1-s2.0-S0959152411002198-main

15
Journal of Process Control 22 (2012) 11–25 Contents lists available at SciVerse ScienceDirect Journal of Process Control jo u rn al hom epa ge: ww w.elsevier.com/locate/jprocont Extending the Symmetrical Optimum criterion to the design of PID type-p control loops Konstantinos G. Papadopoulos a,, Nikolaos I. Margaris b,1 a ABB Switzerland Ltd., Department of Power Electronics & Medium Voltage Drives, CH-5300 Turgi, Switzerland b Aristotle University of Thessaloniki, Department of Electrical & Computer Engineering, GR-54124 Thessaloniki, Greece a r t i c l e i n f o Article history: Received 17 July 2011 Received in revised form 23 October 2011 Accepted 23 October 2011 Available online 23 November 2011 Keywords: PID control Design Tuning Optimization Process control Control engineering Industrial control a b s t r a c t An extension of the Symmetrical Optimum criterion for the design of PID type-p closed- loop control systems is proposed. Type-p control loops are characterized by the presence of p integrators in the open- loop transfer function. For designing a PID type-p control loop there should exist an PI p D, or PI (p1) D, or PID and so on, if the controlled process is of type-0 or type-1 or type-(p 1) respectively. A type-II control loop achieves zero steady state position and velocity error, a type-III control loop achieves zero steady state position, velocity and acceleration error and therefore a type-p control loop is expected to track both faster reference signals and eliminate higher order errors at steady state. For deriving the proposed control law, a transfer function containing dominant time constants and the plant’s unmodelled dynamics has been considered in the frequency domain. The final control law consists of analytical expressions that involve both dominant dynamics and model uncertainty of the controlled process. For justifying the potential of the proposed theory, simulation results for representative processes met in many industry applications are presented. © 2011 Elsevier Ltd. All rights reserved. 1. Introduction The need to design higher order type control loops has always been challenging and critical over the academic and industry soci- ety [1,2]. From the theory, it is known that higher order type control loops have the advantage of tracking fast reference signals [1]. In order to achieve this target, and if the design of the control loop is based on the frequency domain, the design of a type-p control loop involves the existence of p integrators in the open-loop transfer function. Hence, if for the design of a type-p control loop, PID type controllers are adopted, then the sum of the pure free integrators of both the process and the controller should be equal to p. Throughout the literature, it is evident that the PID control law offers the simplest, feasible and yet most efficient solution to many real-world control problems, see [5–22]. More than 90% of industrial controllers still implemented, are based around PID algorithms, see [17–23]. However, the demanding problem of designing higher order type control loops has been treated by many researchers after employing or modifying well established control schemes such as the IMC 2 principle [1,25,26] or the Smith predictor Corresponding author. Tel.: +41 58 5 893242; fax: +41 58 5 892580. E-mail addresses: [email protected], [email protected] (K.G. Papadopoulos), [email protected] (N.I. Margaris). 1 Tel.: +30 2310 9 96283; fax: +30 2310 9 96447. 2 Internal Model Control. [27]. Main drawback of the above approaches is that the proposed control laws are restricted to the design of type-II control loops and whatsmore, for verifying their control laws potential, simple process models are employed [9,24,32] (first order plus dead time process, first order reduced integrating process). Over the litera- ture, a first attempt of designing type-III control loops can be found in [3,4]. There, it is shown that by applying the principle of Symmet- rical Optimum criterion along with the use of PID type controllers, robust type-III control loops can be designed. In similar fashion and since the aim of this work is to present a feasible control design method that can be applied in many industry applications, once more the simplicity and widespread application of the PID con- trol law will be exploited. Therefore, the approach proposed in that work, is faced with the following problem. Tune a PID type controller, such that the output of the final closed-loop control sys- tem eliminates higher order steady state errors, position, velocity, acceleration etc. For developing the proposed theory, the principle of Symmet- rical Optimum criterion will be adopted. In the sequel, it will be shown that for achieving the design of type-p control loops along with the aid of the Symmetrical Optimum criterion, the conven- tional design principle for tuning PID type controllers has to be adopted. This conventional principle implies that exact pole-zero cancellation has to be achieved between the process poles and the controller’s zeros. Therefore, for applying the proposed theory all dominant time constants of the process have to be measured accurately. The Symmetrical Optimum criterion is an extension of 0959-1524/$ see front matter © 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.jprocont.2011.10.014

description

type 3 control

Transcript of 1-s2.0-S0959152411002198-main

  • Journal of Process Control 22 (2012) 11 25

    Contents lists available at SciVerse ScienceDirect

    Journal of Process Control

    jo u rn al hom epa ge: ww w.elsev ier .com

    Extending the Symmetrical Optimum criterion toloops

    Konstanta ABB Switzerla itzerlb Aristotle Univ essalon

    a r t i c l

    Article history:Received 17 JuReceived in reAccepted 23 OAvailable onlin

    Keywords:PID controlDesignTuningOptimizationProcess controControl engineIndustrial control

    timumops a

    PID tyf typed ver and

    ordeant tiThe ncer

    lts for

    2011 Elsevier Ltd. All rights reserved.

    1. Introduction

    The needbeen challeety [1,2]. Froloops have order to achbased on thinvolves thfunction. Hecontrollers of both the

    Throughlaw offers to many reof industriaalgorithms,designing hresearchersschemes su

    CorresponE-mail add

    kpapadop@en1 Tel.: +30 22 Internal M

    [27]. Main drawback of the above approaches is that the proposed

    0959-1524/$ doi:10.1016/j. to design higher order type control loops has alwaysnging and critical over the academic and industry soci-m the theory, it is known that higher order type controlthe advantage of tracking fast reference signals [1]. Inieve this target, and if the design of the control loop ise frequency domain, the design of a type-p control loope existence of p integrators in the open-loop transfernce, if for the design of a type-p control loop, PID typeare adopted, then the sum of the pure free integratorsprocess and the controller should be equal to p.out the literature, it is evident that the PID controlthe simplest, feasible and yet most efcient solutional-world control problems, see [522]. More than 90%l controllers still implemented, are based around PID

    see [1723]. However, the demanding problem ofigher order type control loops has been treated by many

    after employing or modifying well established controlch as the IMC2 principle [1,25,26] or the Smith predictor

    ding author. Tel.: +41 58 5 893242; fax: +41 58 5 892580.resses: [email protected],g.auth.gr (K.G. Papadopoulos), [email protected] (N.I. Margaris).310 9 96283; fax: +30 2310 9 96447.odel Control.

    control laws are restricted to the design of type-II control loopsand whatsmore, for verifying their control laws potential, simpleprocess models are employed [9,24,32] (rst order plus dead timeprocess, rst order reduced integrating process). Over the litera-ture, a rst attempt of designing type-III control loops can be foundin [3,4]. There, it is shown that by applying the principle of Symmet-rical Optimum criterion along with the use of PID type controllers,robust type-III control loops can be designed. In similar fashion andsince the aim of this work is to present a feasible control designmethod that can be applied in many industry applications, oncemore the simplicity and widespread application of the PID con-trol law will be exploited. Therefore, the approach proposed inthat work, is faced with the following problem. Tune a PID typecontroller, such that the output of the nal closed-loop control sys-tem eliminates higher order steady state errors, position, velocity,acceleration etc.

