obtained in step 2 and eqn. (12.85) yield the re Example 12.8: Find lA) = A! for a-[2 J Solution, The characteristic equation is ited rey A -T q@)=1Al-Al -[3 pig/t@+DA+2)<9 Matrix A has distinet eigenvalues 4, =~ 1, 4, =-2. Since A is of second-order, the polynomial R(2) will be of the following form: RU) = a+ aA The coefficients , and a, are evaluated from equations fA) = ai Ry + Ay % =~ 1022, a, = - 1023 The result is From eqn. (12.85), we get fA) = As ols oA =/- 1022 - 1023" “| 2046 2047 ‘The Cayley-Hamilton technique allows us to attack the problem of computation a where A is a constant n x n matrix. ‘The power series for the series e*, et ale ate 4 MaTsary At aI converges for all A and for all finite ¢. Therefore the matrix polynomial (A) A eanbesse ‘asa polynomial in A of degree (n — stn } using the techniques presented earlier. Thisisillos™ below with the help of an example. Scanned with CamScanner