You are on page 1of 112
Fye basic characteristics of the transient performance of a closed-loop system are closely related 10 ine closed-loop poles, Ifthe system has a variable loop gain, hen the location of the closed-loop poles depends on the value of the loop gain chosen. The root locus technique, introduced in the previous } plore graphic tool that helps us know how the closed-loop poles move the plane as the loop scrip is varied. From the design viewpoint, smople gain adjustment may ove the closed-loop poles er dome systems to desired locations wnile simultaneously satisfying the steady-state performance equrements. If the gain adjustment alone does not yield the desived result, modifying the feedback atGTSy iidy Become nec- a structure and/or incorporating additional coniTor devices (compensa ea hs objective in this chapter isto learn procedures for compensétor design WN 100! locus ots The procedures for lead and lag compensator design, and PAD conto! design will be developed. “Tn Chapter 12, we will look at another method of design ‘based on frequency response plots. INTRODUCTION .Resall the basie Feedback system block diagram defined in Chapter 2 and shown i” abridged form as Fig, 91a, When we speak of a controller D(s) our concesn is mainly with ike information handling * devices; the sources of information axe the reference input x(), Which is derived from the corfimiand 5 Sry and the feedback signal B(), which is produced by the feedback path elements Als) fromm the = controlled output »(¢). The 0m ling devices—ctvato controlled process, itself—are represented by GG) in the block diagram of Fig. 9.12 “aris often possible and Convenient 10 Manipulate the block diagram into a iinity-feedback model shown in Fig 9.1b, where the reference inpat (is equal tthe commanded valve oF the controlled « Qutputs the system error appears explicitly as the actuating signal reise te fst chapter devoted exclusively to control system design. The design procedures dovel- coped inthis chapter are based on shaping the thre aspects ofa system's response discussed at length in Choplers 5 and 6, namely stability, spend, and steady-state accuracy ofthe closed-loop response Fora tai ; LICH be FE oF : 7 : z 0+ | 2) [Law + @ (b) Fig. 921 (a) A non-unityfodback system (0) A unity feedback systers y_Contrl Systems: Principles ond Design "table closed-loop system, all the closed-loop poles must strictly lie in the left half of the s-plane. Since ime objective of design, we always ensiire tat our Controller DG) StbIZES the Systern +y relocating the poles such that they all lic in the left half ofthe s-plane. Once we have ensured closed ‘op stability, we then refine the controller (without violating the constraint on stability) to achieve the vability is the desired performance. ‘The perform: g the spe Tous chapters is thot of turnii formance indices ean be used in this step. At several places in the preceding chapters, we desired result. As Subsystem D3) in the feedback loop (Fig. 9:1); nuroller/eompensator) The selection of a compen where n is the number of dk current chapter and in Chapter 12, respectively. 9.2 | RESHAPING THE ROOT LOCUS A simple control-system configuration of Fig. 9.2 is crcated By providing feedback around the plant G,{s) and adding an error-signal amplifier with gain K,, The amplifier gain js adjusted to provide acceptable performance, if possible. The system of Fig. 9.2 is offen referred to as having propor- sional compensation. If the plant dynamics are of such a nature that 8 satisfactory design cannot be obtained by @ gain adjustment alone, some dynamic compensation for ce is often specified in terms of figures of merit, such as peak overshoot, setting ime, steady-state error constant, ....., which we have already met; and bandwidth, phase margin, goin agin, _., which we meet later in this book. One of the ideas which has been emerging froma the fications into a target closed-loop characteris inial and then selecting the controller D(s) to match this polynomial. An appropriate choice of a target closed-loop polynomial is 10 use a second-order pole pair in the quadratic factor (2+ 2Cto,5-+ 2 and all other real/complex poles faster than this quadratic system is dominated by the second-order response characteristic. Overshooi, setiling time or other per- ic polyno- lt $0 that the response of the higher-order ¢ used gnin setting to improve the performance ofa feedback system. Mecely by gain setting, it may be possible to meet the given specifications of simple contol systems, However, in most practical cases, the gain AdjUSUnEnT does nor provide the usually the case, increasing the gain reduces the steady-state ersor but results in oscillatory transient response or even instability. Under such circumstanees, itis possible to introduce appropriate corrective action, integral or derivative, to force the plant to meet the given specifications. The job of integral or derivative actions is to compensate for the deficiency in the performance. The 1)is thus a compensator. The transfer Function D(s) will, in the sequel, be referred to as a controller transfer function or a compensator transfer Function: both the or) will point to D(s) in the feedback loop. ‘oF Dis) is basically a search problem in n-dimensional parametric space ign parameters of D(s). Points in the search space correspond to different selectiongof compensators. By choosing diferent points ofthe parametric space, we can generate feed- back systems with different performantes- A compensator that meets the given performance specifica- tions'€an be determined by moving in this search space on trial-and-error basis. Classical control theory. provides systematic methods for solving this trial-and-error search problem. The two most successfaf practical methods, the root-locus methed and the frequeney-response method; will be presented in the Ys) Gs) ier Plant Hs) Fig. 92 A simple control system configuration Sensor For we K wit dynan the ne tation toran Th selerr Fig. 9 We design Compensator Design Using Root Locus Plots _f@B vee be plt dynamics is required. A simple comolsystem Rs) + F> | vs) em _ Configuration tbat includes. dynamic compensation is Das) ed Shown in Fig, 9.3; the compensator D(s) is placed in = the ancade With the forward-path transfer fonction G,s). Cone ‘Tho system of Fig 9.3 s offesrreferred to as having cas- 7 He) ing caile compensation coamt ain “The dynamic compensator will generally also have the again K that has to beadjusted to improve steady-state pay oie ai dgiaile no- accuracy, ue, the transfer Function Ds) is typically of Fig 22th con see ee a compensation cae AE the form 2 a Ris) + Hs) a K][G+2) — KGs) aa D&Asy Tle (9.1a) - Plant (cr) nce @ a His) sof | =KD() (9.16) Sensor isl For the purpose of simplifying the design procedure, (@) i FE ve normally associate the adjustable goin parameter 7 aa Kowith the plant transfer finetion and then design the WO Dts) fra ) a dynamic compensator D(s) given by Eqns (9.1) based on SY chsmler Cham ta the new plant trarisfer function KG,(3)-At the implemen ioates sah tation stage, the gain K'is associated with the compen al tor and the transfer fimetion Ds) is realized. ere oe ‘The configuration of Fig. 9.42 will accordingly be ‘Sensor ey iefered to as the uncompensated system and that of (b) ede Fig. 9. 25 compensated system.’ - Fig. 9.4 (a) Unncompensated system fica ‘We introduce root-locus design through selected () Compensated syster design problems. cory ssfal Bea] : athe 992°| Design Problems FH Consider the systenm of Fig. 9.5 which has an open-loop transfer function k, G(s)= A 02 : wovn( eri] me From Eqn. (9.2), we have 7 5k, K 6) = = 13 " d= ern(srs) sr49) saad “The erm “cascade compensation’ used in Chapters 9-12 refers tothe use af compensator (controller) in cascade (oores) with te forwar-path transfer Fenton. It may be pointed out here that ia process contol fick the frm ation ‘taseade control is frequently used for feedback configurations having ‘Toop within 2 loop’ 8D Convel Systems: Prinples and Design ‘The root locus plot ofthe uncompensated system : eppears in Fig, 9.6. RG) Assume that the relative stability specification calls for a damping ratio £ = 0.45 (peak overshoot 1%). The Cline intersects the root locus at ~ 0.404 + 0.802. For the-deminant root pair, the undamped natural frequency a, is 0.898 rad/sec. Phe set time (= 4/0 foie espose i sbot a See The Uncompensaled system, teTee, bas i sluggish transient response. In order toimprovethe spel ofresponse,theroo! Jocus must be reshaped so that itis moved farther to the left aay from theimaginary axis. Toaccomplish this goal, Jet us consider the proposition of adding a compensating zero, Figure 9.7 shows the root locus plotofthesystem with the compeissatingzeroplacedat 5=~ 1.5. Notice that the effect ofthe zero is to move the locns to the le, toward the more sible part ofthe i plane. The { (= 0.45) line intersects the root locus at : ~2.2443, Forthe dominant root pair, the undamped : natural frequency «is 4.85 rad/sec. The sting time ! {or the response is therefore about":7 see, Thus, The addon ota compensating Zer WENGE To improve the speed of response for ¢= 0.85 >A further observation about the effect of adding \; compensating zero is made fiom Figs 9.6 and 97. K J ) r Without the compensating zero, the real root in Fig 26 Roo! lacus pol of 1+ Fs ‘additibA to the dominant complex-root pair is always [ocated Ihe Te oF the oper oop pole at s =~ 5. On the other hand, in the presence of compensating zero, this root moves over to the right of s =~ 5 gni the dominance condition is therefore weakened. Since the rel mot (which is the closed-loop pole) will. be Jocated close to the compensating zero, the amplitude of the response term due to this root will be small. Also, the time constant of the real root will be comparable to that of complex dominant root pair; therefore, its effect on settling time may be tolerable. 3 :n acceptable transient response can generally be obtained by trial-and-errér adjustments of the cofn- |pcnsating zero on the real axis. It can easily be verified that if we place the compensating zero too far to the left (say, near the open-loop pole at s=— 5), the dominance condition is improved but complex branches of the root Jocus are pressed toward the right and the dynamic response becomes poor. If, on the other hand, the zero is placed close to the origin (say, to the right of the open-loop pole at =~ 1), the complex branches are advantageously far into the left, but a seal root close to the origin appears. This root will also be close to the compensating zero; therefore amplitude ofthe response term due to this root may be small. However, since the real root will bave a large time constant compared to that of dominant complex root pais, its effect on settling time may be appreciable, - ‘No general rule can be laid down for selecting the location of compensating zero that satisfactorily meets all the design requirements. Trial-and-error procedure is the best tool in hand. However, the following guidelines can be of considerable help: 4 4 | | Compensator Design Using Roo! Locus Fels _ ABB jo Eth EO Piace the compensating zero. on the real axis el the region below the desired closed-loop pole Iocation such that it les clase to the left of any open- Ioop pole in this region. This technique ensures that she clasei-foop pole an the cent axis caused by the introduction ofthe compensating ero wil be located ‘lose to it with a time constant comparable to that of desired dominant closed-loop pole-pair, and would < therefore make a negligible contribution to system Pah iiynamics. In case, the mncompensated system does ~~ not have any open-loop pole in the region below the = —-— desired closed-loop poles; the dominance condition 5 ust be checked by locating the real-axis closed- loop pole created by the compensating zed. “CBIE-adtition ofa zero in the open-loop transfer fonction can be achieved by placing a compensator Di)astzes+ iit (9.40) in cascade with the forward-path transfer function, “Thiis results in (refer to Fig. 9.3) #15) Fig. 97 Root focus plot of e218 K+ 9 (0.46) Plott Dist DAs) It is seen that D,(s) given by Eqp. (9.4b) is a proportional plus derivative (PD) convoller: Duo= E04) = KOT) +5 Sand ss : beat ‘An op amp circuit realization ofthis controller is ofthe form shown in Fig. 9.8 (refer to Fig. P7.18 - for circuit details). 7 a ‘The problem with PD compensation is that the inevitable i high-frequency noise present at the input to the Ds) will 7 ‘be greatly amplified. The usual method for overcoming this ia problem is to add a pole to the compensator transfer fine- | Proportional control} + tion which Jimits the high-frequency gain. The pole must of o--—J circuit é couse be added such that it has relatively negligible effect i lon the root locus in the region where the two dominant ee losed-loop poles are to occur. Figure 9.9 shows the root locus plot in the presence of Derivative control both a compensating 2er0 and a pole. The root locus near the dominant closed-loop poles gets somewhat modified by the eee presence of the compensating pole. The effectoon the transient Fig. 9.8 A physienl reelization of PD response of the system is not very pronounced. The (= 0.45) controller e 2 2 x = os g Ss Se aaa baa ey Jing intersects the root locus at s =~ 1.531 + j3.039. For the dominant root-pair, he undamped natura} ie frequency @, is 34 rad sec. The setiling time for the response is about 2.6 sec. ie ‘AS can be easily seen from Figs 9.7 and 9.9, the effect of the compensator pole is to press the locus fae toward the right. However, if the pole is added far away from the jaraxis, it has relatively negligible effect on the root locus in the region where the obt two dominant closed-loop poles are to occur. ae 2 sys! ‘The choice of the compensator pole loce- . tion is a compromise between the conflicting 7 tro effects of noise suppression and compensa tor effectiveness. In general, if the pole is to0 - close to the cero, then the root locus moves hack too far toward its uncompensated shape, ‘and the zero is not successful in doing its job. The On the other hand, if the pole is to0 far to the syst left, the magnification of noise at the output of Xs) is too great and the motor or other acnuator of the process will be overheated by mse noise energy\in many cases, « pole located at Roo SF TisaAETFOm 3 10 10 times the value of the - zero location resolves the conflict. Vele We see from the above discussion that the Und South ofthe compensating zero; both the pole and the lzero confined to the left half of the s-plane. The : : Jranserfinction of such a firs-onder compen: pig 9.9 Root locus plot of 1+ —-KS-+1S) {ee then becomes ‘Note that cz< 1 ensures that the pole is located to the left of the zero. The contribiision of Ds) to'the angle criterion of the root locus is always positive. This type of compensator is called lead cov tor. The name ‘lead compensator has a clearer meaning with respecr to which is covered later in Chapter 12. —— pole must be located to the left s+ Dis + 5)(s413) s+ = sa sip svila@r DUs)= ‘The compensator (9.5a) results in (refer to Fig. 9.3) Root seit =x(S )eerroo PAs «(S5) a Velo. Itis seen that D,{s) given by Eqn. (9.5b) is 0 form of proportional plus derivative contvoller wherein, the derivative action is bandwidth limited. ‘Anfbp amp realization of this controller is of the form shown in Figé9.10 (refer to Fig. P7.23 for circuit details). 7 Kp Proportional control ‘| + cus ible ‘The following is the summary of the results cirenit obtained for uncompensated-and compensated systems. Kos ‘The system of Fig. 9.5 has an open-loop =< oe transfer function, Bandwidih-limited derivative contol circuit (s+ I)(s+5) Fig. 9.10 A physical realization of lend compensator as) ‘The root locus plot of this uncompensated stem is shown in Fig, 9.6. A damping ratio ¢- 45 yields the following pertinent date: 0.404 4 j0.802 Dominant roots, Root locus gain at s,, K= lols, + fs + 5]= 4.188 838 Yelocity-error constant, 898 radisec ‘Undamped natural frequency, vee), a 9 sec Settling time Figure 9.9 is the root locus of lead-compensated system with open-loop transfer Finetion m5 a, K(s+.3) i D(s)a¢s57 AS) _ 9.58) (9G) = aise Ss4 15) r to the A damping ratio = 0.45 yields the following pertinent date: onsa- asign, Dominant roots, 1,931 + 3.039 7 eS aeree tie Root locus gain at s,, afl til +515 #151 _ 999 154131 9.5b) Kas) locity error constan K, =lim sD(s)6(s)= 4a Velocity-error constant, = Him s D(S\G(s) = herein ‘Undamped natural frequency, a radlsec Settling time (ine positive angle contributed by the lead compensator shifis the root locus towards the left in the plane, this results in improvement inthe dynamic response of the system. The lead compensator also helps to increase system error constant, thongh fo limited extent\ Ko Ge + pattie a | Contro! Systems: Principles and Design |_Design Problem 2 nasa ou Consider the system of Fig. 9.5 which has an open-loop transfer function oe 5K, K = G(s)= At = * 9.6) GO" Brnisss) eres) ea 7 gk ‘The soot locus plot of the uncompensated system appears in Fig. 9.6. Suppose that the relativestability 3 —( specification calls for a damping ratio C= 0.45, The ¢-line interseets the root locus ats=~0.404+j0.802 ZF to fora value of K = 4.188. Therefore, the undamped natural frequency @, of the closed-loop system is fT 0.898 rad/see, and the settling time (= 4/€,) for the response is 9.9 sec. We assume that the dynamic jf co response of the uncompensated system with K' = 4.188 is satisfactory. tf co For the uncompensated system with K'= 4.188, the velocity-crror constant is q : q im sG(s)=lims: —_K__}_K 9305 gq 0 eo |siststs)) 5 ‘and thus with a unit-ramp inpot (1 rad/sec), the steady-state error 1 aiid KR which is too large. It is therefore required to raise the system X,, To accomplish this, consider the “3p ~ proposition of adding a proportional plus integral (PI) controller in cascade with the plant transfer fonction in the forward path. A PL controller has transfer function of the form O72) The PI controller thus, ads a pole ats = 0 andia ero at s=~2 10 the open-loop transfer function. : System ype increases by one becanse of the aided open-loop pole at the origin. Hence, for gible An op amp circuit realization of the Pf controller given by Eqn. (9.7) is: oF ‘the. form shown in Fig. 9.11 (refer to Fig. P7.19 for circuit details). Let us now investigate the effects of the pole-zero pair of the Pr controller fon_transient response a of the systein. If can easily be verified that the out ppole alone would move the root locus tothe right, stea thereby making the system unstable. The zero must Kp= a be near the pole in order to minimize the rt on See aa 1 the transient response. Figure 9.12 shows the root & ee @a vesu 1680S plot of the sysiem with the compen: Pa amp placed at s =~ 0.01; the compensating pole - * deta the origin. Notice that the angle contributicin of the Ky _KelTy iA compensating pole-zero pair at the desired closed- eS ont loop pole - 0.404 + 0.802 is negligibly small, and eee impr therefore the oot locus in the vicinity ofthis point is Integral control circuit apf not appreciably disturbed, Fig. 9.11 A physical realization of PI controller “BR Pens: JOBE —— see : Compensator Design Using Root Locus Pols _ AB) [A farther observation about the effect of adding pole-zero pair is made from Figs 9.6 and 9.12, With: “at the compensating pole-zero par, the root locus ha a real rot in action to the desired dominant complex root-pair — 0.404 + j0.802. The real.root is always located t the Jeft of the open-loop pole at 5 a therefore ils effet on transient response ofthe system will be negligible. On te other haat jn the presence of compensating pole-zero pair, there are two real roots jit addition to the desired co plex 1001 par; one Jp the le of open-loop poe at =~ 5 andthe other onthe reak-ax at between Mio? and 0.01 (The additional real root contributed by the compensating pole-zero pair is to0 close qo the origin and therefore, will have a vesy-largetime-constant, affecting the settling time appreciably: “pete is however, a zero very lose otis eal oot and therefore, the magnitude ofthe response ro contibuicd by his Toot Will be negheRoly small. The closed-loap pole-cero dipole tniroauced conmroléF will not contribute significantly to the transient response of the close 9.6) iliey 302 mis mie the asfer. Oe ction. table K(s+0.0) Fig. 9:12 Root locus plot of 1+ i ene (s+ e+5) <9. Itis desirable to place the compensating pole-cero combination as far to the left as possible with- out causing major shifts in the dominant root locations and giving required improvement in she steady-state response. The compensating pole-zera pair near the origin-(the pole not necessarily af the origin) will give rise toa closed-loop pole-cero dipole shified relatively to the left compared (0 the cuse of Pl controller; therefore, the time constant of the additional root will be relatively less resulting in a smaller effect on the setiing time of the system. In system response, a long sail of small amplimde introduced by closed-Foop dipote will decay relatively faster (refer to Example 9.4 Jor more 3 details). 