    For developing the proposed theory, the principle of Symmet-rical Optimum criterion will be adopted. In the sequel, it will beshown that for achieving the design of type-p control loops alongwith the aid of the Symmetrical Optimum criterion, the conven-tional design principle for tuning PID type controllers has to beadopted. This conventional principle implies that exact pole-zerocancellation has to be achieved between the process poles andthe controllers zeros. Therefore, for applying the proposed theoryall dominant time constants of the process have to be measuredaccurately. The Symmetrical Optimum criterion is an extension of

    see front matter 2011 Elsevier Ltd. All rights reserved.jprocont.2011.10.014inos G. Papadopoulosa,, Nikolaos I. Margarisb,1

    nd Ltd., Department of Power Electronics & Medium Voltage Drives, CH-5300 Turgi, Swersity of Thessaloniki, Department of Electrical & Computer Engineering, GR-54124 Th

    e i n f o

    ly 2011vised form 23 October 2011ctober 2011e 23 November 2011

    lering

    a b s t r a c t

    An extension of the Symmetrical Opsystems is proposed. Type-p control loloop transfer function. For designing aand so on, if the controlled process is oachieves zero steady state position anposition, velocity and acceleration erroreference signals and eliminate highera transfer function containing dominconsidered in the frequency domain. both dominant dynamics and model uthe proposed theory, simulation resuare presented./ locate / jprocont

    the design of PID type-p control

    andiki, Greece

    criterion for the design of PID type-p closed- loop controlre characterized by the presence of p integrators in the open-pe-p control loop there should exist an PIpD, or PI(p1)D, or PID-0 or type-1 or type-(p 1) respectively. A type-II control looplocity error, a type-III control loop achieves zero steady state

    therefore a type-p control loop is expected to track both fasterr errors at steady state. For deriving the proposed control law,me constants and the plants unmodelled dynamics has beennal control law consists of analytical expressions that involvetainty of the controlled process. For justifying the potential of

    representative processes met in many industry applications

  • 12 K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25

    the Magnitude Optimum criterion, introduced by Oldenbourg andSartorius [28], and is based on the idea of designing a controllerwhich will render the magnitude of the closed-loop frequencyresponse as close as possible to unity in the widest possible fre-quency range. In succession, Kessler suggested the SymmetricalOptimum criterion [29], which in reality is the application of theMagnitude Optimum criterion in type-II control systems. The nameof the Symmexhibited bity, the Symbut the appcontrol syst

    The desimum and Sand Kesslertages: (1) ththe setpoinHowever, etude Optim

    For the dclear presencontrollers Section 3. Btype-III contheory is exknowledge is provided 6), represenare employedominant t

    2. Denitio

    Accordine(s) = r(s) function T(s

    T(s) = bmsm

    ans

    then the res

    e(s) =(

    ans

    an

    where cj = aand if e(s) is

    e() = lims0

    s

    If r(s) = 1/s t

    e() = lims0

    which becoS(s) = y(s)/dby

    T(s) = smbmsnan

    S(s) = s [ans b2) +

    3 S(s) standsby S(s) = y(s)/d

    respectively. If (5) and (6) hold by, the closed-loop control systemis said to be of type-I. In similar fashion, if, then the velocity erroris equal to,

    e() = lims0

    (ansn + + cmsm + + c1s + c0ansn + an1sn1 + + a1s + a0

    )1s

    (7)

    becoerror

    s(t) =

    com-loop-loop

    smbmsna

    s2an

    ing tf type fo

    (ansn

    bmsm

    ansn

    tivelteriz6), (1he adate h

    con

    us nos), ythe inance

    ry ap

    Tms(

    Tm iant t26]. L

    vectnsta

    is thrquere, t

    f the nt. Fior co, kp s

    : statetrical Optimum criterion comes from the symmetryy the open-loop frequency response [29,30]. In real-metrical Optimum criterion is not something different,lication of the Magnitude Optimum criterion to type-IIems.gn of control systems both with the Magnitude Opti-ymmetrical Optimum criteria of Oldenbourg-Sartorius

    respectively presents at least two important advan-ey do not require the complete plant model [30] and (2)t response of the closed-loop system is satisfactory [2].xcluding the German bibliography [3336], the Magni-um criterion is rarely referred today [30].evelopment of the proposed theory and for the sake of atation of this work, the conventional tuning of PID typevia the Symmetrical Optimum criterion is presented inased on this principle, in Section 4 the design of PIDtrol loops is presented. Finally, in Section 5 the proposedtended to the design of type-p control loops. Preliminaryregarding the denition of the type of the control loopsin Section 2. For testing the proposed theory (see Sectiontative processes met over the literature and industryd, processes with time delay, processes with equivalent

    ime constants and non-minimum phase processes.

    ns and preliminaries

    g to Fig. 1 the error e(s) is given byy(s) = [1 T(s)]r(s) = S(s)r(s) . If the closed-loop transfer) = y(s)/r(s) is dened by

    + bm1sm1 + + b1s + b0n + an1sn1 + + a1s + a0

    (1)

    ulting error e(s) is dened by

    n + + cmsm + + c1s + c0sn + an1sn1 + + a1s + a0

    )r(s) (2)

    j bj (j = 0, . . ., m). According to the nal value theorem stable, e() is equal to(

    ansn + + c2s2 + c1s + c0ansn + an1sn1 + + a1s + a0

    )r(s). (3)

    hen(c0a0

    ), (4)

    mes zero when c0 = 0, or when a0 = b0 . Hence, sensitivityo(s)3 and closed-loop transfer function T(s) are dened

    + + s2b2 + sb1 + a0+ + s2a2 + sa1 + a0

    , (5)

    n1 + an1sn2 + + (am bm)sm1 + (am1 bm1)sm2 + s(a2 snan + sn1an1 + + s2a2 + sa1 + a0

    for the sensitivity of the closed-loop control system and is denedo(s) when r(s) = nr(s) = di(s) = nr(s) = 0, Fig. 1.

    whichof the

    limt

    evs

    and beclosedclosedtively,

    T(s) =

    S(s) =

    Accordto be ohave th

    S(s) = sp

    and

    T(s) =

    respeccharacS, see (have telimin

    3. The

    Let e(s), u(error, disturbindust

    G(s) =

    wheredominstant [used intime cowhichload tothermoloop oconstaIf vectdrives)

    4 SFOC a1 b1] (6)

    mes nite if c0 = 0 or a0 = b0. As a result the nal value is given by

    lims0

    (c1a0

    )= lim

    s0

    (a1 b1

    a0

    )(8)

    es zero when c1 = 0 or when a1 = b1. In that case, the control system is said to be of type-II. Sensitivity S and transfer function T take the following forms respec-

    + sm1bm1 + + sa1 + a0n + sn1an1 + + sa1 + a0

    , (9)

    sn2 bmsm2 bm1sm3 + a2 b2snan + sn1an1 + + s2a2 + sa1 + a0

    . (10)

    o the above analysis, a closed-loop control system is saide-p when sensitivity S and complementary sensitivity Tllowing formp + an1sn1p bmsmp bm1sm1p + ap bp)

    snan + sn1an1 + + s2a2 + sa1 + a0(11)

    + + apsp + ap1sp1 + + a1s + a0+ + apsp + ap1sp1 + + a1s + a0

    (12)

    y. Also, one could argue that type-p control loops areed by the order of zeros at s = 0 in the sensitivity function0) and (11). According to the above, type-p control loopsvantage of tracking fast reference signals, since theyigher order errors.

    ventional Symmetrical Optimum design criterion

    w consider the closed-loop system of Fig. 1, where r(s),(s), do(s) and di(s) are the reference input, the controlput and output of the plant, the output and the inputs respectively. An integrating process met in manyplications can be dened by (13)

    11 + Tp1s)(1 + Tps)

    , (13)

    s the integrators plant time constant, Tp1 the plantsime constant and Tp the process parasitic time con-et it be noted that such type of modelling is frequentlyor controlled induction motor drives. More specically,nt Tm stands for the mechanical subsystem of the motore mechanism that involves the electromagnetic and, the difference of which, makes the shaft rotating. Fur-ime constant Tp1 is involved in the inner current controlelectrical drive and represents the stator winding timenally, Tp stands for the motors unmodelled dynamics.ntrol4 is to be followed (control of induction motortands for the pulse width modulators gain (kPWM) which

    or eld oriented control; RFOC: rotor eld oriented control.