7h: car easily be verified that if the compensating pole-cero pair is placed too far to the left (say, on the real-axis segment to the left of desired dominant complex closed-loop poles), no significant improvement in the steady-state response is possible. 7 igure 9.13 shows the root locus plot inthe presence of compensating pole at s = ~ 0.005 and the coms pensating zero at s =~ 0.05. The angle contributed by the compensator pole-zero palr at = = ODA + 40.802, the desired dominant roo, is about 2.6° and is therefore acceptable. For the damping ratio ponse roller y_Lento yelems: Pips and Design & = 0.45, the new locus is close 10 the locus of uncompensated system. The new dominant roots are ¢ S, := 0.384 £ 0-763; thns, the roots are essentially unchanged. x « geoas [7° 7 K eek Th 0 i - fo i os 3 x > An : for cin For the compensated system adjusted toa damping ratio of 0.45, the Following results ate obtained! = ‘ Dominant coos, 572-0384 + 70.763 Root Focus gain a x Sls tila Sil =401 The 15, 0.05] K(5+0.05) Velocity-error constant, K, = tims4——_X( #005) _ 58° |s( + I)(=-F5) (5+ 0.005) : E00) gop ; 5%0.005 Undamped natural frequency @, ~ 0.854 rad/sec. : 7 ‘A comparison of the uncompensated system and the compensated system shows thatthe undemped | natural frequency has decreased slightly from 0,898 to 0.854 rad/se, but there is considerable increase : in velocity-error constant: from 0.838 to 8.02. This increase in K, is in roximately. equal 10 [olip| where s = =z is the location of the compensating vero anid’ 7 = =p if the location of the compensating pote. eee We see from the above discussion that the compensating ze10 must be located to the left of the 3]. ~ compensating pole; both the zero and the pole confined to the left half of ihe s-plane, The transfer function of such a first-order compensator then becomes sez shit pz Dis)=S*2 = Es1e>0 9.80 (= Ep supe Om” (980) 7 amped crease, equal of the of the Compensator Design Using Reo! Lecus Plots gh oS) [Note that B > } ensures that the pole is located to the right of the zero, ie, nearer the origin than the sero. The contribution of Dis) tothe angle criterion ofthe root locus is always negative. This type of cempensator is called lag compensator. The name “lag compensator” has. clearer meaning with respect to frequency-domain design, which is covered later in Chapter 12 ‘The compensator (9.8a) results in (refer to Fig. 9.3) exile Day=K o 46) 8) (253i b> he (9.8) It is seen that D,(s) given by Eqn. (9.8) is a form of proportional plus integral controller wherein the integral action is low-frequency gain limited = ra K(B-1), é Proportional contro) - oy-x KBD : POE Fes Eset 8 Kr ystl Gain-limited integral ‘An op amp realization of the controller is of the control circuit form shown in Fig. 9.14 (refer to Fig. P7.22 for circuit details)- Fig, 9:14 A physical realization of lng compensator “The purpose of reshaping the root locus generally falls into one of the following categories 1. A given system is stable and its transient response-issatisfictory, but is steady-state erfor is too Tavge. Thus, the corresponding error constant must be increased while nearly preserving the tran- sient response. This can be achieved by a cascade Pi/lag compensator: 2. A given system is stable, but its transient response is unsatisfactory Thus, the root locus must be reshaped 50 that it is moved farther tothe let, away fiom the imaginary axis. This can be achieved by a cascade PD/lead compensator. : 3, A PD/lead-compensated system has satisfactory transient response but its steady-state error is too lage, The steady-state error can be decreased by cascading a PV/log compensator to the PDY jead-compensated system. This is equivalent to using a second-order compensator. Consider first the cascade of PI and PD compensators: x DAa=| Kit} J+ Kas) Ki) Ky Ki Ky + ats K Ks Kp+ Abe Kys i Q_Contal systems: Princes and Desi ‘Thus, cascading a PI compensator with a PD compensator is completely equivalent to a PID compen ‘sator, en op arp realization of this compensator is of the form shown in Fig. 9.15 (refer to Fig. P7.20 for circuit details. Consider now the cascade of lag and lead compensators Fille, \( sty =k af St¥8 | p31 0,4> a (2) eres Pirate This isa second-order compensator, called the{fag-lead compensator) Expressed explicitly in the form of proportional, integral, and desivative control actions, Ds) looks like K; Kps 2a9-(in +f) (nee - ) ‘An op amp sealization of this D,(s) is obtained by cascading a (low frequency) guin-limited PI com- pensator with a bandsvidth-limited PD compensator (refer io Fig. P7.23 for eircuit Jetals) “Tithe next three sections, we will focus on the design procedures for lag, Jead, and lag-leatl compen- sators. The design of PI, PD and PID compensators follows from these procedures ina straight forward way. Placing the pole of the compensator at the origin gives us a PI compensator (4 PI compensator js recommended for a system if compensator pole at the origin does not degrade closed-loop stability “appreciabhyy- Placing ihe pore oT a Tead compensator at ~ » (equivalently, designing a lead compensator by appropriately placing 2 zero only) gives us a PD compensator (4 PD compensator is recommended if the feedback loop is free from high-frequency “alse problems. This ts unlikely in reaP-life appl ‘afions), An example of PID compensation will be gived later in Section 9.6. Proportional eontrol ‘The systems, which-are-type-2-or-higher, are circuit, usually absolutely unstable. For these types of sys- ams, clearly PD/lead compensators are required, & Fee eee O hu Kp a5-only a PD/lead compensator increases the Kils margin of stability. Jn type-I and type-0 systems, Integral contol f-oparation is alas possible the gain © circuit saflicmly reduckd. In such cases, any of the : three compensators, namely, PWiag, PD/lead, or [kp PiD/lag-Iead, may be used to obtain the desired performance. The particular choice is based upon Derivative control transient and steady-state response characteristics of uncompensated system. Eig. 9.15 A physical realization of PID controller “53 | CASCADEL As has been pointed out in the previous section, lead compensation is generally used to improve the transient response of a system, In tis section, we will consider how the root Tocus may be used to esign a compensator s0-a8 to reelize a desired improvement in transient response. AD COMPENSATION | coe As) ed to Consider a unity-feedback system with a forward-path transfer function G(s). The transfer function G(s) is assumed to be unalterable except for the adjustment of gain. Let the system dynamic response specifications be translated into the desired jocdtion, for the dominant complex closed-loop pales as shown in Fig. 9.16. ae ‘AEE angle criterion at sis not mel, i. ZG(5,) # 180°, the “uncompensated root locus with variable open-loop gain will not pass through the desired root location, indicating the need For Compensation. Let us introduce a first-order lead compensator with zero and pole in the left half of the complex plane: <1,t>0 0.9) P ‘The compensator D{s) has to be so designed that the compensated root locus’ passes through In terms of the angle that criterion, this requi Compensator Design Using Root Locus Pots _ {Pm incascade with the forward-path transfer function G(s) 2s shown” Fig, 9.16 Desired dominant in Fig. 9.17. closed-loop poles Ms) Gs) = LDIs,)G(s,) = ZD(s,) + ZG(s,) = * 180° Ris) + O De 180° Z6(s) (0.10) é or 208, pensator pole-zero must contribute an angle @ given by Ego. (9.10) and shown in Fig, 9.18. For a given angle ¢ for lead compensation, there is ne unique, location for the pole-zero pair. Out of tions, we select the one that satisfies the following additional requirements: ‘The least-damped complex poles of the resulting closed-loop transfer function cor- respond to the desired dominant poles, and deat vesy close to_the-openlooe 22708 oF relatively far away from the jo-axis. This ensures that the poles other than the dom nant poles make 2 negligible contribution to ‘the system dynamics. 2. Changing the locations of the compensator pole and zero can produce a range of values, for the gain while maintaining the desired Guces maximum gain (and hence results in wv Ui) roduve -mar gain. 7 all other closed-loop poles are either located line “Thus, for the root locus of the compensated system Fig. 9.17 A feedback system with cascade tg.phss through the desired root location, the lead com- compensator Jf many possible solu- jo roots, The pole-zero pair location that pro- 7 ie aii sninimum steady-state error) is preferred. Fig. 9.18 Angle contribution of tead compensator Riighno vic Ubewhon : oe as B_Contol Systeme: Prnsples ana Design ee 3. Realization of requirements | and 2 may demand a large value of the ratio pz. IFwe choose this, ratio to be too large, high frequency noise problems can, and probably will, become significant . A rule of thumb that is sometimes used is to limit this ratio to 10 (pole is located at a distance 10 times the value of the zero locotion), but this depends on the phy/sieal systeni Under Comsivier “2 ation and particularly on the noise present. In the light of the above discussion, the steps for lead compensator design are summarized below. Translate the transient response specifications into a pair of complex dominant roots. 2. Sketch the pole-zero plot of the uncompensated system and determine using angle eriterion whether the desired root locations can be realized by gain adjustment alone. If not, calculate the angle deficiency @. This angle must be contributed by the lead compensator if the root locus is to pass through the desired root locations. 3. Place the compensating zero on the real axis in the region below the desired dominant root 5, If the uncompensated system has an open-loop pole on the real axis in the region below 5,, the ‘compensating zero should lie lo the left of this pole. 4. Determine the compensator pole location so that the total angle contributed by compensator pole zero pair at 55 9. 5. Evaluate the gain at s, and then calculate the error constant. 6. Repeat the steps if ransient performance is unsatisfactory because of a weak dominance condi tion and/or if the error constant is not satisfactory. Seep feedback type-2 system with Let us first consider the example of ‘open-loop transfer function K s Gs) It can be seen from its root locus plot that the closed-loop poles always lie on the jeoaxis.”As explained earlier, lead compensator is the only choice for this system. It is desired to compensate the system so as to meet the following transient response specificatip: t Seitling time, 1, $4 see Peak overshoo! for step input $ 20% These specifications imply that 2045, and o,21 We choose a desired dominant root location as 3, =o, + jo, Ji? =-1+ 72 igure 9.19a shows the pole-zez0 map of G(s), and the desired dominant root location s,. From this, figure, we find that LG(54)= 2X 116? =— 232° ‘Therefore, the angle contribution at s, required of the lead compensator pole-zero pair is == 180° ~ (~ 2329) = 52° : Place the 2er0 of the compensator on the real- be ‘axis segment direcily below sy, at s=—2=— 1 a join the compensator zer0 to 3, and Jocate the compensator pole by making an angle of = 52°, as shown in Fig. 9.192. The location of the pole is found to be ats==p =— 3.6. With the addition of the cascade lead compensator stl 536" the operloop transfer function of the system becomes DH 5 te Ks) ie Disjots) = KD (9) 0553.6) The goin K is evaluated by measuring the vector lengths from the poles and zeros to the root location sj: po O29829 1 2 Therefore,.the openloop transfer function of the compensated system is die Fig. 9.19(a) Casende lend compensation (Example 9.1) vith Bs) 3543.6) oy Dis\G § ye root locus plot of the compensated system is shown in Fig. 9.190. The third closed-loop pole sntroduced because ofthe led compensator lies onthe rezI-axis soot locus between =~ 3.6ands=— 7 Tha to this closed-loop pole, there will be an exponential term in the transient response jn-addition to the oscillatory term due to the complex pair of closed-loop poles. The specifications fs the (rane ‘Sent response were originally expressed in terms ofthe overshoot and ihe settling ime of the syste ‘Those specifications were translated onthe basis ofan approximation of the system by a seconchorder system, 10 an equivalent € and @, and therefrom to the desired closed-loop pole locations. However, 1s the original specifications will be satisfied only if the poles selected are dominant. Often the designer will simulate the final design and obtain the gctwal 1 transient response of the system. If the peak over- shoot and settling time indices of the transient response do not compare well with the specified ‘values, the system is redesigned ‘Computer simulation (MATLAB Module 9 in the Appendix) of the unity-feedback system with 0 open-loop transfer function (9.11) resulted in a peak overshoot of 46.7% and a settling time of 3.78 sec) fee for a step input (Fig. 9.19b). The difference in ned ons: Response Fig. 9.19(b) Simulation results (Exarnple 9.1) 2 x ¥ Cit STE PRES Bd Design ® the overshoot fiom the specified value is due to the third closed-loop pole, which is not negligible, A second attempt to obtain a compensated system with an overshoot of 20% wonld utilize a compen- sator with zero at say 5=- 2. This approach would move the third closed-loop pole farther to the left in the s-plane, reduce the effect of the third closed-loop pole on the transient response, and reduce the overshoot Finally, he error constants of the system are evaluiated. We find that the system under consideration and therefore will result in zero steady-state error for step and ramp input signals. The accel- is typi eration error constant is St Laas 36 The steady-state performance of the system is quite satisfactory. With the compensotor zero shifted farther to the Jeft ina redesign attempt, the error constant K, will increase marginally. funetion s¥2) It is desired to design a cascade lead compensator such that the dominant closed-loop poles provide a damping ratio € = 0.5 and have an undamped natural f¥equency @, ~ 4 rad/sec. Our specifications provide the location of the desired dominant closed-loop poles. An splane diagram showing the desired pole location s, and the open-loop poles of the uncompensated transfer function is given in Fig. 9.20. The desired closei-loop pole is at fre Hae : The angle of G(s) at the desired closed-loop pole, obtained from Fig. 9.20, is LG(s,)=- 210° r ‘Thus, if we need to force the root locus to go through the desired closed-loop pole, the lead compen- solor must contribute @ = 30° at this point Z Letus locate the compensator zero at s =—2.9, Le, in the region below the desired dominant closed- Joop pole location and just to the left of the open-loop pole at s = — 2. Join the compensator zero to sy and locate the compensator pole by making an angle @~30° as shown in Fig. 9.20. The location of the pole is found to be at s =" 5.8, ‘The open-loop transfer function of the compensated system becomes (Sales ‘The gain K at s,, evaluated using magnitude criterion, is 18.7. It follows that 2+ j3.46 5q=~ $e, + je, D(s)G(s) , _18.1(542.9) PS)GKs)= s(s+2)(s+5.4) ompensator Design Using Root Locus Plots jon ele ane Eas Fig. 9.20 Root locus plot (Example 9.2) +-Lrhe root locus plot of the compensated system is shown in Fig. 9.20. The third closed-loop pole is at |:=~3-4 which js lose to the added ero ats =~2.9. Therefore, the effect ofthis poke on the ansient response iseapected to be'small. This should be verified by simulation study on the closed-loop transfer Funétion %O) yyy 18.1s-+2.9) ae RO G24 BAG\(s+2— 3.46) (543.4) ‘Simulation study (MATLAB Module 9 in the Appendix) shows that the résponse has a peak overshoot sede (0f 20.8% and a settling time of 2.05 see for a step input. : - ‘The velocity-error constant of the compensated system is the : 18.7x2.9 1 ee = BIR29 _ 5.02 set BSD sa ‘Although a certain change in the value of K, can be made by altering the pole-zero location of the lead ‘compensator, if'a large increase in the value of K, is desired, then we must alter the lead compensator to a Jag-lead compensator: erecraheEsareSesarase ~ ‘A nity-feecback system with open-loop transfer function Gs) © F(st15) js to be compensated to meet the following specifications: ¥ i RB _Contrl Systems: Principles ond Desion Settling time, 1, < 5 see Peak overshoot for step input = 25%. For a second-order system, damping ratio = 0.45 gives 20% overshoot, and with undamped nat- ural fiequency @, ~ 2.2 radisee, the settling time 1, = Aw, = A sec. These specifications provide the location of desired dominant roots 12 = Gy alte Atsy==) 4/2 (refer to Fig, 9.21), 265) =~ 2% 116°— 75° =~ 307° “Therefore, the angle contribution required from a lead compensator is = ~180°~ (~ 307°) = 127°. “The large value of @ here is an indication that ‘4 double lead compensator’ is appropriate. Each section of the compensator has then to contribute an angle of 63.5" at sy In the region below the desired root, there is an open-loop pole at ¢ = ~ 1.5. We may place 2er0 of the compensator at s =~ 2 = ~ 1.5, canceling tie plant pole. Join the compensator zero to 5, and locate the compensator pole by making an angle of 6 = 63.5°, as shown ip Fig. 9.21. The location of the pole is found to be at s = ~ 10. Witit the cascade double lead compensator DE= er the open-loop transfer function +10" of the compensated system becomes, r JO, Fig, 9.21 Root locus plot (Example 9.3) Kissy K(st1.5). S (sti Sy(s+10Y s?(s+10)* D(s\G(s) “The gain Kat sy, evaluoted using magnitude criterion, is 247. It follows that 2a(s+1.5) S(sr10)P 'A rough soot locus plot of the compensated system is shown in Fig. 9.21. By locating all the closed- Joop poles and carrying out a simulation study on the closed-loop system, we can determine whether the dominance of the complex pole-pairis ensured, i.., whether the actual transient response compared ‘well With the specified one. Simulation stady shows that the response has a peak overshoot of 53.1% nd a settling time of 3.96 sec fora step input. More attempts on compensator design are, therefore, required. D(s)G(s)= Aa important consideration governing this choice i the inherent noise in control system. Refer to Example 12.2 Compensator Dasign Using Root Locus Prots £°§.4 | CASCADE LAG COMPENSATION n Section 9.2ca lag compensator jmproves the steacly-state bel n, while nearly preserving its transient response. Tn this section, we consider how the root locus may be usedT Wo desigirs Comipensaior so as to desixed improvern ‘Consider a unity-feedback system with 2 forward-path transfer funct ‘As pointed out ear 2 (9.12) pert assume th 3 acer we “this system has satisfactory Lransient-respense;-he--its-raot locus plot passes through (or is close.to).the.desired closed: oop pole locations, indicated in Fig. 9.22. Itis required to improve the system error constant K, K, or.X, (depending on whether the system is type, -1 or -2, respectively) to a specified value without impairing its transient response. This requires that after compen- sation, the root focus should continue to pass through “jj, while the error constant ats, i raised tothe specified “dine. To accomplish this, consider adding a lag com- pensator pole-zero pair close to origin, with zero to the sielt of the pole. If pole and zero of this pais are located close to each other, it will contribute a negligible angle ‘Aats, such that s, continves to lic on the root locus of the compensated systetn, Figure 9.22 shows the location py ee of such a lag compensator pole-zero pair. See eee “The gain of the uncompensated system at, is given by Pov Pi tet Thee, s B54) 2 0.13) Tiss For the system with a cascade lag compensator ser sellt oi D(s) sep stWBr the system gain at sy is (refer to Fig. 9.22) _Leveot Systems: Principles and Design oF where a= sy Vfl, ond B= isp Wt : Since the compensator pole and zero are located close to cach other, they are nearly equidistant from : 5,6... I, therefore, follows from Eqns (9.13) and (9.15) that ‘The error constant K, (which i uncompensated system is given by oe. The the I s \ 1 t vert oT (9.16) root “Thus, the P-parametet of the lag compensator (refer to Eqn. (9.14))is nearly equal t6 the ratio of the baad desired error constant to the error constant of the uficompensated system. ed i Since the lag compensator does contribute a small negative angle 2 at sy the compensated root locus oa | «will not pass through 5,. The effect of the small lag angle 2 is to press the root locus slightly to the right. aaa { The intersection of the compensated root locus with the specified G-line will be at s having slightly eS y i lower value of @, compared to the one corresponding to s,. Also the actual error constant will somewhat bake | fall short of the specified value if B obtained from Eqo. (9.16) is used. ate | ‘The effects of lag angle 2.can be anticipated and counteracted-by taking design parameter 0,10 be | T 1 somewhat larger than the specified value. Also ia the design process, we may choose a somewhat aaa larger than that given by Eqn. (9.16). ‘ In the light of the above discussion, the steps for lag compensator design are summarized below. L. Draw the root locus plot for the’ uncompensated system. 2. Translate the transient response spetifications into a pait-of complex dpminit roots and locate these roots on the uncompensated root locus plot. I is assumed that acceptable transient per- ; formance can be obtained by gain adjustment; the desired dominant root will, therefore, lie on (or close to) the uncompensated root locus. 26, ot Compensator Design Using Foot Locus Pots 7 +3, Catculate the gain of the uncompensated system at the dominant roots, and evaluate the corre sponding error constant. 44, Determine the factor by which the error constant of the uncompensated system should be increased to meet the specified value. Choose the P-parameter of lag compensator 10 be somewhat greater i than this factor, —__ : 5, Select zero (s =~ 2) of the compensator sufficiently close to the origin. As a guide rule, we may construct 3 Tine making an angle of 10° (or less) with the desired ¢-Hine from s,¢ the intersection of this line with the real axis gives location of the compensator zero. 6. ‘The compensator pole can then be located at s=~ p= ~2/B. Its important to note that the com. jpensntorpole-zeo pair should contribute an angle less than 5” at. 80 thatthe oot locus plot in the region of ss not appreciably changed and hence satisfactory transient havior of he systern is preserved. Reconsider the uncompensated system of Example 9.2, where the uncompensated open-loop transfer function is K GO Ker “The system is to be compensated to meet the following specification Damping ratio, ¢= 0.707 Settling time, ¢, 5 see ‘Velocity-error constant, K, 24 __$The uncompensated root locus is a vertical line at 5 =— 1 and results in a 16) oot on the ¢ = 0.707 line at sj=~1 +i. as shown in Fig. 9.23a. Setiling time = Unsompensated locus tthe corresponding to this root is 4/¢¢2,= 45 ~ ‘The gain at s, is K = 2. Therefore, the Compensated locus oe uncompensated system with “the gain : a adjusted to aalue of 2 meets the transient io response specifications. Slight reduction oe in @, due to the small lag angle of the compensator required to improve steady- 8 ~ ae state performance, is also permissible. 2 | i i Saat ‘The velocity-error constant of the a | i uncompensated system is | : ‘ Ky I 7 ! vente “The desired velocity-ertor constant is \ per Z 1 eon eee Fig. 923a) Roo! locus plot (Example 9.4) 4 _Loniol Systems: Principles and Design 2 ‘Thus, the velocity-error constant must be increased by at least a factor of 4. Lag compensator with B= Sis expected io meet this requirement. We might set compensator zero at s=~2=~0.1. Note that {| is much smaller compared to |, V2. The compensator pole is located at s = ~ p = ~2/B = ~ 0.02. ‘The difference of the angles from compensator zero and pole at the desired roots, is about 3°; therefore, 52-1 + J] will ie close to compensated root locus?” ‘The open-loop transfer function. the ae system becomes s Hl USGL) (33 ts] ‘The root locus plot of the compensated system is shown in Fig. 9.23a. It intersects the £= 0.707 line at «= ~0.96 + j0.96. The gain X at 5), evaluated by magnitude criterion, remains approxiniately equal ie to 2. It follows that 25%0.)) Piet (5+ 0.02)(5+2) From Fig. 9.23a, we find that corresponding to K = 2, the third closed-loop pole is at s=— 0.11. Note that the real pole has a time constant of t= 1/0.11=9.1 sec, which would appear to make the system so slow as to be unusable, However, there is a zer0 close to this additional pole. The net effect is that the settling time will increase because of the third pole, but the amplitude of the response term contributed by this pole will be very small. In system response, a fong tail of small amplitude will appear which may not degrade the performance of the system appreciably. This should be verified by simulation study on the closed-loop transfer function, FO) As+0.1) RCs) (5+ 0.11) 540.96 j0.96)(s+0.96— 70.96) i Simulation study (MATLAB Module 9 in the Appendix) shows that the response has peak overshoot of 12% and a settling time of 15 sec (Fig. 9.23b). | 1 4 Response eecee_c o 5 w 15 20 25 130 - 1(sec) Fig. 9.23(b) Simulation resulis (Example 9.4) ote he ed ch of Compensater Design Using Root Locus Plots 46%) CASCADE LAG-LEAD COMPENSATION tn the preceding seetions, we have seen thatthe lend compensator is suitable for systems paving unsa jsfactory transient response. However, it provides only a Timited improvement in steady-state response. iilhe steady-state behavior fr highly unsatisfactory, he Teal compensator may not be the aset Onthe ther hand fer systems with satisfactory transient response but unsatisfactory steady-state responds, the Jag compensator is found to be a good choice. Vien beth ransientand steady-state responses are unsatisfactory, we must draw upon the combined powers of lag and lead compensators in order to meet the specications ‘The transfer function for the Jag-lead compensator is of the form D(s)=| 222 \(e3) (9.172) 5 pa Nop, stil \/stin 52 {stile (see Fetn>0,720 11 (eae) Beata fae Tap Secon Lead Sao In designing a lag-lead compensator, we frequently choose B = 1/or (this is not necessary: se € or cnuree choose 12 17). The reason for this choice is that withthe reduetion in number of design porarneters, the désign problem becomes simpler. With B= 17a, the transfer Fancion (9.17)of the lag- Jead compensator reduces 10 lad % malig Zen elisa Uy HOR Sten 4 pia =( 22 (22 rete > 0 Bowes (0.18) alt, }\ sever, Verdin) ee be treba Bs ace Taam by Op , By Shee eb We first design the lead section to realize the required Gand @, for the dominant closed-léop potestfnxy 5¢ Fee me eneis then determined For the Jead-compensated system, say itis Ky. EK is suf Tonal bigh to give the Jesired steady-state response, we need not go further. Lead compensition i all that is required. If, however, KE is not large enough, we consider adding the lag section with B= l/c. his results in an error constant hairs kiko ike (0.19) Af KIS, as given in Eqn. (9.19), is not large ‘@nough to meet steady-state specifications, we choose i> L/eeand redesign the lag section. a 4 ' (605) Gs) “We iesign a feedback system to meet the following specifications: een 1. Damping ratio of dominant closed-loop poles, R(s)+ 205 ~( i 2. Undamped natural frequency of dominant Le i closed-loop poles, «, = 5 rad/sec. see i eared Fig, 924 Control configuration (Example 9.5) We choose the nity-feedback control configura- sion shown in Fig 9.24. Root locus plot of the system with A > 0 and D{s)= 1 is shown by dashed line aa in Fig, 9.25, The £~ 055 line intersects the root locus at the point s=~ 0.25 + 0.433. The undamped ok 9.24 with natural frequency corresponding to this root is 0.5 rad/sec. Therefore, the configuration in Fig D(s) = 1 cannot meet the specifications. D ‘We introduce a caseade lead compensator. From the performance specifications we find that the 7 dominant closed-loop poles must be at s =~ 2.5 + fa.33. Consider the desired location, : 2 will, sy=254j033 Since (refer to Fig. 9.25) ZU) 115° = = 235°, the lead compensator must contribute 55° so that the root locus passes through the desired root s, To design the lead compensator, we first determine the location of the compensator zero. There Lead are many possible choices, but we shall compensated hore choose the =0.5 sothat root locus it cancels the open-loop pole at s =~ 0:5. Once the compensator zero is chosen, the compensator pole ean be located such that the angle contribution of the pole-zcre pait js 55°, BY simple graphical analysis (refer 120° g=05 ti Uncompensated root locus, dagen Th to Fig. 9.25), we find that the pole must be Jocated at 5 =—p=~ 5. Thus, with the lead ‘compensator Dysy= UL #95, star, 5457 the open-loop transfer function of the system becomes, cs sus plot (Example 9.5 pyootey=—Eet09_2 _* ay 1-925 Root locus plot (Example 9.5) Se+ 0545) 45) “The value of K obtained by magnitude criterion, is The d the me Kafsylhsy #51525 select! Root Jocus plot of the lead-compensated system is shown in Fig. 9.25. Let us evaluate the steady- backs state performance of this system, anes Th = lim sDy(s)G(s)= other Ky Tey SNe) invoiv Compensator Design Using Root Locus Pots sal :_ ig “The desired K, is 80. A lorge increase in K, cannot be obtained by adjustment of pole-zese-pairotthe read compensate, We, therefore, must alter the lead compensation fo lag ead compensation Talus design a lag compensator e «sty P= Tipe, OP 15) {or the lead-compensated system. IF we choose B = 1/a= 10, the velocity-error constant will increase bya factor of about 10, Le, the velocity-error constant ofthe lag-tead compensated system will become about 50. The choice = I/c, therefore, does not meet our requirements on steady-state accuracy. ‘The velocity-error constant of the lead-compensated system is Kf = ‘The desired velocity-error constant is KJ = 80. “Thus, the velocity-crror constant must be increased by a Factor of 16. The lag compensator with B= 16 ‘will meet this requirement. We now choose the pole-zero pair of the lag compensator, We want that ine ped vith the 33 5° < LDyl sy) = £5 +8, L5y + Py

] G)= =p changesin therefrenceinputthaleoukdpreduee = P10) FT plant input saturation, In such cases, a practical implementation of compensation is realized by means ofa minor loop around the plant imple- ‘menting derivative term, with a proportional (P) ‘or a Jag (PD) compensotor in the forward path (Revisiting Section 75 will be helpful). ig- 9.29 Rale feedback compensation {EGP Te feeahack system shown in Fig, 929 provides the opportnity for varying PIC 27! three parameters: 4, K, and K,. The larger the number of variables inserted into 2 em, the better the opportunity far achieving a specified ance. However ie design problem becomes much MGFE complex The characteristic equation of the given system is (Fefer to Fig. 9.29) 14 409), __** _-9 2 (0.20) TE5K,G) eHI0)+ KK, inging, we pet P4104 KK s+ = 0 Gompensator Design Using Root Locus Pots _ #8) “The parameters to be selected are c= AK, and f= KK, The characteristic equation in terms of and B parameters is 8410s? +Asta= 0 riours) to study the total effect of both ct and fon the ystem (refer to Review Example 5.3) c equation, we can use single-parameter root loci for We may plot a family 6f Footdoci (the root cor roots of the characteristic equation of the given s ‘By appropriate partitioning of the characterist the design of three parameters (4. K, K,) ofthe given system Consider the following partitioning: Al KK MED g (9.21) s¥(5+10) net effect of rate feedback is to add a zer0 ats suitable location for this additional zero to s ‘This equation shows that the procedure calls for a search for tisly the specifications. zi aaa thatthe system i tobe compensated to meet the following specifications vibe Percent peak overshoot < 10% som: Setting time e4sée onse “The damming Cand natural Fequency t, ofthe dominant rots, obtained fom rhe elations eeNE 0; and A fo, 0.6, «, = 1.67 radisec. are ‘The desifed dominant foots are then given as wally gyno VELBA, ator). 2B“ yrrom Fig. 9:30, we sce that the angle “Contribution of the open-loop poles at s4 is = 2(127°) ~ 8° = ~ 262°, Therefore, for the point s, fo be on the root locus, the compen- sating zero should contribute an angle of $ =180° — (-262°) = 82°. The location of the compensating zero is thus found to be at s = =1.2 The open-loop transfer function of the ‘compensated system becomes KK,(541.2) $%(s+10) Fis) (9.22) ‘The root locus plot of the compensated system is shown in Fig. 930. It is now observed that forall values of KK, the system is stable. The value of KK, ats, is found to be 174. The third closed-loop pole is at s = -8 which is far away from the imaginary axis compared to dominant closed-loop poles. Hence, its effect on the transient behavior is negligible. ying intoa bem 129) ‘e209 Fig. 9.30 Roo! lacus plot for the characteristic equation (9.21) 22 el Control Systems: Principles and Desi a y ik ‘on From Fig, 9.30, we observe that the root locus intersects the ¢ = 0.6 fine at two points, P and Q. Thus, two valves of AX, will give the damping ratio (of the closed-loop poles equal to 0.6. KK, at point P results in a pair of dominant complex-conjugate closed-loop poles which satisfy the transient response specifications, The third closed-loop pole is at s=~ 8 which has insignificant contribution to system dynamics. At point Q, the dominance condition will be weakened the third closed-loop pole on the real axis will force a slow overdamped response. It is important to point out thai the zero at s=— 1.2 isnot the zero of the system of Fig. 9.29. Its the open-loop zero of Fs) (refer to Eqn. 9.22) which was introduced in the process of partitioning, the characteristic equation such that the adjustable variable KK, appears as a saliiplying factor Let us now investigate the steady-state behavior of the compensated system. From Eqn. (9.20), the velocity-error constant K, is given by ln abe sik, =12 S80" | S2(s4]0)+KK,s | KK, If this value of K, is acceptable, then the design is complete. Otherwise, the design procedure given above cannot satisfy the steady-state and transient specifications simullaneously. A tr the stendy state and transient behavior becomes necessary We have three parameters and two equations: AIR, = 1.2; KK, = ina 0.4, then the other two parameters become: A = 0.48 and K.= 43.5. K can be dec Iwve select X, 8 gives 4 = 0.96 and K = 21.75. with no change in performance by increasing and K; e.g, K, 9.8 REVIEW EXAMPLES SGI The block diagram of Fig, 9.31 represents a position control system, The BE! open-loop transfer function of the uncompensated system is RG) + z Ms aS OX PO.) {2 ss+ Doe) Compensator Plant The specifications of the system are as follows: Fig. 931 A position control system Damping ratio, 05 Undamped satural frequency, 2 radlsee Velocity-error constant, K,25 sec? ‘The desired dominant closed-loop poles are required to be located at fo, jo, F K Cr e plant G(s) =. closed-tor 0 plot ofthe plant Gls) = 7 anda desired closed-loop 1/73 Sia Figure 9.32 shows the pole~ pole at s, = —1-+ 1.73. From this figure, we find that LG (5) = —120° = 90° —30° = -240° ‘The point s,, therefore, does not lic on the uncompensated root locus and transient response specif cations cannot be met by gain adjustment alone. We propose to add a cascade lead compensator to meet the fre po wh loc cal at she the compensator Design Using Root Locus Plots f wo. iperransen sponse speciation. In rd forte ox ls ogo Hough tear onion fiom the lead compensator pole-zero pair must be =~ 180° — ZG(s,) =— 180° (- 240°) = 60" rorthed itis observed thot the open-loop pote ats = ~ 1 Ties directly below the desired closed-loop poe location, It appears that the best transient response due (0 th= complex closed-loop pole will occur aoe rhe zero of the compensator cancels the open-loop pole ats = ~ ls there will be no real closed: Joop pole near the imaginary axis in this case. This ideal situation, however, € anmot be realized pra of imperfect cancelation. A real cl josed-loop pole near the imaginary axis will occur but cally because cy lt be very else to tthe dominance condition wil, therefore, be ensured sree the compensator Zero 10 sy and Tocate the compensator pole making an anele of @ = 60° as jobe ats == 4 shown in Fig. 9.32. The location of the pole is found t f=05 eased 75. —s nThe 0.25 |-0.06 Fig. 932 Root locus plot (Review Example 9.1) With the addition of the caseade lead compensator sein, sea,” Dls)= d-loop * 7 a the open-loop transfer Function of the system becomes DAs) specifi. omect AB_Conirol Sysiems: Prnelptes and Design The value of K at s, is found 10 be 24. Therefore, 24 DIG et ssa)" The velocity-error constant of the lead compensated system is This does not meet the specified K, > 5. A lag section D,(s)= will increase A. by about four times, which then satisfies the sp Ur) s+Wty stalt, stir, RE = lim 5 D(s)G(s) = 1.5 ification on steady-state performance. ‘The line drawn from sy making an angle of 6° with the desired ¢-line intersects real axis at ~ 0.25, which gives the location of the zero of the lag section. The pole of the lag section is then found to be at~ 0.25/4 =— 0.06, ‘The open-loop transfer function of the lag-lead compensated system then becomes Dyls)DMs)OUs) = K*(s+0.25) (5+ 0.06)(s+4)) ‘The rot lee plot fr the Ing-tead compensated system is shown in Fig. 9.32. From this figure, the it at sj (which is slightly shified from s, due to introduction of the lag section) is given by K* Therefore, the open-loop transfer Function of the fag-lead compensated system is Dy{s)D(s)O(s)= poeta MstDG+5) a) Design a feedback system to meet the specifications: |. Velocity error as small as possible 2. = 0.707 3. Settling time < 4.5 seconds Solution Asa first try, we choose a unity-feedback system shown in Fig, 9.33. The root locus plot of this system for 4 > 0 is shown in Fig. 9.34. In order for the feedback system to have settling time Jess than 4.5 seconds, all the closed= loop poles must lie on the Jefi hand side of the vertical line passing through the point ~ 4/1, = ~ 4/4.5 = — 09. 