  • K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25 13

    Fig. 1. Block d the care the output ls at thplants dc gain

    is supposed(0 1 p.u.)back path othe sequel,

    Back to F

    C(s) = (1 +Tis

    is adopted. design metcontrollersapplied, becble. This is jform

    C(s) =Tis(1

    is applied, t

    T(s) =TiTm

    where TpT

    T(s) =TiTm

    From (17), missing, see

    C(s) = 1 Tis(1

    is employethe convencellation mconstant Tp

    T(s) =TiTm

    which is uAppendix B

    Assuminrately measby (14), T

    5 In many in

    etricaes th

    TiTm

    gnit

    =

    nom

    (TiT

    +[(

    com

    Tn

    23) a

    8T3

    s

    lly a in

    8s3

    specta) ansed-in thiagram of the closed-loop control system. G(s) is the plant transfer function, C(s) is and input disturbance signals respectively and nr(s) and no(s) are the noise signa

    and kh is the feedback path.

    to remain constant all over the whole operating range regarding output frequency.5 Parameter kh is the feed-f the output measurement and as it will be proved inkh should satisfy condition kh = 1.ig. 1, for controlling (13), the PID controller dened by

    Tns)(1 + Tvs)(1 + Tcs)

    (14)

    For its tuning, the conventional Symmetrical Optimumhod is employed. Time constant Tc stands for the

    parasitic dynamics. If Tn = Tv = 0, I control cannot beause the closed-loop transfer function becomes unsta-ustied as follows. If for controlling (13), I control of the

    1 + Tcs)

    (15)

    hen the closed-loop transfer function is given by

    kps2(1 + Tp1s)(1 + Ts) + khkp

    (16)

    c 0 and T = Tp + Tc. From (16) it is evident that

    kpTp1Ts

    4 + TiTm(Tp1 + T)s3 + TiTms2 + khkp. (17)

    it is clear that T(s) is unstable since the term of s is Appendix B. In similar fashion, if PI control of the form

    + Tns + Tcs)

    (18)

    d, then for determining controller parameter Tn viational Symmetrical Optimum criterion, pole-zero can-

    Symmbecom

    T(s) =

    The ma

    |T(j)|

    The de

    D() =

    and bewhen

    kh = 1,

    Using (

    T(s) =

    or naresults

    T(s) =

    The reFig. 2(the clo43.4% ust take place, Tn = Tp1 . Therefore, the dominant time1 has to be evaluated and in that case, T(s) becomes

    kpTs4 + TiTmTs3 + TiTms2 + khkp

    , (19)

    nstable again for the same reason as for (17), see.g again that the dominant time constant Tp1 is accu-ured and considering a PID controller as that describedv = Tp1 is set (pole-zero cancellation, conventional

    dustry applications kp stands for the plants dc gain at steady state.

    frequency dfrequency rfrequency

    try of the

    where its slc = 1/(2T

    Fol(s) = 1

    8s

    In order toence input The great ozero of the ontroller transfer function, r(s) is the reference signal, do(s) and di(s)e reference input and process output respectively. kp stands for the

    l Optimum design). The closed-loop transfer functionen equal to

    kpTns + kpTs3 + TiTms2 + khkpTns + khkp

    . (20)

    ude of (20) is given by

    kpkp[1 + (Tn)2](kpkh TiTm2)2 + 2(kpkhTn TiTmT2)2

    . (21)

    inator of (21) is equal to

    p1T)26 + TiTp1 (TiTp1 2kpkhTnT)4

    kpkhTn)2 2kpkhTiTp1 ]2 + k2pk2h (22)

    es minimum, see [30,31], in the lower frequency range

    = 4T, Ti = 8kpkhT2

    Tm, Tv = Tp1 . (23)

    long with (20) results in

    1 + 4Ts + 8T2

    s + 4Ts + 1

    (24)

    fter normalizing the frequency by substituting s = Ts

    1 + 4s+ 8s2 + 4s + 1 . (25)

    ive step and frequency response of (25) are shown ind (b). From there, it is clear that the step response ofloop control system exhibits an undesired overshoot ofe time domain Fig. 2(a), and a peak overshoot in the

    omain Fig. 2(b). This is also justied by the open-loopesponse Fig. 3 where the phase margin in the crossoverc = 1/(2T) is m 35 < 45. Note also the symme-critical frequencies

    (1

    4T, 1T

    )exhibited by |Fol(j)|

    ope is equal to 1/deg around the crossover frequency), Fig. 3. The open loop transfer function is given by

    + 4s2(1 + s)

    . (26)

    overcome the obstacle of 43.4% overshoot, the refer-is ltered by adding an external controller Cex(s), Fig. 4.vershoot of the step response in (24) is owed to thetransfer function, N(s) = 1 + 4s. This can be removed by

  • 14 K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25

    Fig. 2. Type-II closed-loop control system. (a) The effect of the two degrees of freedom controller to the step response of the closed-loop control system. Step response (solidblack) and ltered step response (dotted black). (b) The effect of the two degrees of freedom controller to the frequency response of the closed-loop control system.

    Fig. 3. Ty

    including thexternal lt

    Cex(s) = r(r(

    is chosen, tnoted that tdynamics, cin the refer

    4. Extenditype-III con

    Accordintems, a simcontrol sys

    integrating process of the form (13) is assumed, where Tp1 standsfor the dominant time constant of the process and Tm, Tp stand forthe integrators time constant and the unmodelled plant dynam-

    pectited, t

    (1 +Tis2(

    Tc1red ting t

    the c

    TiTm

    T =

    Fig. 4. Two dethe closed-loope-II closed-loop control system. Open loop frequency response.

    ics resevalua

    C(s) =

    wherecompaassumtion of

    T(s) =

    whereat zero as a pole in the reference lter. In that, if aner of the form

    s)s)

    = 11 + 4s (27)

    he overshoot decreases from 43.4% to 8.1%. Let it behe rise time increases from trt = 3.1T to trt = 6.6T. Suchan for sure be improved by adding additional dynamicsence lter.

    ng the Symmetrical Optimum design criterion totrol loops

    g to the design of type-II closed-loop control sys-ilar methodology for the design of type-III closed-looptems will be proposed. For the following analysis, an

    |T(j)| =

    where A0 =dened by

    D() = (TiT+ k

    + (k

    One way toj = 2, 4, 6, . .

    grees of freedom controller. Controller Cex(s) lters the reference input so that the undesirp transfer function T(s) and not the output and input disturbance transfer functions So(s)vely. Supposing that the dominant time constant Tp1 ishe proposed I-PID controller is dened by,

    Tns)(1 + Tvs)(1 + Txs)1 + Tc1 s)(1 + Tc2 s)

    (28)

    , Tc2are known and sufciently small time constants

    o Tp1 . By setting Tx = Tp1 (pole-zero cancellation) andhat Tc = Tc1 + Tc2 , Tc1 Tc2 0, the transfer func-losed-loop control system is equal to

    kpTnTvs2 + kp(Tn + Tv)s + kpTs4 + TiTms3 + kpkhTnTvs2 + kpkh(Tn + Tv)s + kpkh

    ,

    (29)

    Tc + Tp . The magnitude of (29) is given by

    k2p(1 TnTv2)2 + k2p(Tn + Tv)22[TiTmT4 + kpkh(1 TnTv2)]2 + A02

    (30)

    [kpkh(Tn + Tv) TiTm2]2. The denominator of (30) is

    mT)28 + TiTm(TiTm 2kpkhTnTvT)6

    pkh[2TiTmT 2(Tn + Tv)TiTm + kpkhT2i T2m]4

    pkh)2(T2n + T2v )2 + (kpkh)2. (31)

    optimize the magnitude of (30) is to set the terms of j, ., in (31), equal to zero, starting again from the lowered overshoot at the output y(s) is diminished. Controller Cex(s) affects = y(s)/do(s), Si(s) = y(s)/di(s).