40.25) s{s +0.06)(s+ 4y Consider a plant with transfer function Fig. 9.33 A control configuration (Review Example 9.2) aiid tion ter wh eq Eq poli foct Gompensotor Design Using oot Locus Pls _ : From the root loci in Fig. 9.34 we see thot jo this is not possible for any > 0. Therefore, the configuration in Fig. 9.33 cannot meet the specifications. ‘As a next ty, we intiSace-an additional tachogenerator feedback as shown in Fig. 9.35. : ‘The characteristic equation of the system : becomes i 4G(8) $= 0 (9.23) 14+5K,G(s) (9223) : where G(s)= =. 5, m (= Sayer) be Rearranging, we get PH62+(S42K)s+2A=0 (9.24) Fig. 9.34 Root locus plot for the system of Fig. 9.33 “The parameters to be selected are @ = 24, and K'= 2K, The characteristic equation in ge) 4 i fo a terms of @ and K parameters, is IO to : P4652 45s+Kobo=0 at which may be rearranged as a root Jocus equation: Fig. 9.35 Tachogenerator feedback contr configuration (Review Example 92) Ks =0 9.25 EE Pros +5sea . i ) We will carry out the iterative design by choosing a value for a, and then plotting the root loci of Eqn. (9.25) as K is varied from zero to infinity. Our frst trial value is a = 10 (+ = 5). The open-loop : poles ofthe root locus Equation (9.25) may be obtained by computer simulation or by sketching locus plot of the equation (refer to Review Example 8. F560) w H+ 6575) For a= 10, Eqn. (9.25) becomes Ks face 0 * GH SAIS F029 4 f) 33 s+ 02! }— 71.33) i B_Controt Systems: Principles and Design The root loci of this equation are plotted in Fig. 9.36. The = 0.707 line intersects the root focus at ro points. Both the points Tie on the left-hand side of the vertical Tine passing through ~ 0.9, From Fig. 9.36, we see that the point ~1,2 + /1.2 thot corresponds.to K = 6.3 (or K, = 3.15), meets the dominance require- rent better compared to the other point. We therefore select K, ~ 3.15, Let us now investigate the steady-state behavior of the compensated system. From xg K= 63 Eqn. (9.23), the velocity-error constant K, is 3 —-——o—— given by 52 3.65 im s| 2648) S40" [TF9K, GS) 10 0 jim $| so [see (et3)t6as) 113 ‘A marginal increase in Kyabove this value cam be achieved by readjustment of the parameter Fig. 9.36 Root locus plot for the system of Fig. 9.35 CONCLUDING REMARKS a best In this chapter, we have considered alternative appronches io the design of feedback control systems aa using root locus plots co 17 What we have learnt? + The root locus is reshaped by adding additional open-loop poles and zeros-via a lead/lagfead-log, compensator or a PID control function in a selected feedback contro] stucture—a classical control confignration with/without minor-oop feedback. The additional poles and zeros mist be located by 3 trial and error so thatthe second-order approximations used in the design are valid All closed-loop 3 7 poles besides the dominant second-order pair, mast yield a response that is much foster than the ae Gesired response. Thus, these poles must be at Teast 5 times as far from.the imaginary axis as the dominant pair. Further, any zeros ofthe system must be close to a pole for pole-zero cancelation of far from the dominant pole par. The resulting closed-loop System can then be approximated by two : dominant poles. If second-order approximations cannot be justified, then a simulation is required to rake sure the design is within tolerance a ‘The dominant pole pair controls the transient response of the closed-loop system. Best results for a transien response design re achieved through lad compensator/?D control function, PD control & fimetion adds an open-loop zero to compensate the transient response. Lead compensator, on the other hand, adds an open-loop pole along with an open-loop zero, giving beiter filtering. 135 sag ontrol ted by Hoop an the as the ion or ay 30 ired 0 sits-for zontrol fon the Compensator Design Using Root Locus Plots _ #F%y + Best resulls for steady-state response design are achieved through lag eompensator/PI control func- tion. PI control fanction adds an open-loop pole at the origin, thereby increasing the system type. jag compensator, onthe other hand, places the pole off the origin but near it Both the methods add a ‘zero very cJose to the pole in order to minimize the effect on the transient response Wecan correct both the transient response and the steady-state error with Jag-lead compenssator/PID control function. + A minor-loop feedback contsol configuration with controller gain in forward path and derivative gain in feedback path, has also been investigated using root locus plots 9.1 Discuss the compensation characteristics of eascade lag and lead compensators using root locus plots. Show that (@) lead compensation is suitable for systems having unsatisfactory transient response, andl it provides a limited improvement in steady-state performance; and (b) lag compensation is suitable for systems with satisfactory transient response but unsalisfac- tory steady-state response. 12 In many control systems, it is desirable 10 use a feedback compensator forming a minor loop around the plant, in addition to a cascade gain compensator which is used to adjust the overall performance. Discuss compensation characteristics ofthe minor-loop feedback scheme N°Si of questions with multiple answer choices is given below. These questions have been designed 10 capture important aspects ofthe subject covered in the chapter Por each ofthese questions, identify the test of the given answer choices, Compare your answers with the Master Key given at the end! of the session. This quiz should serve the purpose of a sclf-appraisal test for readers. 9A°‘the transfer function of a lag compensator is Lrors: = 720 PO ee ‘The valne of cris given by : (Ayal @®)a>1 Oa! (D) cris any constant 9.2. A compensator is of the form n= K (SE ao.g.2 0.250 i as41 ‘This compensator may be realized using opamp circuits for Vp (A) proportional plus integrator action : (B) proportional plus gain-limited integrator action 1 (C) proportional plus derivative action (D) proportional plus bandwidth-limited derivative action 2_Contto! Systems: Principles and Design 9.3 compensator is of the form This compensator may be realized using op amp circvits for (A) proportional plus integrator action {B) proportional plus gain-imited integrator action (C) proportional plus derivative action (D) proportional plus bandwidth-limited a tive action 94a compensator is ofthe form ‘ DI) = Ky K,>0, K,>0,1,>0 ast 5 This isa (29 Tag compensator {B) lead compensator * (C) lag-lead compensator i i (D) None ofthe answers in (A), (B), and (C) is correct : 9.5. compensator has a transfer function i Dis)= Ky A K,>0,K,>0,4>0 5 ee ; (A) |g¢ compensator (Bead compensator (C) lag-lead compensator AD) None of the answers in (A), (B), 2 9.6 Consider the following statements: he effect of compensating pole is to pull the root locus towards left. (Gi) The effect of compensating zero is to press the locus towards right. (A) None of the above statements is true. {B) Stotement (is tre but statement (ii) is false. (©) Statement (j) is False but statement (ii is true (D) Both the statements are true. 17 Consider the following statements: 7 (i) For a noise-less system, PD compensation is a better choice than Tead compensation for tran- sient performance. (i) Pl compensation isa better choice than lag compensation for steady-state performance, if both yield acceptable relative stability. (A) None of the above statements is true. (B) Statement (jis true but statement (i) is false. (C) Statement (j) is false but statement (ii) is true (D) Both the statements dre true. ¥-8- Consider thesfollowing statements: {i). Replacing Pl compensator with lead compensator will increase system sensitivity 10 noise (i) Replacing PJ compensator with Iag compensator will result in smaller negative effect om transient response. 1 (C) is correct oT on 9.1 96 tran ifborh Feet on ee Gompensotor Design Using Root Locus Plots fy (A) None of the above statements is true. {B) Statement (i) is true but statement (i) is fase. (©) Statement (i) is false but statement Gi) is tre. (D) Both the statements are true. 9.9 A long tail of small amplifade may appear in the output response with (A) Pl compensation (B) Lag compensation (C) Either of the two {D) Neither of the two 940 A closed-loop pole-zero dipole near the origin will appear with (A) PD compensation (B) PI compensation (C) Either of the two (D) Neither of the two "PP A process control system has type-D plant. For an accurate step-command following, a recom- ‘mended cascade compensation scheme for the system employs. (A) a log compensator {(B) a lead compensator (© either a lag or a Jead compensator (D) neither a lag nora lead compensator MASTER KEY ~ ; 91 © 92 (0) 93 (By 94 (ay 93. (B) 96 (Ay 97D) 93 (©) 99) 9.10 (B) 9.11 (A) ‘A umnity-feedback system has an open-loop transfer function, y= —* Ser DG+S) Draw the root locus plot and determine the value of K that yields a damping ratio of 0.3 for the dominant closed-loop poles. A phase-lag compensator having a transfer function, _ 1Os+1 1005-1 is now introduced in tandem. Find the new value of K that gives the same damping ratio for the dominant closed-loop poles. Compare the velocity-rror constant and settling time of the original and the compensated systems. MATLAB Exercise You have approximated the closed-loop system by a second-order system, assuming the dominance condition is satisfied. Generate root locus plots of the original and com- pensated systems in MATLAR window and determine all the closed-loop poles for the two sys- i tems, Also generate step responses in the MATLAB window and find settling time and steady-state error in the two cases. Consent whether dominiince conditiontvas satisfied. 2) 9.2 A unity-feedback system has forward path transfer function j 1.6/4 using lead compensator with Ieisrequited that he eosei-oop poes be leated ats, 123 ster Function D(s) = Determine the values of ccand XK to locate the closed-loop poles as required. WI of the third closed-loop pole? _ DIATLAB Exercise Attempt this problem using MATLAB dialogs. 9.3. Consider the feedback system of Fig. P9.3. Note that the plant is open-loop unstable {a) Plot the root loci to determine whether the system can be stabilized by proportional control, ie. Ds) = K. {b) not, could the unstable pole of G(s) be canceled by a zero of D(s) to stabilize the system, and if not, why not? is the location {) Choose a compensator Dis)=*E*2 and sketch root loci. How does this phase-lead $ compensator affect the transient as well as steady-state behavior? MATLAB Exercise Thepole of the physical plant a¢5) TT] He) ‘may be different from the one given by its model. DE) HH Gis) Cancelation of the pole of the model by compensa- st tor zero may not yield stable closed-loop when the Compensator Plant compensator iscascadedto the physical plant. Show this by generating root locus plots in MATLAB Fig. P93, window vith compensator zero at 5 = 2, and the plant pole varying between s= 1.5 and s = 2.5, 9.4 A unity-feedback servo system has open-loop transfer Function, (5) = 1000109) at SQ-+0.25+0.2557) A second-order cascade compensator S 408544 ae PO Gy 0.38a)(5 1042) cancels the complex poles of G(s). Using root-locus analysis, compare-the transient perfor- mance of compensated and uncompensated systems. What is the effect of D(s) on steady-state performance? : : 9.5. Consider the type-I system of Fig. P9.5. We would like to design the compensator D(s) to meet the Following specifications: (3) Damping ratio ¢ = 0.707 (ii) Settling time $2 sec (a) Show that the proportional control is not adequate. (b) Show that proportional plus derivative control will work. (c) For D(s) = Kp-+ Kps, find gains K, and K, to meet the design specifications. MATLAB Exercise After_solvin jem RIF Ye) y ereise Af 1g the prob O—fom 5 using hand sketches, attempt it using MATLAB. 5+) Determine settling time and peak overshoot of the L compensated and uncompensated systerns. Fig. P9S 98 Compensater Design Using Rect Locus Riots 6B sin 9.6 Determine the values of K, z, and p of the system(s) +, shown in Fig. P9.6, so that (3) the dominant ‘closed-loop poles have damping ratio ¢ = 0.45 and as (ii) velocity-error constant K,= 20. MATLAB Exercise Verify your solution using Tig P96 MATLAB, and if required refine your design. Due to uncertainties in the lond, the time-constant of the plant may vary by 108. Check the robustness of your design by determining transient and steady- ate performance for various values of this time-constant 9.7 The controlted plant of a unity-feedback system is rion a1, sand k oe (s+ I)(s+5) tt is desired to compensate the system so as to meet the fol Jowing transient response specifications: Settling time, t, $3 see Peak overshoot for step input $20 % Design a suitable cascade compensator (choose the compensator zero $0 as to cancel the pl pole at s=— 1), What is the velocity-error constant of the compensated system? lant MATLAB Exercise Attempt this problem using MATLAB dialogs. 9.8 The conifolled plant of a wnity-feedback system is K G(s)2 A eres “4 this desired to compensate the system so as to meet the following transient response specifications: ‘Damping ratio, {= 0.707 Settling time, 1, = 1.4 see ro Design a suitable first-order or second-order (two first-order sections in cascade) ead compensi- tor for the system. 9.9 The controlled plant of a unity-feedback system is 3.06 SF DG+2) verfor state Gs)= et the ‘Sketch a root locus plot of the system, and determine the domiinant closed-loop poles and the vyelocity-error constant. This desired to inerease the velocity-error constant to about 5 sec~’ without appreciably chang F ing the location of the dominant elosed-loop poles. Design @ firstorder cascade compensator Do)= or to meet this specification. ; MATLABExereise Attemplthisproblemusing (5) + 7 i MATLAB dialogs. “O-|D6i aos 9.10 Consider the type-0 system of Fig. P9.10. We ‘would like to design the compensator D(s) 10 ree! te following specifications: | B\_Contot Systems: Principles ond Design SS a (i) Damping ratio, £ = 0.6 Gi) Time constant ¢=1/f02, = 10.75 (ii). Zero steady-state error fora slep input. (2), Show that the proportional control is not adequate 9.5 (b) Show that proportional plus integral control will work K, (©) For D(s)= Kp-+5, find gains Kp and K, to meet the design specifications. i MATLAB Exercise After solving the problem using hand sketches, attempt it using MATLAB. Determine settling time and peak overshoot of the compensated and uncompensated systems. 1 9.11 The controlled plant of a unity-feedback system is : K G(s)=——*___ (= aT) 69 2 Determine the value of K so thai the damping ratio of the dominant closed-loop poles is 0.6, For this value of K, determine the velocity-ervor constant K,. 7 lis desired to increase K,, by a factor of 10. Itis also desired to keep the damping ratio of the dominant closed-loop poles at 0.6. A small change in the undamped notural frequency of the : dominant closed-loop poles is permissible. 9.16 (a) Design a suitable cascade lead, tag, oF lag-lead compensator to sealize the given objectives, (b) Design a suitable cascade PD, PI, or PID compensator to realize the given objectives } MATLAB Exercise Attempt this problem using MATLAB dialogs. ; 9.12 The controlled plant of a unity-feedback system is 5 K : G(si= = cI sts 10)° ; It js specified that velocity-error constant ofthe system be equal to 20, while the damping ratio of : “the dominant roots be 0.707. ft {a) Design a suitable cascade Jead, lag, oF Iag-lead compensator to realize the given objectives. r (0) Design a suitable cascade PD, Pl, or PID compensator to realize the given objectives. 7+ ’ MATLAB Exercise Attempt this problem using MATLAB dialogs. ‘ 9.13 Controlled plant ofa unity-feedback system is K G()=-* : On Teta) é It is desired to compensate the system so as to meet the following specifications: Damping ratio of dominant roots = 0.45 Undamped natural frequency of dominant roots = 3.5 rad/sec Velocity-error constant = 30 sec"! Design a cascade lag-lead compensator to meet these objectives (choose the zero of the lead section of the compensator to cancel the plant pole at s =~ 1). MATLAB Exercise Verify dnd refine your design using MATLAB, Determine settling time and peak overshoot of the compensated system. : 9.14 Controlled plant of a unity-feedback system is Gompensator Design Using oot Locus Plo: _ Bs a3. (@) Design a PD controller such that the dominant closed-loop poles are located at ‘What is the position-crror constant of the compensated system? (b) Itis desired to reduce the steady. state error to step inputs to zero, Design a PID controller that L meets the requirements on both the transient and the steady-state performance. |9.45.A unity-feedback system is characterized by the open-loop transfer Function eee O)= ea a NSD) LAB. (a) Determine the value of K if 20% overshoot to a step input is desired. bust (b) For the above value of K, determine the settling time and velocity-error constant K, (©) Design a suitable cascade lead, lag or lag-lead compensator that will give approximately 15% overshoot to a unit-step input, while the settling time is decreased by a factor of 2.5 with K, 2 20. (@) Design a suitable PD, PI, or PID compensator to realize the objectives given in (c) MATLAB Exercise You have employed second-order correlations to solve this problem using ofthe hand sketches. Verify your solution using MATLAB, and if required refine your design. of the 7 9.16 Fora plant with transfer function G(s) = the following specifications: et 1s 06. 3 «9 feedback system is to be designed to satisfy (@) Steady-state err fora ramp input $ 35% of input slope. (b) Damping ratio of dominant roots 2 0.707. = (©) Settling time of the system $3 sec. The structure of the control system is shown in =} Fig, P9.16. Select the parameters K and K, to meet the given specifications. MATLAB Exercise The parameter X, is subject fo Fig, P9.16 ‘variations by 10%6. Check by simulation the robustness Of the closed-loop system to these variations by determining setting time and K, for various values of K, Rosy e+, O77 Hs atioof ives. a feedback system is to be designed to K 9.17 For a plant with transfer function G(s)=———, OED satisfy the following specifications: (i) Dominant closed-loop poles at ~1 +2. (i) Ramp error constant, K, 21.5. The structure of the control system i8 pes) shown in Fig. P9.17 O re tend (a) Selectthe parameters.4,K,andK, ~ to meet the given specifications me and (6) Find the closed-loop transfer function of the system. Control Systems: Principles and Design 8.19 @=2.8; closed-loop poles:=I £j1.732,-7 8.20 There exists a circular root locus with center at (2, 0) and the radius equal to iz. & 9.1 Uncompensated system: K= 7.0 Ky=1:4,1,= 11.6 sec Compensated system: K = 60, K, = 12.0, (,= 12.7 see 92 a=5.3;K=23.2; third pole at s=—3.1 193. (c) System is stable for K> 19.2; steady 9.4 D(s) improves transient performance cons "95 (0) Kp= 8; Kp=2 9,9 -0334/0.58; K,=0.53 9.1) K=820; K,=41 e error to unit-ramp input = ~16/K. lerably; no effect on steady-state performance 9.44 (a) D(s)=3-+ 2s; K,=3 9.15 (a) K= 56 (b) 1, =3.6 see: K, 10.2 (2) Unstable (b) Stable 103 (i) Yes (i) No (iil) Yes 104 @) Two (b) Two (©) Stable (d) Stable (e) Nyquist criterion not applicable, (1) Stable 10.5. Gy Stable (i) Unstable 10.6 (a) Stable for K < 5/4 (b) Stable for all K (c) Unstable for tl {@) Stable for K < 2.65 10.7 K>0.75 108 K>6 4 109 K>2 10.10 (a) K= 0385 (b) Unstable for all values of K 10.11 t= 1.2 sec ee 10.12 «, = 6.2 rad/sec; @, = 15.9 rad/sec; DM =31.7°; GM = 5.5 10.13. GM= 0.7; @M = 10° 10.14 (a) 16rad/sec —(b) 102rad/sec—(c) 10 ra'sec—_(d) 26 rad/see 10115 (a) @,=7.86 rad/sec; 0: phase never reaches ~ 180°; DM = 51.8% Gif =< (b) «= 6.8 radlsec; (0, = 10 radlsec; DM =21.4°; GM = 6 dB ‘ad/sce; 0; Phase never reaches - 180°; PM= 73.4%; GI = ee radlsec; ©, = 13.65 radl/see; DM = 22°; GM = 4.2 dB (©) gain never reaches 0 dB; 2, = 7 rad/sec; DM =<; Gi (Dc, = 3.16 radlsec; «9g phase never reaches — 180°; OM. 10.16 (i) Stable system with M=21" and GM= 8 dB Gi) Unstable system with of = -30°, and GM =~12 dB 10.17 (a). K< 400 (b) K<55 (c) K< 1.67 10.18 (0) GM=12dB; PM=33° —(b) GM =2.5 dB; DM = 18° (c) t= 0.078 sec 109 @ AOts2) ® ——*¥ iw Sassen oso (1+ 8/10) G+ 5/2) (+10) 0+ 5/30) ‘s(1-+ 0.0255) 250 19.68" © © fs ‘s(0+ 0.45) (10.0258) (2s) 0+) (1+ 025) x a zc NYQUISTIBODE FREQUENCY RESPONSE PLOTS AND SYSTEM STABILITY Preview In Chapters Sand 8, w -ysis) to determine stability and to assess rel Shewwing how frequency response methods can be used 10 was used earlier to study relative stability of feedback control systems as a fo this objective in domain by using transfer function models of subsystems in ler re discussed stability and developed various tools (Routh criterion, root locus anal Jative stability. We continue that discussion in this chapter by westigate stability. The root locus method onetion of certain design parameters. We realized tl the Feedback loop, and root locus plots of closed-loop characteristic equation when a system param ig varied from zero to infinity. Here in this chapter, we use sinusoidal transfer function models and ‘Nygquist/Bode plots to study relative stability of feedback sysiems. INTRODUCTION A dynamie system is excited by an input, which is a function of time; the response of the system isa Furction of time; and the dynamic behavior (system response to excitation) is governed by physical laws that lend to differential-equation model ofthe system. Time" is, thus, the independent variable on which the signals and systerhs are dependent, Why did we use transformation from -damain to s-domain? Transformation is beneficial only if i helps in our understanding ofthe characteristics of signals and systems and/or understanding in transformed ‘domain leads to better analysis/design results involving these signals and systems. Laplace transformation. ‘as we have seen, involves turning differential equations into algebraic equations, thus greatly facilitating the analysis process. Being able to look at signals and systems in the s-domain helps us understand better hhow our systems really fmction; the stability, the transient and steady-state response characteristics are “understood easily from transfer function models, This understanding has given birth to one of the most important design tools—the root locus. We use Laplace transformation from f-comain to somain, obtain theanalysi/desiga results in s-domain cashing on better understanding, and then apply the inverse Laplace transform to the result to transform it back to the rcdomain, ‘ ‘We now turn our atlention to Fourier transformation, which is a process of decomposition ofa signal ino sinusoidal components of various Frequencies. A time-varying signal may be resolved into a sum ofsinusoidal fanctions 4 sin(c-+@) of different frequencies ay. Fourier transformation tells us what frequencies make up a signal (refer to Section 2.9). ‘What is the motivation for conducting control sysiems analysis and design To conduct a frequency-domain analysis of a linear control system does not imply that the system it only be subject to sinusoidal inputs. It may never be. In fact, it is the frequencs-selective Characteristics of signals in control systems that provide the motivation for transformation to in frequency domain? frequency domain. A control sys represerted by r sented by w, (no! signals. T nents ofthe inpw of design reduce In Chapter 9, eeaph poles ofthe syste cn visual We can shape » in control syste 2 sttitable con wshieh isa eihe ‘quency jqueney response inverse Fourier tr b ‘wansformation (rt ive design proce relationship (2.96 sransforin of sta Frequeney respi and apy the output (7, Ae ress aginst mod application of phy uncertain. and ca achieved. ¢ som response inodels in Section 10.8) 1 experimentally A plant tor v ‘practical view have we cone we control systems a Fit the industry. T system for the pl ‘om extaet the transfer fimetion The question: maturity of under present and the s ‘The present cb stability criterion tion of stability of terionalsagives stability eriterion oF ral hat tive 100 | yquistBode Frequency Response Plots and System Stabilly ge ‘A control system is required to process two kinds of input signals: command (reference) signals, represented by 7(0; and disturbances represented by w and x, (refer to Fig, 10.1). The signals cepre- cated by w, (noise), are generally of high frequencies, while rand w generally represent low-frequency Siunals. Therefore, a feedback control design is required to stop (atenvate) the high-frequency compo ents ofthe input and “selectively’ pass the low-frequency components. Ths, the alternate viewpoint ifdesign reduces the consrol-design problems to art appropriate design of a lowpass filter in Chapter 9, the shaping of the response (0) of a control system was dane by the root locus method, which is 9 graphical tool that manipulates the controller transfer function 2%) to adjust the closed-loop poles ofthe system (refer to Fig, 10.1) giving desired shape to the output y(). fn fiegueney davai ws eon viswalice the response (in terms ofits frequencies rather than an output itil changes with finn als ler. By using Feedback loop and We can shape (0) by manipulating its frequencies, Frequency-selective characteristics F's in control systems make this design approach simple and mosily by ‘a suitablé controller within the loop, we reshape the fre- quency response of the closed-loop system; and this fre- Dey {qnency response is translated back into time domain through inverse Fourier transformation, to see if he desired y() has At boon achieved. Complex computations of inverse Fourier 2O—pa}- [sal -O7- transformation (refer to Section 2.9), however, make this iter- oy iz ative design procedare difficult. Thre isa way out. Using the He relationship (2.99) between Fourier transform and Laplace (eee transform of stable transfer functions, we will convert the Wn frequency response model of the closed-loop system to Fig. 10.1. A Jeedhack control system s-domain and apply inverse Laplace transformation to study the output y(2) Another important acorn favor of design based on frequeney-reponse models she bat nse against model uncertainties. As we have seen in Chapter 3, transfer function moxlels abtsine! Fromm sptication of physical las are based on many approximating sssumptions. The modeling errors invade ‘uncertainties, and affect the robustness of the performance of the resulting control systems, The frequency- response models for stable plants can be obtained directly from experimental ‘measurements (discussed hater gp Sesto 10.8), therefore uncertaintes related to modeling errors will be minimized, The design bfS&l on experimentally obtained frequency-responsé models is expected to be relatively more robust ’A plant for which the feedback system is to be designed may, however, be an unstable plant [A practical viewpoint in favor of ignoring this kind of limitation is that: hony many unstable plants ‘have we eome across in process industry’onnfacturing floors? Since most ofthe plants in industfial ‘control systems are stable, the robustness aspect of frequency response based design continues to bene- fr the industry Theoretically, however, we may argue that fa plant is unstable, create a stable feedback system for the plant, make the frequency response measurements on this feedback system. and there from extract the frequency response data for the plant using the relationship between the closed-loop transfer function M(o) and the plant transfer GU). “The question: why frequency-domain design?, will get naturally answered in a better way as our maturity of understanding the control system characteristics in Frequency domain increases through the present and the subsequent two chapters “The present chapter is concerned withthe sablity analysis in Frequency domain, A frequency-domain stability eiterion was developed by H Nyquist in 1932 and remains a fundamental approach to the fnvestign- tion of eabiliy of linear contol systems. In addition to answering the question of absolute stability, this terion also gives some useful resll on relative stability. The relative stability measures given by the Nysqus stability eriterion are, in fat, central tothe great importance of frequency-domain design methods, Geel Systeme: Prnepns and Deson \We first develop the Nyquist stability criterion, TWwo common types of frequency-domain plots (equi ploss, Bode plots) are then introduced, and using these plots, absolute and relative stabil ity ate investigated on the basis ofthe Nyquist stability erterion. ‘The use of the Nyquist eriterion in frequency-domain design procedures will appear in the next two chapters. i a DEVELOPMENT OF THE NYQUIST pee STABILITY CRITERION In order to investigate the stability of a control system, we consider the closed-loop transfer Function, Ye) Rs) 1+ G(s) (s) (0.1) keeping in mind that the transfer fonctions of both the single-loop and the multiple-loop contra! sys ene can be expressed in this form, The cheracterstic equation of the closed-loop system is obtained by setting the denominator of YsV(s) o zero, whiel is same as setting thé numerator of | + GUS)H(8) to zero. Thus, the roots of the characteristic equation must satisfy 1+ GS)HEs) =0 (10.2) {We assume at this point that Gls)f/(s) can be expressed by « ratio of finite algebraic polynomials ins “This assumption has been made for convenience; the Nyquist criterion applies to more general situa- tions, as shown subsequently, Let Kis aioe Ei a Net enh msn (103) (SF AMS Pe) (OF Py) G(spHs)= Gis)H{s) isthe product of plant, compensator, and sensor transfer functions; its pole and zero locations due assuned to be known since these transfer functions are generally available in factored form. Substi- tuiing for G(s)A(s) from Eqn. (10.3) into Eqo. (10.2), we obtain K(stai(st2y)- (+26) (5 AMS P2) (5+ Pa) (E+ BU Py) (0 py) + K(s Sst) (OE) (s+ Pys+ Pa) (8 Pad 14 G(s)H(s)= 1+ (s+ 28+ 22)- (s+z,) =o 4 (st ae) SPs) tee Itis apparent from Eqn, (10:) thatthe poles of 1+ G(s)H{s) are identical to those of G()I, jen the open-loop poles ofthe system; and the zeros of 1 + G(s)H(s) are identical to the roots of the characteristic tqnation, Lethe closed-loop poles of the system, For the elased-loop systems to be siable, the zeros of 1 + G(eI(e) mms lie in the left hal ofthe plane. Itisimportat to note that even if some ofthe open-Je0p poles lie in the right half-plane, al the zero of 1 + G{s)F7), i. the closed-loop poles may le in the Teft hulfs-plane, meaning thereby thet an open-loop unstable system mnay lead to closed:-togp stable operation ‘Reval! that we have introduced two methods of checking whether or not all zefos of 1 + G(s)H(8) have negative real parts. The zeros of I + G(s)H{(s) are roots of the polynomial (8 pls py) (St p+ Ks +2(5+ 24) (+z; and » asafi zeros, terior used Ce Tt point the p by th N cont! NyquistlBiode Frequency Response ‘Another method is to plot the root loci yo (s4 “Gr pM Pa) Pe) ‘and we may apply the Routh test asa funetion of K. In this section, we shall introduce yet another method coros of 1+ G(s)H(s) Si terion, is based on the prin ‘used in the Nyquist exite ‘Consider a rational function, ciple of argument in the theory of complex ¥ on is explained below. sy Le G(s Mts) = OS SP, “The pole-zero map of Q{s) is shown in Fig, 10-28 ‘This figare also point or an arca is said to be enclosed by a the path is traversed in the clockwise direction. The pole -P1 by the contour I} in Fig. 10.20. and the 2% s-plane @ cassineguesuat uaa lagasanunris ei i cide the open left hall s-plane. The methot, called the jatiables [32]. The basie concept Josed path if it is found to lie to t rots and Systom Stability of of checking whether or not all Nyquist stability cri= ssa closed contour Fy. A fhe riyht of the path when are therefore enclosed show! ero (py lane of F, we have 10.2 (a) A contour which encloses both the pole ant the 2-70 4) {(b) A contour which encloses only the pole ‘Now we see that os the point s follows ths preseribed path (i.e clockwise direction) an the s-Pk “ contour Ij, both Z(s-+2,) and 2(s + py) decrease continuously. For one clockwise traversal ic of 5, L(s4 52-28 op : 526s p= 2 where 6,2 indicates the change innate 28 Ty '8 traversed, 5, £018) = 6p, Zeb a)~ SLO Ps =—2n-(-2m)=0 ) @ eee ‘Thus. the change in ZO(s) ass traverses 17 (which is a completely general contour in the s-plane ‘excep! that it encloses both ~2, and ~p,) once in clockwise direction, is zero. Using these arguments on the contour Fin the s-plane which encloses only ~p, and not ~2; (Fig. 10.2b), we find that (for the _ not enclosed by the contour 3, 6;, 2(s+2,)=0 for one clockwise traversal of >) zeros 5,,Z0(s)=0-(-28) = 2 We thus sce that a change in ZO{s) that one s-plane sing a contour in the s-plane, is Md zeros gets in ti siricily a function of how many pole: of Gls) are enclosed by the s-plane contour, Let vs consider 4 yoneral O(s) = 1+ GUA), given by Eqn. (10.4). for which » typical pole-zero plot might be made as shown in Fig. 10.3. The figure also shows an s-plane contour J; which encloses ‘os and P poles of (s). Note that F does not through any of the poles or zeros of Q(s). If ke one clockwise traversal of the contour may be seen that ar x -P, 5p, LO(s)= 2-20) = P(-2m) =(P-2D2n (10.5) Fig 10.3 A general pole-zero plot with contour I, Another way of looking at the result of Eqn. which encloses Z zeros and P poles (103) is to consider Os) given by Egn, (104) ‘evalated on the «plane contour F; of Fig. 10.3 and plotted in the Q(s)-plane. For every point 7 + Joon thes plane contour [7 we obtain Q{s) = ReQ+ lO. Alternatively, it ean be stated thatthe function Qts) pigps the pain jain the s-plane into the point ReQ + jlmQ in the. Ots)-plane. It follows that for the-cfosed contour [yin the s-plane, there corresponds a closed contour Tp in the Ofs)-plane. A typical pis shown in fig. 1DALTheatrowheads on this con ace tourindi¢ate the direction that Qfs) takes as s moves on J, in the clockwise direction. Now as we traverse Ip onee inthe py direction indicated by arrowheads, contour J; im the s-plane A, jg traversed once in the clockwise direction. For the example y of Fig. 10.4, traversing [ once in the clockwise direction gives a change in £Q(s) of ~ 4zr, since the origin is encircled twice in the negative direction. Thus, Eqn. (10.5) gives ImQ Q-plane 4s) 8, 20s) = 2n(P ~ 2) =e ReQ or Z-P=2 In general, Z-P=N (10.6) Where A is the number of clockwise encirctemenis of the Fig, 10.4 ((5) eonluated on the contour origin that Fy makes in the Q{s}plane, Tt may be noted Trin Fig. 103 wee poin ofa #1 te 1+ we d of th cont ne on he ne + cS that we are not interested in the exact shape of the O(s}-plane contour. The important fact that concerns uss the encirclements ofthe origin by the Q(s)plane contour ~~ The relation (10:67 Between the enclosure of poles and zeros of O(s) by the plane contour and the encirclements of the origin by the O(s)- plane contour is commonly known as the princf- ple of argument which may be slated as follows, ‘Let Q(s) be a ratio of polynomials in s: Let P be the number of poles and Z be the number of zeros of Q(s) which are enclosed by a simple closed contour in the s-plane, Fig: 105 Evaluating 1+ G(s)H(s) from the GE)ES) map multiplicity accounted for. Let the closed Contour be such that it does not pass through any poles or zeros of O(s). The s-plene contour then maps into the Q(s)-plane contour as a closed curve. “The number W of elockwise encirclements of the origin of the O(s)-plane, as a representative point s traces out the entire contour in the s-plane in the clockwise direction, is equal to Z-P. L—» Re io @ Since Olsy=1 + GH). #L to each we can obtain O(s)-plane contour from the G{s)H(s)-plane contour simply by adding poist of Gls)H(s)-plane contour, Figure 10.5 shows atypical G(si(s)-plane contout Foie The effect eradding +1 to each point of [ey to obtain Q(s)-plane contour [7 is accomplished simply by PH io the scale of the real axis, a8 shown by the numbers in parentheses in Fig. 10.5. If = ajo point of the Gis)H(s) map Tey, corresponds to the origin of the Q(s) map Tiy For convenience, wve tesignate the ~] + j0 point of the G(s)M{s}-plane as the evitical point. Thus, the eneirclements ofthe origin by the contour [is equivalent 10 the encirclemients of the critical point ~| +0 by the contour Jy. In the light of these observations, we can express Eqn. (10.6) as follows: f =N (10.7) Z- where sumber of zeros of t + G(s)H{s) enctosed by the s-plane contour I-3 mmber of poles of G(s)H(s) enclosed by the s-plane contour Ij} and "umber of clockwise encirclements of the critical point ~1 + j0 made by the G(s)H(s)-plane contour Tarp : In general N can be positive (Z > P), zero ( encirclements of the critical point ~1 +0 in clockwise zero nef encirclements of the critical point by the Fey contour. of the critical point in counterclockwise direction by the I, contour. “A convenient way of determining N with respect ta the critical point ~| +j0 of the G(s)F(s)-plane is to draw a radial line from this point. The number of net intersections of the railial line with the Ty, contour gives the magnitnde of N. Figure 10.6 gives several examples of the method of = P), or negative (Z 0 corresponds to N nét direction by the Fi, contour. N= 0 indicates ‘<0 corresponds to N net encirclements determining N. CRP__Coniros Systems. riricapwus cine Soro” oe GHeplane Im Gti-plane | ™ GH-plane | Fig. 10.6 Examples of determination of N 10.2:1.