  • K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25 15

    Fig. m. (b)

    Fig. 6. Ope

    frequency rleads to

    Ti =2kpkhT

    Tm

    In similar fathe aid of (3

    4T2 4(TnIf Tv = nT

    Tn = 4(n n 4

    Proper selefeasible I-Pclosed-loop

    T(s) =8n(n

    Normalizinwhich nal

    T(s) =8n(n

    Note that tterms of s

    j,2. The respeferent value

    6 thn is

    4n

    m Fintaryeter and m

    everncy r). Sinin th), whex(s)

    endi con5. (a) Step response and disturbance rejection of type-III closed-loop control syste

    n-loop frequency response of type-III closed-loop control system.

    6

    in Fig. functio

    Fol(s) =

    Froplemeparamn = 4.1 by. Forfreque(50%shoot Fig. 5(alter C

    5. Exttype-pange. Setting kh = 1 and the term of equal to zero

    nTvT . (32)

    shion, setting the term of 4 equal to zero along with2), leads to

    + Tv)T + TnTv = 0. (33)

    is chosen, then (33) becomes

    1)T. (34)

    ction of parameter n (n > 4 must hold by) leads to aID control law. Substituting Eqs. (32) and (34) into the

    transfer function results in

    4n(n 1)T2s2 + (n2 4)Ts + n 4

    1)T4s4 + 8n(n 1)T3

    s3 + 4n(n 1)T2

    s2 + (n2 4)Ts + (n 4)

    .

    (35)

    g again the time by setting s = sT where s = ju, (s = j)ly leads to u = T, (35) becomes equal to

    4n(n 1)s2 + (n2 4)s + (n 4) 1)s4 + 8n(n 1)s3 + 4n(n 1)s2 + (n2 4)s + (n 4)

    . (36)

    he control loop dened in (36) is of type-III, since the j = 0, 1, 2, are equal, a0 = b0, a1 = b1, a2 = b2, see Sectionctive step and frequency responses of (36) for two dif-s of parameter n, are presented in Fig. 5. In addition,

    Accordinsis for tuninregarding tp stands forTherefore, l

    G(s) =Tmsq

    consisting constant. Adened by Tics by Tsk (kgenerality w

    nsk=1

    (1 + Tsk s

    where Tstime constacontrol systin Section 4

    C(s) =nm

    j=

    Thus, accorcontroller hconstants (achieved. Min order thtion is a ful Frequency response of type-III closed-loop control system.

    e open-loop frequency response is shown. Its transfergiven by

    (n 1)s2 + (n2 4)s + n 48n(n 1)s3(1 + s)

    . (37)

    g. 6 it is concluded that the magnitude of the com- sensitivity |T(ju)| is practically independent of then. Moreover, sensitivity |S(ju)| becomes maximum ifinimum, if n = 7.46. In that case (n = 7.46), Tn = Tv holds

    y other value of parameter n, the shape of the open-loopesponse is preserved exactly as presented in Fig. 5(a),ce the phase margin is m = 35 < 45, an undesired over-e step response of the closed-loop system is expected,ich can be decreased along with the aid of an external

    as mentioned in Section 3.

    ng the Symmetrical Optimum design criterion totrol loops

    g to the analysis presented in Section 4 a similar analy-g the PID type controllers parameters will be presentedhe design of type-p control loops. Note that parameter

    the free integrators of the open-loop transfer function.et the process be dened by

    1nmj=1(1 + Tmj s)

    nsk=1(1 + Tsk s)

    , (38)

    of q integrators and Tm one of the integrators timessuming that the plants dominant time constants aremj (j = 1, 2, . . ., nm) and the process unmodelled dynam-

    = 1, 2, . . ., ns) we can substitute in (38), without loss ofith the approximation

    ) = 1 + Tss (39)

    =ns

    k=1Tsk stands for the process small unmodelled

    nts. Since the target of the design is the nal closed-loopem to be of type-p, according to the analysis presented, the proposed PID type controller is given by

    1(1 + Tmj s)p1

    r=1 (1 + Tnr s)Tispq

    ncz=1(1 + Tcz s)

    . (40)

    ding the design of type-II, III control loops, the PID typeas to contain nm zeros equal to the Tmj dominant timej = 1, 2, . . ., nm) so that exact pole-zero cancellation isoreover, it is proved after some calculus in T(s), that

    e denominator of the nal closed-loop transfer func-l polynomial in terms of the sj coefcients, p 1 zeros

  • 16 K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25

    must exist. Furthermore, the controller must introduce pq integra-tors, so that the nal closed-loop is of type-p. Finally, in order thecontroller transfer function is strictly causal, denominators ordermust be greater or equal to p 1 + nm. The unmodelled controllersdynamics a

    ncz=1

    (1 + Tcz s

    where

    Tc =nc

    z=1T

    In that case

    Fol(s) = kpk

    or by substi

    Fol(s) = kpk

    where T =transfer fun

    T(s) =TiTm

    After some

    T(s) = bap+1

    where

    bp1 =p1j=1

    T

    b2 = kpp1

    i /= j=

    and

    ap+1 = TiTm

    a3 = kpkhi /=

    a1 = kpkhpi=

    According t

    kh = 1. Since the athe magnituFor every o

    Ti = 2kpkhT

    T

    This can beconstant de

    loop is of type-II p = 2, the PID type controller (according to theSymmetrical Optimum criterion) is given by

    C(s) = (1 + Tn1s)(1 + Tn2s)T

    , (54)

    ich T, the

    kpk

    e clo

    TiTm

    ing tonsta

    a1a3

    as a]. Aft

    pkhT

    T

    Tn1 =ordinenl loop

    (1 +

    ing p)(1ecom

    kpk

    ore t

    TiTm

    ing tted v

    a2a4.

    , aftent is

    pkhT

    T

    imilad by , the

    (1 +

    ing ting t

    2kpkre represented by

    ) = 1 + Tc s (41)

    cz . (42)

    , the open-loop transfer function becomes

    hG(s)C(s) = kpkhp1

    r=1 (1 + Tnr s)TiTmsp

    nsk=1(1 + Tsk s)

    ncz=1(1 + Tcz s)

    (43)

    tuting (39), (40), (41) and (42) results in

    h

    p1r=1 (1 + Tnr s)

    TiTmsp(1 + Ts)(44)

    Ts + Tc and TsTc 0. Finally, the closed-loopction is equal to

    kpp1

    r=1 (1 + Tnr s)Tsp+1 + TiTmsp + kpkh

    p1r=1 (1 + Tnr s)

    . (45)

    calculus in (45) it is concluded that

    p1sp1 + bp2sp2 + + b3s3 + b2s2 + b1s + b0sp+1 + apsp + ap1sp1 + + a3s3 + a2s2 + a1s + a0

    (46)

    pj = Tp1Tp2 . . . Tpp1 , b3 = kpp1

    i /= j /= k=1TniTnj Tnk , (47)

    1

    TniTnj , b1 = kpp1i=1

    Tni , b0 = kp, (48)

    T, ap = TiTm, (49)p1

    j /= k=1TniTnj Tnk , a2 = kpkh

    p1i /= j=1

    TniTnj , (50)

    1

    1

    Tni , a0 = kpkh. (51)

    o (A.9), if a0 = b0 then

    (52)

    im is to determine parameters Ti, Tnr (r = 1, . . . , p 1)de of (46) will be optimized according to Appendix A.

    rder p, the optimal integral gain is given by

    m

    p1r=1

    Tnr . (53)

    proved as follows. For a process of one dominant timened by (13) where (q = 1), then in order the nal control

    for whIn that

    Fol(s) =

    and th

    T(s) =

    Accordtime c

    a22 = 2

    is set, [30,31

    Ti = 2k

    and if Acc

    stant dcontro

    C(s) =

    Assum(1 + sTFol(s) b

    Fol(s) =

    Theref

    T(s) =

    Accordcalcula

    a23 = 2

    Finallyconsta

    Ti = 2k

    In sdenetype-p

    C(s) =

    Accordregard

    Tik1 =is(1 + Tcs)

    n2 = Tp1 and (1 + sTp)(1 + sTc) 1 + sT have been set. open-loop transfer function is given by

    h(1 + Tn1s)

    TiTms2(1 + Ts)(55)

    sed-loop transfer function is then given by

    kp(1 + Tn1s)s2(1 + Ts) + kpkh(1 + Tn1s)

    . (56)

    o the analysis presented in Section 3, the integratorsnt is calculated if

    (57)

    nother means of optimizing the magnitude of (56),er some calculus it is obtained

    mTn1 (58)

    4T then Ti = 8kpkh(T2/Tm), see Section 3.g to Section 4, for a process of one dominant time con-

    ed again by (13) where (q = 1) then in order the nal is of type-III p = 3, the PID type controller is given by

    Tn1s)(1 + Tn2s)(1 + Tn3s)Tis2(1 + Tcs)

    . (59)

    again pole-zero cancellation, Tn3 = Tp1 and + sTc) 1 + sT the open-loop transfer functiones

    h(1 + Tn1s)(1 + Tn2s)

    TiTms3(1 + Ts). (60)

    he closed-loop transfer function is equal to

    kp(1 + Tn1s)(1 + Tn2s)s3(1 + Ts) + kpkh(1 + Tn1s)(1 + sTn2 )

    . (61)

    o (57) and since n = 2, the integrators time constant isia

    (62)

    r some calculus it was shown that the integrator timeequal to

    mTn1Tn2 . (63)

    r fashion, for a process of one dominant time constant(13) and if n = k 1, in order the nal control loop is ofPID type controller is given by

    Tn1s)(1 + Tn2s) . . . (1 + Tnk s)Tisk1(1 + Tcs)

    . (64)

    o the analysis presented previously, it can be claimedhe integrators time constant Tik1 , that

    hTTm

    k1j=1

    Tnj . (65)

  • K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25 17

    Therefore, for n = k, it has to be proved that

    Tik = 2kpkhTTm

    kTnj

    = 2kpk

    According ttroller is giv

    C(s) = (1 +

    for which Tcellation. Sclosed-loop

    Fol(s) = kpk

    T(s) =TikTm

    or

    T(s) =Tik TmT

    respectively

    a2k+1 = 2ak+or

    (TikTm)2 = 2

    or TikTm = obtained

    Tik = 2kpk

    = 2kpk

    which is eqresults in

    T(s) =2kpkhT

    For determithe magnituconstants T

    4T2 p3

    i=1

    This is justing parameresults in

    k2pT2n1

    = 2kpor nally

    Tn1 4T =

    In similar fashion, in type-III control loops for determining parame-ters Tn1 , Tn2 we make use of a

    22 2a3a1 + 2a4a0 = 0, see (A.11). This

    results in

    2 Tn2 2 2

    lly,

    4Ting t

    is of follo

    2ak

    ak+1a

    es thes.reforsed-

    iTmT

    it waain

    2rkhich

    4Tr

    then

    rk

    then2kpkhe obt

    2rk

    r som

    2 4

    p2i=1

    ove e

    lly

    pi=

    is truiousetersditio

    e-V c

    n2 + T

    n1Tn

    Tn2Tnj=1

    hTTm

    k1

    j=1Tnj

    Tnk = Tik1Tnk

    . (66)

    o the design of type-p control loops, the PID type con-en by

    Tn1s)(1 + Tn2s) . . . (1 + Tnk s)(1 + Tnk+1s)Tisk(1 + Tcs)

    (67)

    nk+1 = Tp1 is set, assuming design via pole-zero can-ince again (1 + sTp)(1 + sTc) 1 + sT, the open and

    transfer functions are given by

    h

    kj=1(1 + Tnj s)

    TikTmsk+1(1 + Ts)

    , (68)

    kpk

    j=1(1 + Tnj s)sk+1(1 + Ts) + kpkh

    kj=1(1 + Tnj s)

    , (69)

    kp(rksk + rk1sk1 + + r2s2 + r1s + 1)sk+2 + Tik Tmsk+1 + kpkh(rksk + rk1sk1 + r2s2 + r1s + 1)

    (70)

    . Then, according to (57), Ti is calculated by

    2ak (71)

    kpkhTikTmTrk (72)

    2kpkhTrk. Finally, along with the aid of (70), it is

    hTTm

    rk = 2kpkhTTm

    kj=1

    Tnj

    hTTm

    k1

    j=1Tnj

    Tnk = Tik1Tnk

    (73)

    ual to (66). In that case, if (73) is substituted into (69),

    kpp1

    r=1 (1 + Tnr s)2p1

    r=1Tnr sp+1 + 2kpkhT

    p1r=1Tnr s

    p + kpkhp1

    r=1 (1 + Tnr s). (74)

    ning now parameters Tnr , it will be shown that in orderde of (74) satises condition |T(j) 1|, controller time

    nr must satisfy condition

    Tni 4T p2

    i=1Tni +

    p1i=1

    Tni = 0. (75)

    ied as follows. In type-II control loops for determin-ter Tn1 we make use of a

    21 2a2a0 = 0 (see (A.10)). This

    (2kpTn1T) (76)

    0. (77)

    4TTn1

    or na

    Tn1Tn2

    Accordsystemk), the

    a2k1 =

    a2k = 2the onquenci

    Thethe clo

    T(s) =T

    In (72)we obt

    k2pr2k1

    from w

    rk1 If n = k,

    rk 4TIf n = k TiTm = (70) w

    k2pr2k

    or afte

    4T2rk

    or

    4T2

    The ab

    4T2Tnp

    or na[4T2

    which

    Obvparaming conin

    Typ

    4(Tn1T

    4(T+ Tn1 4TTn1Tn2 (Tn1 + Tn2 ) + Tn1Tn2 = 0 (78)

    (Tn1 + Tn2 ) + 4T2 = 0. (79)o the above, and based on (65) if the closed-loop control

    type-p, then for determining parameters Tnj (j = 1, 2, . . .,wing optimization conditions are claimed to be,

    2ak 2ak3ak+1, (80)

    k1 (81)

    at satisfy condition |T(j) 1| in a wide range of fre-

    e, if n = k 1 then controller C(s) is dened by (64) andloop transfer function is given by

    kp(rk1sk1 + rk2sk2 + + r2s2 + r1s + 1)sk+1 + TiTmsk + kpkh(rk1sk1 + rk2sk2 + + r2s2 + r1s + 1)