|) The Nyquist Contour _ [Atthis point the reader may place himselfTherself in the position of Nyquist many years ago, confronted ‘vith the problem of stability of the closed-loop system that has the transfer function of Eqn. (10.1), ‘hich ig equivalent to determining whether or not the function 1 + Gte)FAsyhas-zeros-in the right half plane. Apparently, Nyquist discovered that the principle of argument of the eomplex-variable theory could be applied to solve the stability problem if the s-plane contour J; is taken to be one that encloses the entire right half of the s-plane. Of course, as an alternative, I; can be chosen to enclose the entire Jeff half of the s-plane, as the solution is a relative one. Figure 10.7 illustrates a [ contour that encloses the entire right half of the s-plane. Such a contour is called the Nyquist contour. Its directed clockwise and comprises of an infinite line segment C, along the j>-axis and are C, of infinite radius. ‘Along C,, 5 =jtwith » varying from — jo to + je. gos), Along Cas = Re with > oad @ varying from +90° through 0° to -90°. SPP As the s-plane contour FT, must avold al poles AC Ge the Nyquist contour defined in Fig, 10.7 are required-wher G(s)M(s), and therefore 1+ G(s)H s-oné or more poles on the imaginary axis (the zeros of 1 + G(s)#%(s) are unknown to us; we will shortly see the implications of splane contour passing through the zeros of 1 + G(s) H{G)). The basic trick, of course, is to take a smal detour around the imaginary axis poles. Figure 10.8 illustrates a modified Nyquist contour when. Cloy ajc Fig. 10.7 The Nyquist contour Fig. 10.8 Indentted Nyquist contour Gls)Hi(o) hos a pole at s= 0. Along the semicircular indent around the pole atthe origin. » = pe! with 1p > O and 9 varying from —90" through 0° to 490°. The indented Nyquist contour in Fig, 10.8 does fot enclose the pole atthe origin. OF course, as an alternative, the Nyquist contour may be indented 10 enclose the pole at the origin 4022|_The Nyau “The Nyquist stability criterion is a direct a contour J;is the Nyquist contour. In principl 2 closed-loop system can be determined by plottin tthe Nyquist contour, and investigating the behavior of the G(s)A(s) plot with respec point —1 +0. The Gls)#I() plot that corresponds to the Nyquist contour is called the Nyquist plor of GOH). With the additional dimension added to the stability problem, we define WV, F and Z as Follows: N= number of cloctovise encirclements of the critical point~| +j0 made by the G(s\H(s) locus of the Nyquist plot. rumber of poles of G(s)H(s), and therefore of | + G(s)H(s), enclosed by the Nyquist contour. Then (refer to Eqn. (10.7) Z= number of zeros of | + Gls)H{s) enclosed by the Npguist contour i =N+P (10.82) pplication of the principle of argument when the s-plane Ie, once the Nyquist contour is specified, the stability of 1g the G(s)F(s) locus when s takes on values along to the critical Closed-loop stability requires that ro (10.86) Qy_Lontrol Systeme: Prinipos and Design S This condition is met if Ne (10.80) Jn the special ease (his is generally the ease in most single-loop practical systems) of P= 0 (e.. ie open-loop transfer function G(s)Hs) has no poles inthe right half ofthe plane) the ‘closed-loop em is stable if N=0 (10.80) 10.2.3 | The Nyquist Criterio} We can now slate the Nyquist stability criterion as follows: ifthe Nyquist plot of the open-loop transfer function G(s)H(s) corresponding to the Nyquist contour in the seplane encircles the critical point ~1 +0 in the counterclockwise direction as many times ‘asthe number of right half's-plane poles of G(s)H(), the closed-loop system is stable inthe commonly oecurring case of G(s)HG) with no poles in the right halfof the s-plane, the closed- Joop system is stable if the Nyquist plot of G(s)H() does not encirele the —| + 0 point When the Nyquist plot of G(s)lls) passes through ~1 +0 point, the mumber of encirclements N is indeterminate. This curresponds to the condition where \ + G(s\H{s) kas zeros on the imaginary axis. i necessary condition for applying the Nyquist criterion is that the Nyquist contour inust not pass through any poles or zeros of \-+ G(s)EIs). When this condition is violated, the value for N becomes indeterminate and the Nyquist stability criterion cannot be applied. 10.3. SE LECTED ILLUSTRATIVE NYQUIST PLOTS, “The following examples serve to illustrate the application of the Nyquist criterion tothe stability study of control systems. Consider a single-loop leedback control system with the open-loop transfer funetion given by i K SOHO Freee lyk? 0.K>0 ao.) GleyH(s) has no poles inthe right-halfs-plane; therefore, stability is assured if thé Nyquist plot oF G{s)H{s) does not encircle the ~1 +0 point. “The Nyquist plot of G(s)H(6) a5 ve know, isthe mapping ofthe Nyquist contour in the s-plane onto Geotts)-plane. In Fig. 10.9 the Nyquist contour has been divided into three sections: C,, C2, and Cy Section C; is defined by 5 =j'@, 0 <=; section C, is defined by s=jo,—e < @S 0; and section Cy isdlefined by s= Re!®, R — c0, and 8 varies from +90° through 0° to -90°. Mapping of Section C, onto G(S)H(s)-plane Substituting s = jer into G(s)HG), we obtain (10.10) GL joy jo)= (jen + jor, +) NyquisttBode Frequency Response Piots and System Stabiliy Im GHt-plane | GHplane ed Fig. 10.9 (a) Nyquist contour; (0) Nyquist plot of GUS)H(S) of Equ. (10.9) () Smapifcd Nyquist plot vis fee xis, | A plot of G(j@)1/(/o) on polar coordinates as «is varied from 0 10 ois the mep of section ©) on the ass] Gisifi(s-plane. This plo is called the polar plot of sinusoidal transfer netion! GL Jonri Jo), nes For application of the Nyquist stability criterion, an exact polar plot of G{ jojo is not essential Often a rough sketch is adequate for stability analysis. The general shape ofthe polar plot of Gl jeovd( jor 4] may be determined from the following information: a 4 1. The behavior of the magnitude and phase of G( ja)H( jo) at o= O-and 2. The points of intersection of the polar plot with the real and imaginary axes, and the values of @ ody at these intersections | 3, A few on-locus points in second and third quadrants in the vicinity ofthe critical point ~1 +/0. sfer | For G(jeaH{ jaa) given by Eqn. (10.10), the magnitude and phase at w= 0 and w= are goleulated {as follows: : : [GLa)HCJ0)) geo KZUO)HUO, = 0.9) + : K leo Vo)... =a] 0 Porye, tof 2 rane imi K oT fered ( 2atseiio. sa a0 \ 1G, \._- oO \ nc “The intersections of the polar plot with the axes of G(s)F(s)-plane ean easily be ascertained by iden- J siying te ret and imaginary pars of GCsadH( Jo). | "This plot also gives the open-loop frequency response of ie system if all he potes of Glas Tie in the lef-hall ‘plans (refer to Eqns. @294-2.99)), The frequency response of stable oper-logp systems obmained experimentally 7 (Geter to Section 10.8) can also be used! in the stability analysis using Nyquist criterion. For Glos) that is not 2.10) Sie, tbe sinosoidal transfer function G{Jo)FI jo) represents a map of a sction ofthe Nyquist contour on GIs)A(s} Planes and using this map, the stability ofthe resulting closed-loop system can be determined using Nyquist criterion Control Systems: Principles and Design _ with & coniot For Gv inFig. There Gio) Hijo) = Som (14 jor + fot, )| I~ jar yl K[d-o'nt,)- jor, +1)] yf; Koxt,-+t) sorta ae) - ims Oddrataerayn oP = Re[G( jap joa)] + fim[G( joyH( joy] When we let in[G(jo)H{ je] to zero, we get = 0, meaning that the polar plot intersects the real axis only at «2= 0. Similarly, the intersection of G{jo}EI{jo)-plot with the imaginary axis is found by setting Re[Gt joy ja)] to zero, which gives w Ti, Koln +t,) JeGeynGo| eRe This s Based on this information, a rough sketch of the polar plot ean easily be made as shown in Fig. 10.9b (ihe portion of the lacus from = 0 to @=+=). Mapping of Section C, onto Gis)H(s)-plane Given Gah jo), 0S © <2, GLJO)EL J) for ~ © < wS 0 is constructed by realizing GE JOH Ja) = [GGOHUOT Where * denotes conjugate. Thus, given the G(ja)H{ja)-locus for 0 < @ < e, we get the G(jo) Hleo}-locus for the negative jo-axis by simply taking the mirror image about the seal axis of the locus for positive ©. The dashed portion of the locus in Fig, 10.9b from (= —se to @= 0 is thus the map of section C, of Nyquist contour. Mapping of Section C, onto G()H()-plane ‘Along section C,, s = Re’; R > eo and 0 varies from +90" through 0° to -90°. Substituting into GANS), we obtain GUNH(S)], go sr locus around the ogi (fer to Fig, 10.9b), represented by The infinitesimal semicit Gitte) 26 nyquistiBode Frequency Rosponse Pots and System Staikty fy (0% is thus the map of section C, of Nyquist with © and crvarying from ~180° through O° 10 +18 contour. For application of the Nyquist stability criterion, G( ils} plane is of no interest to us. We ean therefore tke Fi}g. 109e. This plot does not encircle the eal point =I + For any positive values of K. ‘Therefore, the system is stable forall positive values OF K, Fy, and F exact shape of the Nyquist plot around the origin of 0; the resulting Nyqaist plot is shown nd Consider now a feedback system whose oper-loop transfer function is given by pene? peo th GH x >0,150 s(is+1) aoa L We note here that for this type-I system, there is on ‘The Nyquist contour is shown in Fig. 10.10a, where a ‘The Nyquist contour has been divided jnto Four sections: Cy, Ca, Cs and Cx Section C; is defined by $200, 0< #0<07 section Cp is defined by s=Jan-—ee < @-< 0; section Cy isdefined by s= Re! R—> sind 0 varies from + 90° through 0° to ~ 90°, and section Cis defined by s = pel? p— 0, and @ varies from —90° through 0° to #90°. ‘Mopping of section C, onto the Gis)H(s ne pole on the imaginary axis, precisely atthe origin. snicircular detour about the origin is indicated >-plane is given by the potar plot of sinusoidal transfer function. (0.12) K GL joy jo) = LQHGO* ToC ei) ‘This sinusoidal transfer function can be expressed as k K G( joy jo)=-—5 sea JOHGO= ee ol oe In FO} s-ptane woo TS Git-plane fa) ib) 5 Fig. 10.10 (a) Nyguist contour; () Nyppist plot of GO)HE) of Eqn. (10.1) {2_Contrl Systoms: Principles and Design Fhe low-frequency portion of the polar plot becomes lim Gj Jo) = to joo frequency portion is (refer to Eqn. (10.12)) and the 02-180" “The yeneral shape of the polar plot i shown in Fig, 10.10b. The G{ Ja jo)-plot is asymptotic to the vertical line passing through the point ~ Ke-+0. Mapping of section C, onto Gls)#}-plane is obtained simply by taking the mirror image about the real axis of the polar plot of Gt jayH( jo). ‘Mapping of section C; onto the G{s)4(s)-plane is abtained as Follows: K GlssH(s)) Fay sone! 1m 180° through 0° to +180°. The Gls)M(s)-locus thus turns at the origin with zero radins fro pe, p30, Mapping of section C, onto the G(s)H(s)-plane is obtained as follows. Along Coy + and ¢ saries from ~90° through 0° to +90" Substituting into G(spFi(s), we obtain K K ow aoe pePeapel® +1) P Gino) ~ (a Kip approaches infinity as p-> 0, and —p varies from 490° through 0° to Pa ‘yquist Gontour. Thos, the infinitesimal-semicircular indent around the moves along seption C, of the Ns viinin theseplane maps ino a semicircular ae of infinite radius on the G(s)H{(s>plane as shown in Fig. 10.10b. rhe complete Nyquist plot For G(s)fs) given by qn. (10.11) is shoven ip Fig. 10.108, tn order to investigate the stability ofthis second-order system, we fist note thatthe mumber of poles of GSH) ih theright-halfs-plane is zero. Therefore, for this systern to be stable, we require thot the Nyquist plot UrGEtits) does not encircle the eritical point -1+0. Examining Fig. 10.10b, we find thot resp arth alu ofthe gain K and the time-constant the Nyquist plot does not encircle the critical point, and the system is always stable. xamplét We now consider a type-2 system with open-loop transfer Function, G(s)Hls)= K>0,7>0 (10.13) S(t) i. 10.109. The Nyquist plot is shown in Fig. 10.11. Clearly, the The Nyquist contour is shown in F + enis simply the polar plotof portion of the Gls)H(s)-locis From «= OF to « 0.14) Gl joy jor= 3-3 Par(jortl Note t ht ca Ga ais; it Hie ‘The ¢ radius fr responds portion « Oristhe for posit and g gives GSH The w=" #180%2 +90 Fror clockw from E the sys 0, he in to %) lot ve at, yqustftode Frequency Response Pls and System Slatiily Note that How ony cust (4 Jct emzsvar nat LO het ‘ p K C ee TD Gl iT =|— 2 - 270° = PZ-270° o or] | Gy Geto) (4) anor -b or] 1 can easily be examined that the locus of Gieyti( je) does not intersect the weal oF imaginary ‘pus; it Hes in the second quadrant forall values of «& The G()H(s)-locus tums at the origin with zero radius from -270° through 0° to +270°. This cor- responds 16 the locus from w= 42 to @= ~22. The portion of the G(s)#(s)Jocus from @= == fo @= 0" isthe mirror image about the real axis ofthe locus for positive «. From « ,8= pe, p> O and @ varies from —90° through 0°, to +90". gives K K aaa sage ae” ene ae pre? (ape!? +) Pp Fig. 10-11 Nyquist plot for GUS) of ‘The portion of the G(s)M(6) locus from @= 0" to Ein, (10.13) OF isa cireular arc of infinite radius ranging from “£180? at «= 0" through 0° to —180° at a= 0" (H180° 5 +90° = 0° + ~ 90° —> -180, From Fig. 10.11, we observe thatthe Nyquist plot encircles the critical point “1” j twice in the clockwise diection, ie, N= 2. Since P= numberof poles of G(s)H() in ight-halh plan from Eqn, (10.8a), 225 4, there ae two roots ofthe closed-loop system in the right-half the system, irrespective of the gain K and time constant 7, is unstable stl OH) = (10.19 Let us determine whether the system is stable when the feedback path is closed. 0.05 ai os 600 98.51 357 inna 3 8 isha sol} 4 cw i (. yo ete ped? 5 2 of in #FR_Contro! Systems: Principles and Design @ From the transfer function of the open-loop system im itis observed that there is one open-loop pole in the Gtplane right half of the s-plane. Therefore, P= | “The Nyquist contour is shown in Fig. 10.1 a The Nyquist plot is shown in Fig, 10.12. Clearly, the por- tion of the G{syFZls)-locus from @= 0" to @= + is simply the polar plot of - Gey jo= 2 — 0.16) Fa'(jo-2 Te following points were used to construct the polar plot low POL” Fig. 10.12 Nyqu 500° Eqn. (10.15) Gjoyi(jo, ist plot for G(S)H(S) of Gov (jo), Gl joy ja), = 0.62Z71GJO)H Jor, 49 = 0.012162" G FOV For|y_,,. PETTY 60 “The plot of G(ja)H{ je) for «2< 0 isthe reflection, with respect to the real axis, oF the plot for @> 0. {Every point om the large semicircle in Fig. 10.19a, is mapped by G(s)A(s) into the origin ofthe G(s)H(s)- plans. A point s = pe!*: p> 0, 6 varying from 90" through O° to +90" is mapped by G(sYH(s) into } cco) = 2(180°- 29) “The small semicircle in Fig. 10.10a is mapped! by G(s) (9) into a cireular are of infinite eadius ranging from 360" at = 0° through 180° to 0° at c= 0" (360° —> 270° > 180? — 90° > 0°). Figure 10.12 indicates that the critical point ~1+-j0 is encircled by the Nyquist plot once in the clock- wise direction. Therefore, N= I, and (refer to Eqn. (10.82) Z= number of zeros of 1+ G(s)/i{s) enclosed by the Nyquist contour N+P=2 Hence, the feedback system is unstable with two poles in the right half of the s-plane. Consider a feedback system with the following open-loop transfer function, eee (10.17) s(s+3)(s+5) G(SNH(s} Letus investigate the stability of this system for various values of K. E Ht Fies set K = 1 and sketch the Nyquist plot for the system using the contour shown jn Fig, 10.10a, For all points on the imaginary axis, At Ne parto yields Finall or 1 Inor the € w D ‘or ih oan eu oleae Nyquisti@ode Frequency Response Plots and System Siabilty a (10.18) o 1S@~« sw? (joy jay= > —=} isa COHUO= Terk Glo" +r Al c= 0, G (jeo)H { jee) = ~ 0.0356 ~ jes Next, find the point where the Nyquist plot intersects the negative real axis, Setting {'5. Substituting this value of eoback into Eqn. (10.18), the imaginary part of Egn. (10.18) equal to zero, we find = yields the real part of ~0.0083. Finally, at =<, GJOH (Jo) = ls) (6), “The Nyquist plot for G(s)His) of Eqn. (10.17) im Gtt-plone with K = 1 is shown in Fig. 10.13. Application of the Nyquist criterion shows that with K = 1, the closed-loop system is stable. If we were to increase the gain by a factor 1/0.0083 = 120.48, all points on the G(ja)H(jo)-locus would increase in magnitude by this factor along radial lines centered at the origin. The length of the vector G(jVI5)H(jV15) would be 0.0083 (1/0,0083) = 1, and, therefore, the curve G(jo)H(jo) would go through the critical point: the closed-loop system would be at the limit of instability. Hence, for stability, K must be less than 120.48, i.c., the stability range is O ‘O and 6 varies from ~ 90° through 0° fo +90" Gia) == Kt eci8r-0 Leeper pe PO ayo, tH) ‘Therefore, the semicircular detour in the Nyquist contour is mapped into a semicircle of infinite radins in the left half of Gts)-plane, as 0.15. When —K,K,<~ Lor, K,>1 140 point once shown in Fi the Nyguist plot encircles the jn counterclockwice direction. Thus, the system is stable whdn K, In the preceding sections, the Nyquist plot was used course, in general, a usable system must be st + fortwo reasons. First, a the model may indicate that a system is stl sve require not only that a Measures of degree 0! no poles in right-half s-plane, STABILITY MARGINS J to determine if a system was stable or unstable, OF able, However, there are concerns beyond simple stability, table system must also have, among other characteristics, response. Also, the mode! that is used in the analysis and design © le, whereas infact the physical system isunstable. Generally system be stable but also that it f stability of a closed-loop system that has open-loo ‘can be conveniently created throug! fom for such a system becomes obvious by the inspection of the Ny Fig. 10.14 A feelback control system ~~ J a0 ig. 10S Nyquist plot for Gts) of Eqn. (10.19) ‘an acceptable transient ‘fa control system is never exact, Hence, be stable by some margin of safety p transfer function with Jb Nyquist plot. The stability informa- /quist plot ofits open-loop transfer "9.05 “0.09. 1 1 s116 223 gai Sead 153.56 fanct Itear to 10.1 pole: trans of th ase the j T poin large tose plot desi any 10.19) — ste, Of wily, snsient ence, verally, a with forma- | ransfer Nyquistfiode Frequency Response Plots and System Stability 9% fanction G(s)H(), since the stability criterion is merely the non-encirclement of the critical point =1+0 Itean be intuitively imagined that as the Nyquist plot gets closer to the critical point, the system fends towards instability ‘Consider two different systems whose dominant closed-loop poles are shown on the s-plane in Figs 10.16s and 10.16b, Obviously, system 1 is ‘more stable’ than system B, sinee its deminant closed-loop ‘away to the left from the ja-axis. The Nyquist plots of open-loop rr nsfor functions of systems 4 and B are shown in Figs 10.16c and 10.16d, respectively. Cormparison air the closed-loop pole locations of these systems with their corresponding Nyquist plots reveals that sea Nyquist plot moves closer tothe eriical point -1+0, the system closed-loop poles move closer 19 the jeo-axis and hence the system becomes relatively less stable and vice verse. Thus, the distance between the Nyquist plot af open-loop transfer function Gi wis 0 can be used as a measure of degree of stability of the closed-loop system. Generally, the the more robust the system=reler poles are located comparatively {s)H(s) and the eritical point Jarger the distance, the more stable the system (the larger the distance toaubsection 11.33 (Fig 11.6)). The distance ean be found by drawing a cele touching the Nyquist 10. 16e and 10.16d, Such a distance, however, is not convenient for analysis and tify the closeness of plot as shown ia Fi design. The measures gain margin and phase margin, are commonly used to qu a Nyquist plot to the critical point system fim jo / | GH-plane plane ® ! @ Fig. 10.16 Correlation between the closed-loop plane root locations andl open-loop _frequency-response curves conta systems Principles and Design 10.4.1 | Gain M Figure 10.37 shows a “typical” Nyquist plot of Gl janHi( Js oO G(s}H(s) is known to have no poles snhe right half of the plane. A closed-loop system sth the GLS)FH(s) 2s ‘open-loop transfer Function, is stable, “The point at which the Nyquist plot crosses the negative real a a that point is designated ©, The point 4 is thus the tip of the vector jeo.)H{ jo,) vith [G{jo)H(o)\and angle 180°. Its seen from Fig, 10.17 that |G jeo,yH(aay|< |, and therefore the tip of the vector OL jea,)H(j@s) does not reach the critical point =14 9 Tre vee to inerease the gain of G{jto)H(je) by a factor Wj0al without altering is phase, all points cn the Nyquist plot of Gl jo)H{( jo) would increase in magni by this focto, along radial lines aoe arth origin. The length ofthe vector GL jap Ht ja) would be OAL MOA = 1 ‘and therefore the curve G( jo} jo) would go through the critical point; the closed-loop system would be at the fimit ratabaity, We call the multiplying factor 19a the gain margin because its the factor by which the » drive the system to the verge of instability. marked 1, and the frequency nde gain can be increased to 4Im GH-plane Gl feng HH jeag) =1 470 eat is eee eS ee 0 YP Re a4 GLiayHGey | oOoM ‘ ajay je o70 Fig. 10.17 Typical Nyquist plot of GodH (je) From the graphical description of Fig. 10.17, we now make the following definitions: Phase Crossover Point Thisa point on the G(s)M{6)-plane at which the Nyquist GUa)HYeyplot intersects ihe negative real axis. Phase Crossover Frequency Os Its the frequency at the phase crossover point, or where ZG jea)H jon) = -180° - (0.20) Goin margin ‘The gain margin GM of the closed-loop system that has GIs)H(s) as iis open-loop transfer function is defined as the number | | Fors bya fav numbe, nA Avunit Nyquis The po G6 critical rw G(joo. phase I the crit the ang vector be dees Fro Gain Itisay Cain tis th: Phase transfe For associ: is aber Cor (G(s) gain n Gis Gio vector theret “2 les on, ay ade tip nts nes ore snit the xis. 921) Joop ie cae psoas tices ues yauistiBode Frequency Response Plots end Systom Stabily _ Bs ou= > (jog HC j5)] 00.22) 2: the loop gain magnitude could be increased For example, if}G(jo,)H(o,)| = 0.9, then the Ga bya factor of 2 before the closed-loop system woud go unstable. For stable systems, GAL is always a number greater than one. “Avnit circle centered at the origin has been drawn in Fig, 10.17 in order to identify point B at whick Nyquist Gl jeyH{ ja)plot has unity magnitude; the frequency ot point B has been desienated TThe point B is thus the tip of the vector GLjey,)H{ ja) with magnitude \GCjo,)H{ ja] = | and angle ZG ja, H(Ja,). Its seen from Fig. 10.17 that the tip of the veetor G(ja,)HC je) does not reach the critical point ~1+ 0. Ifwe were to decrease the phase of G(o)#i(j«) without altering its gain, all points on the Nyquist GUa@)H{ jo)-plot would rotate clockwise about the origin an angular amount equal to the increase in phase lag. In the light of Fig. 10.17, we see thatthe tip of the vector G(ja,)H(j@,) would be placed at the ctitical point if the increase in phase lag of Gt ja)H{( joo) were made equal to the angle @M. We call the angle PM, measured positively in the counterclockwise direction from the negative real axis to the vector Gl jc, H{ Je), the phase margin because itis the angle by which the phase of G( jo) fo) can be decreased to drive the system to the verge of instability. From the graphical description of Fig. 10.17, we now make the following definitions: 10.4.2 | Phase Margin _ Gain Crossover Point Ihis.apoin on the G(o){sy-plane at which the Nyquist GUo)HYeor-plot has unity magnitude Gain Crossover Frequency 0, Itis the frequency at the gain crossover point, or where [Gia JH jagl= 1 (10.23) Phase margin ‘The phase margin @M of the closed-loop system that has G(s)H/(s) as its open-toop transfer function is defined as oM = 180° + ZG(ja)H(ja,) (10.24) For example, if £G(jo,)H{ je) = 135°, then Bh = 45°; an additional phase lag of 45° could be associated with GGia)Hja) before the closed-loop system would go unstable. For stable systems, Dit is always positive. ‘Consider now an unstable closed-loop system with Nyquist G{ jeyfi( jo)-plot of Fig. 10.18 (G(@) HE) is known to have no poles in the right half of the $-plane). Applying the definition of tcain margin to this plot we find that G(J@,)H{ ja) is a pamber greates than one and therefore Gi ig a number less than one. Applying the definition of phase margin, we observe that the vector G( ja,)Fi( jo) lies in the second quadrant. The angle measured fFom negative seu! ax!5 10 this vector is woigative (counterclockwise measurement is token as positive by d.finition) and therefore, the phase margin @M is negative Controt Systems: Principles and Design Git-plane i Re ky SUH) GC janet joo) Fig. 10.18 Nyquist plot of an unstable Fig. 10.19 Infinite GM, poor dM closed-loop system Jn general, the gain margin and the phase Im margin are mutually independent. For exam- setae GHE-plane ple, itis possible for Gf jaH{ ja) to reflect an o : «excellent gain margin but a poor phase margin. : : Such is the case depicted in Fig. 10.19, where the GAP =e but the WM < 15°. In such a case, the phase crossover frequency ,is undefined. Conversely, Gi jayH( jo) cam reflect a poor but an excellent phase margin, as depicted in Fig. 10.20, where the Ml = co but the GM is only slightly greater than 1. In such a ease, the gain crossover fie quency ©, is undefined. In the light of these observations, gain and hase margins should be used in conjunction with one another when determining the degree &f stability of a closed-loop system. Together, Fig. 10.20 Poor GM, infinite PM tiey represent a measure of the distance of G(JoyH( ja) from the critical point -1 + j0. When we eannot define gain margin (|GGa)| > 0 as 26(jed) + -180°), then only phase margin is used to determine the degree of stability. Only gain nargin becomes the measure of degree of stability when phase margin cannot be defined. ee 10.4.3 | Some Constraints and Cautions Gain margin and phase margin are valid measures of relative stability of a closed-loop system only if its open-loop transfer function G(s)H{(s) has no poles in the right half of the s-plane. \f the open-loop tcansfer function has poles in the right halfs-plane, itis safer to examine the complete Nyquist plot and/ cr oot locus for interpreting relative stability of the closed-loop system. neg val gail sys NyquistiBode Frequency Response Prots and System Stability 11 should be observed that the sketches of Figs 10.17~10.20 each cross the negative real axis or the anit circle atmost once. For systems with numerous zeros and poles, the Nyquist plot may eross the tegative real axis several times, or it may cross the unit circle more than once. For such curves, several values of gain margin and phase margin are defined, as shown in Figs 10.21 (a) and (b). Caution is sue tested when interpreting the lability of such systems. Itis safer not to depend! on the gai margin and phase mangin for relative stability analys it should also be observed that the sketches of Pigs 10.12-10.20 correspond to systems where increasing ain leads to instability. There are certain practical situations wherein increase in gain can make the System stable. Consider, for example, unity-Feedback system with open-loop transfer function, 2 etsy = RHI) his isa system for which increasing goin causes a transition from instability to stability (The Routh criterion shows that the closed-loop system is unstable for K <5 and stable for K> 3). The Nyquist plot in Fig. 10.2le has been drawn for the stable value K= 7. From this figure, we observe that the phase gain margin is less than one. According to the rules for stability discussed cl conflicting signals on system stability. Caution is suggested when inte in margin and phase margin for margin is positive and the earlier, these two margins yiel preting the stability of such systems, It is safer not to depend on thé gal relative stability analy saute 030 @ ) © Fig. 10.21 Nyquist plots requiring careful interpretation of GM and i Oas gain A unity-feedback system has open-loop transfer function, § (10.25) (s+ 25+ 28-42) G(s) Control Systems: Principles and Design are only ia the left half of the s-plane, the Nyquist criterion tells us thal we Since the open-loop poles In such eases, closed-loop stability can be evaluated want no encirclement of {+ /0 point for stability. from Nyquist Gjorplot Ga), = 1520; ja], = 020° G{ jo) = ts = HAC= aw) jut} (10.26) 16-0) #wo'(6-0 siting the imaginary part to zero, we find the phase crossover frequency @, V6 radisec. At this ‘joo in Eqn. (10.26) is calculated to be ~ 0.3 (refer to Fig. 10.22). Thus, the frequency, the real part of GI ppart becomes 1. The gain margin is GM = 3.33 ain can be increased by (1/0.3) before the re: Fig. 10.22 Nyquis! plot for GC jo) of Eqn. (10.26) “To find the phase margin, first find the gain crossover frequency @,: the frequency for which the magnitude of G{Je) in Eqn. (30.26) is unity. As the problem stands, this ealculation requires computa- tional tools, Later in this chapter, we will simplify the calculation process by using Bode plots. ‘The gain crossover frequency «, = 1,253 rad/sec. At this frequency, the phase angle of Gi jw) is =112.33°. The difference between this angle and 180” is 67.67°, which is the phase margin DM. | THE BODE PLOTS ‘A sinusoidal transfer function may be represenied by two separate plots; one giving the magnitude versus frequency, and the other the phase angle versus frequency. A Bode plot (named after Hendrick W Bode) consists of two graphs: one is a plot of the logarithm of the magnitude of the sinusoidal transfer sa plot of the phase angle in degrees; both are plotted against the frequency in function, and the other logarithmie scale. shina iricsisoa ny senna 3s Ina Jog 1G( decibe. The conver form Thi -1 K=lir by Eqn K,=K of thei The The Equ simple comple The quency inform: online Asare Sfreques of thet the Boc ciently Prograr intuitio check 1 Nyquist!Bode Frequency Response Plots and System Stability » a Bode plot, the logarithmic magnitude of sinusoidal transfer function Glo) is represented. \G{jo), where the base of the logarithm is 10. The unit used in this representation of the magnitude is decibel (one-tenth of a ‘Bel’: 1 unit named after Alexander Graham Bell), usually abbreviated as JB. The main advantage of using the logarithmic plot is that the multiplication of magnituiles can be converted into addition. Consider a typical rational transfer function G(s) factored in the time-constant form: K (14 stg 5%) 0.27 Os) = (best, Ml sty) [1+ (201 0,)s-+ 5° 13) This transfer function has zeros at s =~ Vt, — It, ... N poles at the origin, real pol I/t,, — 1/4, .., and complex poles at Co, + je, We note that the constant multiplier Hims" Gs), where Wis the type number or the by Eqn. (10.27). Thus for a type-0 system, K, = K. Fora type-I system, K,= K, and fora ty K,= K. IF G(s) has complex zeros, then even though it is not shown in Eqn. (10.27), a quadratic term ofthe form [1 + (2¢/@,\s + s°/«2,] will also appear in the numerator. ‘The magnitude of G( oo) in UB is obtained by multiplying the logarithm to the base 10 of Gt jeo] by 20 1GCj@)|yy= 20loz|Gj@)| = 2D logh +20 log [1 + jeor,| + 20 log [I + feoz,| 20 log Yok ~ 20 log ft + jeos)~ 20 log [+ For ~ 20 log} 1+ j2Ca! a, ~ 0 fa |= (10,28a) ‘The phase of Gj) is ZG(oo) = tana, + tan here, ++ = N(90®) ~ tan bor, — tan hor, ~ stan"! Sono, / (wis =") =. 3(10.28b) Equations (10.28) express the magnitude and the phase of G(jo) as linear combinations of relatively simple terms. The curves for each term can be added together graphically to get the curves for the he complete transfer function : ‘The logarithmic scale used for frequency in Bode plots, has some interesting properties. The fre~ quency in a typical control system application varies over many (twolthree) powers of ten so that most information would be compressed near the origin if a linear scale were used, The logarithmic scale is nonlinear, that is, the distance between | and 2 is greater than the distance between 2 and 3 and so on. Asa result, use of this scale enables us to cover a greater range of frequencies; both low and high frequency behavior of « system can be adequately displayed in one plot. The other interesting property - “fof the use of log scale for frequency, as we shall see shortly, is that the straight-line approximation of the Bode plots becomes possible. This property allows the plotting by hand that is quick and yet suffi- ciently accurate for control-system design, Most control-system designers will have access to computer programs that will diminish the need for hand plotting; however, itis still important to develop good intuition so that erroneous computer results are quickly identified, and for this one needs the ability to check results by hand plotting. de fer GB conta systems Principles and Design ot, ZG(jea) versus log 0, has 2 linear seale for phase in degrees “the phase angle graph of the Bode pl joo) yg versus log ©, and lopaithmie seae for frequency «. The magnitude graph of the Bode plot, |G haga lincar scale for magnitude in JB and a logarithmic seale for frequency © ih general rational transfer Function Gls) ean contain just four simple types of Factors: Real constant: K> 0 Poles or zeros at the origin of order NV (** Poles or zeros at s=~ Vrof order q: (+50) Complex poles or zeros of order r: (142637 @, 457/04)" indicate one of the unique characteristics of Bode plot—each of the four types ste plot; the individual plots are then algebraically notion, Bode plots of the four types of factors plot of the given transfer Equations (10.2% of fuctors listed can be considered as a se ‘added to yield the total plot of the given transfer fun sro, therefore, the basic building blocks for the construction of the Bod function 10.5-1 | Magnitude Plot: S ‘We shall now discuss how the |G(jO) ly Yersus log oo graph can be sketched with very lle effort and then use these blocks using straight-line approximation, We first examine the basie building blocks, to build the magnitude plot of the given transfer function. ‘The dB versus log @ graph can be sketched on a linea “ever, use af semilog graph paper is more convenient as it eliminates the need t numbers. The logarithmic scale of the semilog graph paper is used for Itequency and the linear scale is used 40 fordB. 30|-—| The ufité used to express frequency bands or fre- quency fntios are the etave and the decade. An octave is a frequency band fom @, to @, where @/0, = 2. There is an increase of one decade from «(0 @, whe, (ogo, = 10. Semilog geaph papers come in two, three, four or five cycles, indicating the range of coverage in Uccades (refer to Fig. 10.23). 1# may be noted that we + rectangular coordinate graph paper. How- 1o take logarithm of many ll cannot locule the point @= 0 on the log scale since O65 172. 5 10 @(rad/sec) —> -. 10.25 magnitude ple ‘K Real Constant K g- 10-23 Bade magni plot of Since the constant K is frequency invariant, the plot of UB = 20!ogK (10.29) is. horizontal straight line. The magnitude plot for K nates. The horizontal axis is tog @; labeled « on log decibels; labeled dB on linear scale 10 is shown in Fig. 10.23-in semilog coordi- jthmic scale. The vertical axis is magnitude in Poles The fa The dt log. Thi range « could : the O-< For The =20N« For and the decade origin, “B 20 10 NyquisliBode Frequency Response Piots and Systern Statily Poles or Zeros at the Origin ® “The factor 1/(je) appearing in a transfer function (yoo) has the 1: aB = 20log] | -201og (10,30) jo “The dB versus log @ plot of Eqn. (10.30) isa straight line with a slope of ~20 dB per unit change in toy (A unit change in log co means Tog(oJeo) = Lor = 100 > aly ‘This range of frequencies is called a decade. Thus, the slope of Eqn. (10:30) is -20 dBidecade, The of range of frequencies c, = 2a is called an octave. Since ~20 log 2 = ~6 dB. the slope of Ean (10.30) oa sani also be expressed as ~6 dBloctave. The plot of Eqn (10:30) is shown in Fig, 10.24a: it intersects the 0-dB axis at = 1 For an Nth-order pole at the origin, the magnitude is B= 2010g|— (1931) 2 I jo) fort ‘The magnitude plot is still a staight line that intersects the O4IB axis at w= 1, but the slope is tow seks ~20N dBidecade. Hor the eace that a transfer fimetion has a zero at the origin, the magnitude ofthe teem is given by cae dB = 20 log [jai = 20 toy (10.32) : vin; itis a straight line with a slope of + 20 dB/ ‘and the plot is the negative of that for a pole atthe o decade that intersects the O-dB axis at @= 1 (Fig, 10.24b). For the case of an Nth-order zero ak the fj origin, itis seen thatthe plot is the negative ofthat forthe Mth-order pote | 20 dB/decade 2 ect . | a. a af s20aRideente | : ie joibioame +6 dBloctave 10 19 s 6 0 . ——- 6 10 at it -10 -20 —20 ~ ol. 1 2 10m Ol 12 0 @ @ © Fig. 10.24 Bode magnitude plot of (a) pole; (b) zero at the origin 0.29) ed 10-8 | Consider the transfer function, K as Conirol Systems: Principles and Design fed pole at the origin, We could add the There are two terms in G(s): the constant K, and the repeat treating Kis” thoce magnitude plots of the Iwo terms 10 get the Bode magnitude plot of G(s), However, gle term turns out to be more convenient, as is seen below. dB = 20log| 20N loge 20log.K (20.33) 1 straight line having a slope -20N dB/decade th log as abscissa, the plot of Eqn. (10.33) is as shown in Fig. 10.25. Further, the ‘and passing through 20 log K dB when log «= 0, i.e. when 2 plo: has.a value of 0 dB at the frequency of uw 0 log K or = (K 20N log w By 40 -40 dBsdecade The 30 a ecura -20 dBMdecade M+ range fireque region The error ¢ se 10 Fig. 10.25 Bode magnitude plot of KA" oe ig ode magnitiute plot of KA ad Poles or Zeros ats =-I/t “The factor 1/() + /002) appearing in a transfer funtion Gi jo) has magnitude ss 2 and (1 8 =2010g|— 20logVi+ ot (1034) T+ jor j Fer low frequencies, such that «><< 1/1, the magnitude may be approximated by 2 The dB =-20 log 1=0 (10.35) § For high frequencies, such that «9>>1/r, the magnitude may be approximated by dB = —20 log(t) = -20 log @~ 20 log t (10.36) ‘The plot of Eqn, (10.35) isa straight line coincident withthe 0 GB axis. The plot of Ean. (10.38) 5 3 alko a suaight line having a slope of -20 dB/decade intersecting the O dB axis at @= 1/5, as shown 4 Sin Fis 10-26. We cal the straight line approximations the asympotes. The low-frequency approximations In Frequency approximation is called the high-frequency by 3 called the low-frequency asymptote and the h anymptote. The frequency «= V/r at which tl break frequency’ fhe two asymptotes meet is ealled the comer frequency OF | Aycuisode Frequency Response Pits and System Stabilly 42) Low- fcguesey Comer ‘eaeney asymptote frequency (_poqp/decade) -@ Fig, 10.26 Bode magnitude plot of 1/(1 + jor) ‘Though the asymptotic approximations hold good for @ << 1/t and @ >>1/z, with some loss of aecuracy these could be extended for «@S I/rand «@2 I/t. Therefore, the dB versus log co curve of Ti{l 17) can be approximated by two asymptotes, one a straight line at D UB for the Frequency range 0 < oS Urand the other, a straight line with a slope ~20 dB/decade (or = 6dB/octave) For the frequency range #5 @< @,,= Is this factor contributes Pie apyidevade such thatthe resultant plot of the tsvo factors i the asymptote 2 of slope +20 dB/decade passing through (6 dB, | radisee) point At = a2 0, the resultant plot has 2 magnitude of 26 dB as shown in Fig. 10.28. Step 3: We now ada tothe resultant plot of Step 2, the plot ofthe factor 1/(1 + 00/10)" corresponding, to the comer frequency @2 = 10. Since this factor contributes Q GB for @ £ My = 10, the ‘resultant plat up to = 10 i the sameas that of step 2 For @ > @h 10, this factor contributes reas B/deeade such that the resultant plot of the three factors i the asymptote 3 of siope (420) + (0) = -20 dBdecade passing through (26 aB, 10 radlsec) point. Figure 10.28 Shows the asymptotic magnitude plot of given (Ja) sponding to the lowest = 1, the resultant sion tnble for Bosle magnitude plot of 201 + JOT jo/l0¥ Asymptote 2 (420 dB/decade) Asymptote 3 30 , (20 dBidecade) Cont! Systems: Principles and Design : To the asymptotic plot thus obtained, corrections are to be applied. The corrections at each corner frequeney and at an octave above and below the comer frequency are usually sufficient, The corner frequency €2, = | corresponds to the first-order factor (1 + je): the corrections are +3 dB at w = 1, +1 UB at o= 05 and +1 dB at @=2. The comer frequency «,, = 10 corresponds to second-order factor I/ (1 + 0/107: the corrections are ~6 UB at «= 10, ~2 dB'at m=5 and —2 dB at c= 20, Table 10.2 lists the net corrections. Table 10.2 Corrections to the asymptotic moguitude plot of 2(1 + jw)/(1 + j@/t0)” Frequency, >| 05% 1 2 5 0 20° Netconection | #1dB 4308 dB -2dB_— 6B -2.0B ‘The corrected asymptotic magnitude plot of the given transfer function is shown in Fig. 10.28. 