    . (82)

    s shown that TiTm = 2kpkhTrk. By applying (80) to (82)

    2kp(2kpTrk1) + 2kprk3(2kpTrk1)T = 0, (83) after some calculus results in

    k2 + 4T2rk3 = 0. (84) we are going to show that

    1 + 4T2rk2 = 0. (85) the closed-loop transfer function is given by (70). SinceTrk then by applying a2k = 2ak1ak+1 2ak2ak+2 toain

    1kp(2kpTrk) + 2kprk2(2kpTrk)T = 0 (86)e calculus

    Trk1 + rk = 0 (87)

    Tni 4T p1

    i=1Tni +

    pi=1

    Tni = 0. (88)

    quation is rewritten in the form of

    p3i=1

    Tni 4TTnp p2

    i=1Tni + Tnp

    p1i=1

    Tni = 0 (89)

    3

    1

    Tni 4T p2

    i=1Tni +

    p1i=1

    Tni

    ]Tnp = 0 (90)

    e, since (83) holds by.ly, the number of combinations of the Tni optimal

    that satisfy (90) is innite. More specically, by apply-n (90) for the design of up to type-V control loops results

    ontrol loops:

    n1Tn3 + Tn1Tn4 + Tn2Tn3 + Tn2Tn4 + Tn3Tn4 )T22Tn3 + Tn1Tn2Tn4 + Tn2Tn3Tn4 + Tn1Tn3Tn4 )T3Tn4 = 0. (91)

  • 18 K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25

    ed-lo

    Type-IV

    4(Tn1 + Tn2= 0.Type-III

    4T2 4(Tn1Type-II c

    Tn1 = 4T.Note that (In similar fsimplicity o

    Tn1 = Tn2 =the respectare given by

    Fol(s) 2np

    and

    T(s) 2np

    The optimaloop we waconsequent

    Type-V c

    n2(n2 16nType-IV

    nn2 12n +Type-III

    (n2 8n + 4With respecPID type conTherefore, i

    Tn1 = Tn2 =according t

    Tn3 =4n(n

    n2 Based on thtions are gi

    [4n3(n

    a5s5

    n3(n

    n2(nn

    n3(n

    n3T3

    n2(nn

    ing aj = bl sysntroncy re typeter mete

    freqncy re th

    S ben2 =s zerlar responFig. 7. Type-IV control loop. (a) Step and (b) frequency response of the nal clos

    control loops:

    + Tn3 )T2 4(Tn1Tn2 + Tn2Tn3 + Tn1Tn3 )T + Tn1Tn2Tn3(92)

    control loops:

    + Tn2 )T + Tn1Tn2 = 0. (93)ontrol loops:

    (94)

    93) and (94) are equal to (18) and (33) respectively.ashion with type-III control loops and for the sake off the analysis, if we choose

    = Tnp1 = nT (95)ive open Fol(s) and closed-loop T(s) transfer functions

    (1 + nTs)p11Tp

    sp(1 + Ts)

    (96)

    (1 + nTs)p11Tp+1

    sp+1 + 2np1Tp

    sp + (1 + nTs)p1

    . (97)

    l value of parameter n depends on the type of the controlnt to design. If we substitute (95) into (92)-(94), we havely,ontrol loops:

    + 24)T4 = 0 nopt = 14.32. (98)control loops:

    12)T3 = 0 nopt = 10.89. (99)control loops:

    )T2 = 0 nopt = 7.46. (100)

    Fol(s) =

    T(s) =

    where

    b3 = 4

    b1 = 2

    and

    a5 = 8

    a3 = 4

    a1 = 2

    Accordsince, controtype cofrequeT of thparamof paraS in thefreque

    NotsitivityTn1 = Ttrollern. Simistep re

    t to the above, for the design of a type-IV control loop, atroller of three zeros in its transfer function is required.

    f we chose

    nT (101)

    o (95), then from (92) it is obtained

    2)8n + 4T =

    4n(n 2)(n 0.536)(n 7.464)T. (102)

    e above, the corresponding Fol(s) and T(s) transfer func-ven by

    an oversho( = 32 < 45For decreastem, the twbe exploiteterms of tim

    T(s) =D1(s

    where

    N1(s) = (1 +op control system for various values of parameter n.

    2)T3s3 + n2(n2 12)T2

    s22n2(n 6)Ts + (n 0.536)(n 7.464)]

    8n3(n 2)T5s5 + 8n3(n 2)T4

    s4

    .

    (103)

    b3s3 + b2s2 + b1s + b0+ a4s4 + a3s3 + a2s2 + a1s + a0

    (104)

    2)T3, b2 = n2(n2 12)

    n 2 T2 (105)

    6) 2 T, b0 =

    (n 0.536)(n 7.464)n 2 (106)

    2)T5, a4 = 8n3T4 (107)

    , a2 = n2(n2 12)

    n 2 T2 (108)

    6) 2 T, a0 =

    (n 0.536)(n 7.464)n 2 . (109)

    to (104), the closed-loop control system is of type-IVj, j = 0, 1, 2, 3, see Section 2. If n < 7.464 the closed-looptem is unstable. As a result, for having a feasible PIDl law, n > 7.464 has to hold by, see (105). In Fig. 7(b) theesponse of sensitivity S and complementary sensitivitye-IV closed-loop is presented for several variations ofn, n [7.5, ). From there, it is apparent that variationsr n do not lead to critical variations of both functions T,uency domain. Sensitivity S is affected only in the loweregion.at, in similar fashion with type-III control loops, sen-comes minimum when all controller zeros are equal,

    Tn3 , n = 10.89, Fig. 10. There, it is shown how the con-os are affected in case of variations in design parametersults are also observed in the time domain, Fig. 7(a). These of the type-IV closed-loop control system exhibits

    ot of 50%, which is justied by the phase margin) of the open-loop Fol(s) frequency response, Fig. 8.ing the overshoot of the nal closed-loop control sys-o degrees of freedom controller structure will again

    d. If n = 10.89, then the closed-loop transfer function ine constants form is given by

    N1(s))D2(s)D3(s)

    (110)

    10.89Ts)3, D1(s) = (1 + 2.3Ts) (111)

  • K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25 19

    Fig. 8. Open-lparameter n.

    Fig. 9. The effresponse of th

    Fig. 10. Variatn.

    D2(s) = (2.2

    D3(s) = (14Thus, by ch

    Cex(s) =(1

    overshoot i

    6. Simulation results

    For justifying the controls law potential simulation examplesof type-II, III, IV, V control loops are presented. According to the

    l lawlling

    (1 +Tis

    hreents ont Tcontrocontro

    C(s) =

    In all tconstaconstaoop frequency response of a type-IV control loop for various values of

    ect of the two degrees of freedom controller structure to the stepe type-IV closed-loop control system.

    ions of parameters Tn1 , Tn2 , Tn3 according to variations of parameter

    74)2T2s2 + 0.99(2.274)Ts + 1 (112)

    .75)2T2s2 + 1.9(14.75)Ts + 1. (113)

    oosing an external controller of the form

    + 2.3Ts)[(14.75)2T2s2 + 1.9(14.75)Ts + 1](1 + 10.89Ts)3(1 + Ts)

    (114)

    s reduced to 14.75%, Fig. 9.

    plants and loops are den has been is calculatethree cases

    6.1. Process

    The proc

    G(s) =(1 +

    is considereloop controis decreaseCex1 (s), Fig.Note that dexternal coof the contrif Cex2 (s) =the overshosystem is oboth ramp

    6.2. Process

    A delay

    G(s) =(1 +

    is assumed not take intthis exampis also testthe controluse of bothrespectivelyCex2 (s) is ofcontrol signFigs. 11(b) a

    6.3. A non-

    Althougexistence oprocess of t

    G(s) =(1 + presented in Section 4 the I-I-PID type controller for a type-0 process is given by

    Tn1s)(1 + Tn2s)(1 + Tn3s)3(1 + Tc1 s)(1 + Tc2 s)

    . (115)

    examples, it is assumed that the sum T of all timef the controlled process is accurately measured. Time=kj=1Tpj + Tc and Tc = Tc1 + Tc2 includes both

    controllers unmodelled dynamics. Since type-III controlsigned Tn1 = Tp1 , Tn2 = 4(n1)n4 T, Tn3 = nT. Parameterchosen equal to n = 7.46. The integrators time constantd through Ti = 2kpkh TTm

    p1r=1Tnr = 2kpkhTn2Tn3T. In all

    Tm = 1 has been set.

    with dominant time constants

    ess described by

    2 s)(1 + 0.84s)(1 + 0.78s)(1 + 0.57s)(1 + 0.28s)

    (116)

    d. From Fig. 11(a) it is apparent that the type-III closed-l system exhibits an undesired overshoot of 87.4% whichd by ltering the reference with an external controller

    11. Settling time remains almost unaltered, tss = 143.isturbance rejection has remained the same since thentroller Cex1 (s) acts only at the reference signal outsideol loop. For manipulating the overshoot of the output,

    1(tn2 tn3 )s

    2+(tn2+tn3 )s+1reference lter is to be used, then

    ot is decreased to 6.2%. Since the closed-loop controlf type-III, the output of the process can track perfectlyand parabolic reference signals, Fig. 12.

    with time delay

    process of the form

    2 s)(1 + 0.99s)(1 + 0.57s)(1 + 0.28s)(1 + 0.1s) e

    s

    (117)

    in this example. Note that the proposed control law doeso account the effect of the time delay and therefore inle the robustness of the method to model uncertaintiesed. If no external lter is used for reference tracking,

    loop exhibits an overshoot of 100.4%, Fig. 13(a). The Cex1 (s), Cex2 (s) eliminates the overshoot to 9.4% and 0%, Fig. 13(a). Disturbance rejection remains unaltered.

    the same form as in the previous example. Note thatal u() is improved in case the reference is ltered,nd 13(b).

    minimum phase process

    h the proposed theory does not take into account thef zeros in the process model, a non-minimum phasehe form

    1.34(1 0.771s) s)(1 + 0.33s)(1 + 0.12s)(1 + 0.056s)(1 + 0.038s)

    (118)

  • 20 K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25

    Fig. 11. Type- em an = 250. Extern ersho

    and output dis

    Fig. 12.

    Fig. 13. Type- = 250. Extern

    and output dis

    is adopted The step reFig. 15(a) acontrol signis used, theundesirableis reduced ttion remaininto Si(s) = y

    6.4. Contro

    For testieter uncertaIII closed-loop control system. (a) Step response of the output of the control systal lter of the form Cex(s) = 1(0.45tn2 tn3 )s2+(tn2 +0.45tn3 )s+1 is used for decreasing the ovturbance di(s) = 0.1r(s) is applied at = 500.Type-III closed-loop control system. (a) Ramp response of the closed loop control system

    III closed-loop control system. (a) Step response of the output of the control system anal lter of the form Cex1 (s) = 1(0.45tn2 tn3 )s2+(tn2 +0.45tn3 )s+1 is used for decreasing the overshoturbance di(s) = 0.1r(s) is applied at = 500.

    for testing the robustness of the proposed control law.sponse of (118) is presented in Fig. 14. In addition, innd (b) the step response of the output y() and theal u() are presented respectively. If no external lter

    overshoot of the step response is 59.9%. Since this is, if r(s) is ltered by Cex1 (s), Cex2 (s) then the overshooto 0% in both cases. Output and input disturbance rejec-

    unaltered since the external lter does not participate(s)/di(s), So(s) = y(s)/di(s) respectively.

    ller tuning without pole zero cancellation

    ng the robustness of the proposed control law to param-inties, a type-III closed loop control system is designed

    where the Therefore, pa is the erro

    G(s) =(1 +

    From Fig. 1measuring of the closedisturbanced (b) control signal u(). Output disturbance rejection is applied atot of the output. Input disturbance di(s) = 0.1r(s) is applied at = 250 and (b) parabolic response of the closed-loop control system.

    d (b) control signal u(). Output disturbance rejection is applied atot of the output. Input disturbance di(s) = 0.1r(s) is applied at = 250

    PID controller does not achieve pole-zero cancellation.arameter Tn1 is determined by Tn1 = (1 + a)Tp1 wherer when measuring Tp1 . The process is given by

    1.23 s)(1 + 0.872s)(1 + 0.367s)(1 + 0.287s)(1 + 0.11s) .

    (119)

    6(a) and (b) it is apparent that if an error of 30% whenTp1 occurs, a small change is observed in the overshootd loop control system. In addition, both input and output

    rejection remain almost unaltered.

  • K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25 21

    Fig. 14. Step response of the non-minimum phase process dened by (118).

    6.5. Comparison between a type-I and a type-III control loop

    For showing the advantages of designing a higher order fastercontrol loop, the following process

    G(s) = 1.23(1 + s)(1 + 0.992s)(1 + 0.692s)(1 + 0.139s)(1 + 0.107s)

    (120)

    is adopted. For this process, a type-I, III closed control loop will bedesigned. For designing the PID type-I control loop the conventionalMagnitude Optimum criterion (see Appendix D) is employed. Notethat for determining controllers zeros, exact pole zero cancellationhas to take place (see Appendix D) [28]. From Fig. 17 it is apparent

    that the type-I control loop fails to track both the ramp and theparabolic reference signal achieving constant non-zero steady statevelocity and acceleration error.

    6.6. A type-IV and a type-V control loop

    From the Laplace transformation it is known that if r(t) = tn thenL{y(t)} = n!/sn+1. Specically, if n = 1 then L{r(t)} = 1/s2 and the sys-tem is of type-II, or if n = 2 then L{r(t)} = 2/s3 and the system is oftype-III. For a type-IV and type-V control loop the Laplace transfor-mation of the reference signal is given if n = 3 and n = 4 for whichwe have L{r(t)} = 3 !/s3+1 and L{r(t)} = 4 !/s4+1 respectively. Accord-ing to the proposed theory for a type-IV, V control loop the proposedPID type controllers are given by

    C(s) = (1 + Tn1s)(1 + Tn2s)(1 + Tn3s)(1 + Tn4s)Tis4(1 + Tc1 s)(1 + Tc2 s)

    . (121)

    C(s) = (1 + Tn1s)(1 + Tn2s)(1 + Tn3s)(1 + Tn4s)(1 + Tn5s)Tis5(1 + Tc1 s)(1 + Tc2 s)

    (122)

    respectively. For determining parameters Tn1 , Tn2 , Tn3 , Tn4 , Ti in(121) according to the proposed theory, we set Tn4 = Tp1 and Tn1 =Tn2 = nT according to (95). For that reason, (92) becomes4(2nT + Tn3 ) 4(n2T + 2nTn3 ) + n2Tn3 = 0 (123)or nally

    Tn3 =4n(n 2)

    (n2 8n + 4)T. (124)

    Integrators time constant for the type-IV control loop is equal to

    Ti = 2kpkhTn1Tn2Tn3T. (125)

    Fig. 15. Type- se of disturbance re

    tn3 )s+1di(s) = 0.1r(s) is

    Fig. 16. Type-cancellation a III closed-loop control system for a non-minimum phase process. (a) Step responjection is applied at = 200. External lter of the form Cex1 (s) = 1(0.45tn2 tn3 )s2+(tn2 +0.45

    applied at = 200 and output disturbance di(s) = 0.1r(s) is applied at = 300.III closed-loop control system. The PID controller is tuned without pole zero cancellation= 0.the output of the control system and (b) control signal u(). Outputis used for decreasing the overshoot of the output. Input disturbance

    : a = 0.3 and a = 0.3. The PID controller is tuned via exact pole-zero

  • 22 K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25

    Fig. 17. Comp the rasteady state ve

    Fig. 18. (a) Re en chosen equal to n = 10.89 according to (99). (b) Response of the type-Vcontrol loop to g to (99).