1O(+ jo! 2) (jo) (+ ja) Consider the transfer function, G( jo “The comer Frequencies of the asymptotic Bode magnitude plot of G(/w) in order of their occurrence as frequency increases are (i) @,= I, due: to simple pole, (i) «= 2, duw to simple zero. Atfiequencies less than ©), only the factor 10/( joo? is effective. Asymplotic magnitude plot of G(jo) is shown in Fig. 10.29. The plot is obtained following the steps given below Step 1; We start with the factor 10/(ja)? cor ~~} magnitude plot is the asymptote 1, having a slope of -40 dB/decade and passing thronyh the point 20 log 10 = 20 dB at w = 1. Asymptote J intersects the 0 dB line ato +Ji0 Let us now add to the asymptote J, the plot of the factor 1/(1 + ja) corresponding to the lowest comer frequency @,, = 1. Since this factor contributes zero dB for «$1, the resultant plot up to «= 1 is the same as that of asymptote 1. For «>1 this factor contributes ~20 dB/ decade such that the resultant plot of the two factors is the asymptote 2 of slope (~40) + (-20) =~60 dB/decade passing through (20 dB, 1 rad/sec) point. At a= @.,=2, the resultant plot has a magnitude of 2 dB as shown in Fig. 10.29. Step 3: We now add to the resultant plot of Step 2, the plot of the factor (1 + ja!2) corresponding to the corner frequency @,, = 2. This gives rise to a straight line of slope + 20 dB/decade for > 2, which when added to asymptote 2 results in asymptote 3 of slope (-60) + (+20) = ~40 dB/decade passing through (2 dB, 2 rad/sec) point, Figure 10.29 shows the asymptotic magnitude plot of given G(Jo). “To the asymptotic plot thus obtained, corrections are to be applied. The corrections due to asymp- totic approximation of magnitude plot ofthe pole factor are: ~3 dB at @= 1,1 dBaLo=0.5 and=I dB 2. To hese corrections, we algebraically add the corrections due to the asymptotic approximation 41 dB at @ = 1, and +1 dB at @ = 4. Table 10.3 sponding to double pole at the origin. Its Shep 2 nitude plot of the zero fac ists the net corrections. Comp Transfe This ter of the v dampin Inc where 1 For it ‘NyquistiBdde Frequency Response Plots and Systein Stability #3 Asympiote 1 (40 dB/decade) | o= | (-60 dBidecady | 4 Asymptote33 a (-40 dB/decade) . i i er lo Ol 05 I 2 4 10 nee Fig. 10.29 Bode magnitude plot for Example 10.10 Table 10.3. Corrections to the asymptotic magnitude plot of 10(T +. joo Yeo? {1 + jo) Tequency, @ : : : : 4 | Net correction 42dB +1 dB eps — a “The corrected asymptotic magnitude plot of the given transfer function is shown in Fig. 10.29. Its sing Complex Poles or Zeros line “Transfer fanctions of control systems often possess quadratic factors of the form, fs ——_1___,.0<¢s1 (10.37) 1+ J(26 le, o-oo, since besides being 2 fimetion of was before, itis also a function This term is slightly more complicated he shape of the Bode plot depends strongly upon what value of of the variable ¢. As may be expected, tl damping ratio ¢ is being considered. In normalized form, the quadratic factor (10.372) may be written as (10.37b) 1 = a, is the normalized frequency. The magnitude of this Factor is 20 tog Jd 0 (Eu? (10.38) jated by where « | 4 tau? + j2bu B= 20 tos such that 1 << 1, the magnitude may be approxi For lovr frequencies, dB =~20 log 1=0 (10,39) geo sguch that u >> 1, the anagnitude may be oP 40 log proximated by For high frequencies, cosa) 0 Jog B nitude plot of the quadratic factor (10-37b) consists of two straight ve | and the other, a line witha slope ~10 dBdecade an approximate mags a= ,, which is the comer frequency of one horizontal Tine at 0.dB for iplotes meet on 0 OB Hine at = Ee Fig. 10.30. y the use of asymplotes c 10.38) and (10.39) Therefore, line asymptotes, foru > 1. The two asym the plot. The asymptotic plot is shown in ‘The error in the magnitude plot caused by 1 1o Eqns Jan be calculated as follows: For 0 <1 tthe error is (ref -20 tog P(2fuy +20 tog (10412) and for $1 <= the error is (refer to Eqns (10:38) and (10.409) _20 tog Y= + 2Gu)? +40 Tog cosy | “The error at the comer frequency (= 1 or @= 04) i8 F 20 10g 26 (0.82) Bt Low Comer j Frequency frequency asymptote High-frequency 7 30 ‘asymptote s Pee 40 (-40dB/decade) “ 50 coh a 0.1 v 0 Fig, 10.30 Bode magnitude plot of Ya+pabu-w) For {= 1, the error atthe commer Frequency 15-20 To 22 ~ 6 oB, Note that for = 1, Ban. (03? has two equal real poles: 4 F| Set cease Ga jy + jo! a, ae J the error is negative for ¢> 0.5 a at the corner frequency, fhude curves for various values of ye magnitucle curves smaller the value o} Core shown in Fig. 10% 9 have pronounced petk ya a F ¢, higher the pea From expression (10.42) we set that fe for €< 0.5. The corrected magni ova this figure, itis seen that for small valves of 6 th Freauiency slightly less than the commer frequen” Ip fact, shall discuss this peak in some detail 31, which shows what might be a typi! plot of (fe/er 10 Eqn. (10.38) aB=~20 logy(i= WF #260" pos Consider Fig. Hl ‘The peak i= be four whieh or The No! quenc? cornet and tht For sion © ‘The pre line fo plot bs Nyquist!Bode Frequency Response Piots and System Stability Fig. 10.31 A typical magnitude curve of T/(1 + j2Gu ~ 1) “The peak value occurs at normalized frequency which is labeled u, (1, = @ey,) and the height of the peak is Mf, This peak will occur at a u where (ai)? + (2Gu)* is at a minimum. The minimum can be found by simple differentiation, Af (Oe which gives au} ~4u, +80, (10.43 or “The corresponding peak value M, is then 20Iog V(t =-20log 261-2? aB aaa (refer to Eqn. (10.44))-and fre- N12, u, is'les8 than the M, Note that as ¢ goes to zero, the peak value M, goes off to infinity quency 1, approaches the corner frequency (refer to Eqn. (10.43)). For 0 << Vs corner frequency. For £> V2, the errors given by expressions (10.41) are negative for all frequencies and therefore magnitude curves do not have a peak. For given values of Cand ,, an exact dB versus loge cnrve is obtained from the following expres sion (refer to Eqn. (10.38)): j 22) (0.45) o dB=-20 log |] 1-| — i (3) ®, The following procedure is useful for hand plotting: Draw two straight line asymptotes: one a 0 dB line for 0.< «@ $e, and the other a ~40 dBidecade line for @, < 0-<=°, The two asymptotes meet at the comer frequency «= 0, Correet the asymptotic plot by MM-dB at the frequency «, where (refer to Eqns (10.44) and (10.43) : e (10.46) (10.466) Gen syste Principles and Design of course, this correction is applied when 6< Wi. Correct the asymptotic plot by ~20 log (20) 8 suihe comer frequency © = e, (refer to expression (10.42). Draw a ‘smooth curve through these points approaching. the low and high frequency asymptores by correcting the asymptotic plotat some other frequencies “The accuracy of the plot can be improved also; say one octave below and above the comer frequency. ‘The errors are calculated from the following expressions (refer to expressions (10.41): 20108 }-(2] (2) for 0 2, this factor contributes +20 dB/decade such that the resultant plot of the two factors is the asympiote ? of slope (~20) + (420) = 0 dB/decade passing through (6 dB, 2 rad/see) point. At @= 3 = 10, ssultant plot has a magnitude of 6 own in Fig, 10.32, Step 3: We now add to the resultant plot of Step 2, the plot of the factor 1/(1 + ja/i0 ~ with a slope 0 + (~ 40) =—40 dB/decade pa: yquistiBtode Frequency Response Plots and System Siebiiy 71100) corresponding to the corner frequency @.= @, = 10. This gives of slope — 40 dB/decade for «>10 which when added to asymptote 2 results in asymptote 3 “To the asymptotic plot thus obtained, corrections are to be applied. Table 10 list BL Asymptotel so} (-20dB/decale) 30} A 2 | ‘Asymptote 2 2b 6 0 -10 1 Asympicte -20 { (40 aBidecadey aaa eee 1 124 ° Fig. 10.32 Bode magnitucle plot for Example 10.11 toa straight line sssing through the point (6 dB, 10 rad/sec). the corrections. of (1+ jay2)/Ga)(l + jeo/10 ~ o7/100) +3dB +1dB +1dB odB es remarks: a Ertor 2-20 log 2. 0 tog2eyt— "The phase angle at any frequency can be obtained as th Ft et Ot Oe The corrected asymptotic plot of the given transfer funetion is shown ini Fig, 10.32. algebraic sum of the phase angles due to various actors in the transfer function, ina manner similar to that used for obtaining magnitude plots (refer to Eagns (10.28), To illustrate the procedure, we consider the transfer function, (L408) Gtjo)= 3 1048) 10) xia form llr U2E 1a, )o— Ve Amie “The phase of Gad is ZG (je0)= tan 9, ~90°— tan"! wy —ton"* 2608, Hod — 0? (10.50) aero syste: Principles and Design ‘The first term on the right-hand side of Eqn. ci The values of for various Fig, (0.33. We can approximate t ‘The straight line approxima and breaks again to a constant value of 90° at 5.72, and occurs at the two corners 1/10%, ‘The second term on the right-hand side in caclly equal 10 -90°. The third term in Eqn. (10:50) but for reversal in sign. We she pair of poles, and is given by (10.50) is due to the real zero, and is given by tants, 1s values of ware given in Ts the exact curve with the straig! ‘on for the phase characteristic break = 10/1, The maximum error in the approxim: and 10/7, qn. (10.50) is due to the pole atthe origin and is identi wall now Jook at the fourth term, able 10.5, and these values are plotted in pt line construction shown in Fig. 10.33. ¢ from 0° atthe frequency @= 1/107), ‘due to the real pole and is similar to the first term which is due to the complex-conjugate oot ow Table 10.5 Phase characteristics of (1 +/@%) that the oF when t i ofone 29" o° Pease, Soaps sr oO q overail 2 ee 3 For i Wee factor 26.8 sis sing 38.7 40.6° compu 45° ase oa? 58.5" pie BP 65° 52.9° 85.6" PGR SRS 81.12 90° : For @ << 0.1@,, we have @ = 0, and for cor>l0@, We get Oy = 180°. At the intermedi- ate Frequency @ = @,, we have dy = 90°. The plots of the asymptotic approximation and the pewal curve for various values of { are shown in Fig, 10.34. It is seen that for small values of 6, ‘the actual phase angle changes very rapidly for @ near the undamped natural frequency ©, “The overall phase angle is now obtained as the algebraic sum of the phase angle due to each factor. The addition, however, is not as simple as inthe case of the magnitude plot due to the fact ee ° s bow 0. 2 io” 20 1 a tT hy 1 Fig, 10.33 Phase charncterisics of (1 +j%) 2d in 2.33. 10%, onis lenti- term agate soar 4 cies yauisuBode Frequency Response Plots and System Stabiity 90° 120°}, -150" 180° 0.10, 100 Fig. 1034 Phase chumcteristics of 171+ (j26, 0-0") shot the straight line approximation consists of three parts over four decades. The only exception arises vthen the various poles and zeros have ‘large’ separation, so that frequency range of phase angle plot rn factor does not overlap with the Frequency range of phase angle plot of any other factor. In this gree, the phase angle plots for the various factors can be superimposed on each other to obtain the overall phase. For a general case, itis usually advisable to make a table of phase angle ag factor and then obtain the total phase as the algebraic sum of these. Better using the exaet values in the table instead of those obtained through the asymptotic ap computational effort to achieve this accuracy is not heavy. ginst frequency for eae’ necuracy is obtained by proximation. The £4 Consider the transfer function given in Eqn, (10.48) for which the magnitude AZ” i af ig plot was obtained in Fig. 10.32. We shalt now plot the phase curve. Solution: ‘The phase angle due to each factor for-ditferent values of «2 is given in Tpble 10.6. ‘The plot of the total phase is shown in Fig. 10:35. : Table 10.6 Phase angle due to different factors of the transfer function given by Eqn. (10. 3) ES =a a oa ART “210132 (continued) ¢_ Control Systems: Principles end Dose? (radé. see: 20 [ ~146.31° = 152.02" 30 Yon, 230 ey! = 170.52" 100 = 174.22 = 175.68" Fig. 10.35 Bode phase plot for Example 10.12 10.6 | STABILITY MARGINS ON THE BODE PLOTS bjective of the frequency-domain design procedures is to shape the Nyquist plot sueh that the critical point ~1+ j0 is avoided with some reasonable stability margins, thereby providing aecept- able closed-loop respousie characteristics, Direct use of the Nyquist plot for design is not particularly Convenient, since chins in parameters other than gain require extensive plot revisions. Fortunately, ‘general charheteristies of Nyquist plots can be visualized with reasonable accuracy in most cases of interest from Bode plots, and Bode plots are easily constructed and modified Since the Boule plot corresponds to only the positive portion of the ji-axis from @=0 to @ in the s-plane, its use in stability studies is limited to the determination of gain anil phase crossover yonding phase and gain margins. Recall that gain margin and phase margin points and the corresp ‘are valid measures of relative stability of a elosed-loop system only if the system has open-loop transfer function with no poles int right half of the s-plane. Since the straight line approximation of the Bode plot is relatively easier to construct, the data neces- sary forthe other freque Bode plot ‘ney-domain plots, such as the Nyquist plots, can be easily generated from the K srA(s+5)" (10.51a) GG Let us determine thw stability of this system implementing the Nyquist stability criterion using Bode plots. Solution transfer { The si Bode The P point for than Od phase is magnitur unstable. @0 log 1 is 200 x Next found by ig 180°, grin req. ution: i aster faction G(s) is expressed as as) (10.5tby The sinusoidal transfer function G(jo)= Cs jos FOC jos Bode magnitude and phase plots for G(ja)) are shown in Fig. 10.36. @(rad/see) —> Fig. 10.36 Bode plot for Example 10.13 xxample tells us that we want zero encirelements of the -1-* j0 point for stability. Thus, we recognise that on Bode magnitude plot, the magnitude must be less than 0 dB (which corresponds to unity gain of G( ja) on polar plane) at that frequency where the phase is 180°. Accordingly, we see that ata frequency of 6.2 rad/see where the phase is -180", the Inngnitude is ~5.5 dB. Therefore, an increase in gain of +5.5 dB is possible before the system becomes instable. Since the magnitude plot was drawn for a gain of 200 (refer to Eqn, (10.51a)), + 5.5 4B {0 log 1.885 = 5.5) represents the allowed increase in gain above 200. Hence, the gain for instability is 200 x 1.885 = 377. The closed-loop system is stable for 0 < K-< 377. ‘ow how to evaluate the gain and phase mnargins using Bode plots. The guin margin is where the phase angle GM, whick is the ‘The Nyquist criterion for this e» veces: m the Next we s found by using the phase plot to find the phase erossover frequency @y s180°. Ac this frequency, look at the magnitude plot to determine the gain margin, ‘magnitude curve 10 0 dB. From Fig. 10.36, we find that = 6.2 radisee; GM=5.5 dB »3ta) ‘gain required to raise the Q_cena Systems Principles and Design ‘The phase margin i found by using the magnitude curve tof the gain eressoner frequency Oy slereiheveuin is OB. On the phase curve at that frequency, the phase margin, PM is the ilifference between the phase value and ~180°. From Fig, 10.36, we find that = 4.5 asec; OM = —157? ~ (180°) = 2° ser fa ‘iXample 10.14 ¢ Consider now o unity-feedback system with open-loop tran: K Gls)= —— i K = 100 S(1¥0.1s)(14 0.25) figure, we find that “The Bode plot of G(ja) is shown in Fig, 10.37. From thi 50! 40 30 dB 20 10 0 4 wy fe (rad/sec) — 40, Fig, 10.37 Bode plot for Example 10.14 (y= phase crossover frequency = 7 rad/sec GM in margin = -16.5 dB ‘@, = gain crossover frequency = 16 rad/sec bat phase margin = 10° The closed-loop system is thus unstable. : Tae were to decrease the gain by 16.5 dB, all points on JB vs «plot would decrease in magnitude wy this amount without affecting the phase plot. The closed-loop system would be a the Hit of insta~ ity The gai K must be les than 100 % x (16.5 = 20 log), Le the stability ange is 0 < K-< 668. Figure transfe where Roo sible, b analysi This tpis sr The dend-ti Bode p Assum becom The Cle: Sine indefin [NyquistlBode Frequency Response Plots and System Stabilly 6 1 STABILITY ANALYSIS OF SYSTEMS g1007 WITH DEAD-TIME Figure 10.38 shows the block diagram of a,system with dead-time elements in the loop. Open-loop transfer function of the system is Gs)His) = Gye" (10.52) where G,(s) is @ rational function. Rls) + —— %) Root locus analysis of systems with dead-time elements is pos- sible, but the method is quite complex. In Section 8.5, root locus analysis was carried out using the following approximation: van _lettp!2 Lest, 12 Fig, 10.38 A closett-loop system ‘This approximation works fey well as lngasthe dente oe eet is small in comparison to the system time constants. The frequency-domain graphical methods provide a simple yet exact approach to handle the dead-time problem since the factor &°*"* is readily interpreted in terms of either the Nyquist or the Bode plot. { ‘0 10.7.4 | Nyquist Plots of Systems with Dead-Tin mnction of the system, ‘Assume that G,(s) in Fig. 10.38 is an integrator. Then the open-loop transfer becomes G(H(s)= The sinusoidal transfer function is pie GC joyH{ jo) = Clearly, enty~ ni2 IG jayH jo) = Vex, 2G j@}H(jo) L Gljoyi(jo)= + (cosety — jsinwry) | jo sinwr, o coset y lim [G{Jo)H(Jen)]= tp — fo Since the magnitude decreases monotonically, and the phase angle also decreases monotonically indefinitely, the polar plot of the given transfer function will spiral into the «@—> «» point al the origin, as é_ContolSystors: Principe ond Desi show? in Fig. 10.39. A curious reader may Hike to iad Im ome of the intersections with real and imaginary axes. i For the first intersection with the real axis, set 24) GH-plane Im [G{jadHi{ ju) = 0 to obtain (tp = #2 Re[G joy Jo] 2tpia Once the Nyquist plot is constructed, the stability of the closed-loop system is determined in the usual manner, From Fig. 10.39, we observe that relative ‘ability of the system reduces as Tp increases Sufficiently large values of Tp may drive the system to instability. The Critical Trajectory ‘Thus far, we have used the point ~1+ 0 of the Gis) Hiskplane as the critical point for stability analysis casing Nyquist criterion. We can extend the critical: point idea into a trajectory if necessary. a tr otthe characteris equation ofthe system shown in Fig, 10.38 satisfy (refer to Eqn. (10.52)) ‘o-~0 Fig, 10.39 Polar plot of o°*">/s 1eGipe” = 0 1 (0.53) or Gsye ‘The righi-hand side ofthe last equation points tothe Fact that ~14+,0 is the critical point for stability anciysis ofthe elose-loop system. Equation (10.53) ean be written as Gis) (0.54) “The corresponding condition for the system without dead-time is , G\@)=-1 (00.55) Comparison 6f Eqns (10.54) and (10.55) reveals thatthe effect of dead-time is simply to shift the crhical ability point —1+-j0 10 ~e'", which deseribes a critical rajectory, When fio 0, the trajec- tory stats atthe point -1+70, and a8 ep increases, te eitical point faces out circle with unit radius ‘centered at the origin of G,(s}-plane, in the counterclockwise direction. Let us determine, with the help of Nyquist stability criterion, the maximom Naloe of tp for stability ofthe closed-loop system of Fig, 10.38 with Gus s(s+ Ist 2) Fron Figt at whic to unit, which Since. we ob whicl Fr Gu we!

You might also like