    In similar fcontrol loonT. Accord

    4(3n2T2 + 3

    and after so

    Tn4 =4n

    n(n2

    Integrators

    Ti = 2kpkhTThe controrespective rtype-IV and

    6.7. Effect operformanc

    The effethis examp

    G(s) =(1 +

    is adopted. depends onmodels thefar from thTp is the pthe processdominant tarison between a type-I, III PID control loop. The type-I control loop fails to track locity and acceleration error is observed.

    sponse of the type-IV control loop to reference signal r(t) = t3; parameter n has be reference signal r(t) = t4; parameter n has been chosen equal to n = 14.32 accordin

    ashion, for the (122) PID type controller and since thep is of type-V, we set Tn5 = Tp1 and Tn1 = Tn2 = Tn3 =ingly, (91) becomes

    3 3 2 2 3 2nTTn4 )T 4(n T + 3n TTn4 ) + n TTn4 = 0(126)

    me calculus results in2(n 3)

    12n + 12)T =4n(n 3)

    n2 12n + 12T. (127)

    time constant for the type-V control loop is equal to

    n1Tn2Tn3Tn4T. (128)

    lled process in this example is dened by (120). Theesponse to r(t) = t3 and r(t) = t4 reference signals for the

    the type-V control loop are presented in Fig. 18.

    f the process unmodelled dynamics to the controle

    ct of the process unmodelled dynamics is discussed inle. The process dened by

    1 s)(1 + as)(1 + a2s)(1 + a3s)(1 + a4s) (129)

    As proved in Sections 4 and 5 the proposed control law pole-zero cancellation and time constant T which

    process unmodelled dynamics (poles of the processe origin), see (32) and (34) where T = Tc + Tp androcess parasitic time constant and Tc Tp . In Fig. 19

    is modelled by a = 0.15 containing a relatively largeime constant and in the next case where a = 0.6 the

    Fig. 19. Step rprocess dene

    parasitic tim

    nant time c

    a = 0.15 the

    Tp = Tp1

    is that thezeros, time(T Tpj ), tof the outp

    7. Conclus

    The Symdesign of tmp r() = and the parabolic r() = 2 reference signal since constantesponse of the PID type-III control loop when a = 0.15 and a = 0.6 for ad by (129).

    e constant of the process is comparable to its domi-

    onstant. SinceTpTp1

    =4

    j=1aj , it is apparent that when

    n Tp = Tp14

    j=1aj = 0.1764Tp1 and when a = 0.6 then

    4j=1a

    j = 1.3056Tp1 . The conclusion according to Fig. 18 less accurate the model of the process in terms of

    delay, poles compared to the dominant time constanthe poorer the performance becomes (see settling timeut and input disturbance rejection Fig. 19).

    ions and discussion

    metrical Optimum criterion has been extended for theype-p control loop. Based on the conventional tuning

  • K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25 23

    for PID type controllers via the Symmetrical Optimum principle, asimilar design technique for type-III control loops was proposed. Itwas shown that type-III control loop achieve zero steady state posi-tion, velocity and acceleration error and therefore they are able totrack faster reference signals than type-I or II control loops. Basedon this technique, the proposed control law was extended for tun-ing PID typsignals is acarried out the processunmodelleddelay constsince nowafor most indated for theapplicationpromising rthe control and process

    Acknowled

    The auththree anonypeer review

    Appendix A

    Let the c

    T(s) = bmsm

    ans

    where m (A.1) we wiThus, by set

    |T(j)|2 = ||

    or

    T(j) = N(jD(j

    Polynomial

    N(j) + b0 + j(

    and

    D(j) + a55

    or

    |D(j)|2 (

    + (a25 + 2a

    2a1a7

    + (a22 + 2a

    6 Time delay

    series esd =

    and

    |N(j)|2 (b28)16 + (b27 b8b6)14 + (b26 + 2b4b8 2b5b7)12

    + (b25 + 2b b 2b b 2b b )10 + (b2 + 2b b + 2b b

    2b1b7

    + 2b, |T(j

    |2 =

    king 2, |Nesul

    2a0

    3a1

    a1a5

    a0a8

    + 2b

    dix B via

    the i

    sTm(

    Tm, T cont

    sTi(1

    ied, t

    s2TiT

    TpT

    s4TiT

    ing tissi

    1 sTi(1

    loyedtionust t

    s4TiT

    is un, PIDoles come-p control loops so that tracking of faster referencechieved. The development of the proposed control isin the frequency domain where the transfer function of

    involves the dominant time constants and the plants dynamics. Future work deals with introducing the time

    ant6 as one more parameter in the proposed control law,days the time delay is straightforward to be measuredustrial processes. The proposed theory has been evalu-

    control of representative plants met in many industrys. The robustness of the proposed control law achievesesults also for the control of processes with parameterslaw disregards, such as non-minimum phase processeses with time delay.

    gements

    ors would like to express their greatful thanks to themous reviewers for their valuable feedback during the

    process.

    . Optimization conditions

    losed-loop transfer function be dened by (A.1),

    + bm1sm1 + + b2s2 + b1s + b0n + an1sn1 + + a2s2 + a1s + a0

    = N(s)D(s)

    (A.1)

    n. By applying the Symmetrical Optimum criterion toll force |T(s)| 1 in the wider possible frequency range.ting s = j into (A.1) and squaring |T(j)| leads to

    N(j)|2

    D(j)|2(A.2)

    ))

    = (j)mbm + + (j)2b2 + (j)b1 + b0

    (j)nan + + (j)2a2 + (j)a1 + a0. (A.3)

    s N(j) and D(j) are rewritten as follows

    + b88 b66 + b44 b22

    b77 + b55 b33 + b1) (A.4)

    + a88 a66 + a44 a22 + a0 + j( a77

    a33 + a1) (A.5)

    a28)16 + (a27 a8a6)14 + (a26 + 2a4a8 2a5a7)12

    3a7 2a2a8 2a4a6)10 + (a24 + 2a0a8 + 2a2a6 2a3a5)8 + (a23 + 2a1a5 2a6a0 2a2a4)6

    0a4 2a1a3)4 + (a21 + 2a0a2)2 + (a0)0 (A.6)

    constant can be introduced in the process model (38) by the Taylork=0

    (1k! s

    kdk).

    + (b22Finally

    |T(j)

    By ma|D(j)|range r

    a0 = b0a21 2aa22 2aa23 + 2

    a24 + 2= b24

    =

    Appentuning

    Let

    G(s) =

    where(B.1), I

    C(s) =

    is appl

    T(s) =

    where

    T(s) =

    Accordof s is m

    C(s) =

    is empconvention m

    T(s) =

    which Finallyplex pT(s) be3 7 2 8 4 6 4 0 8 2 6

    2b3b5)8 + (b23 + 2b1b5 2b6b0 2b2b4)6

    0b4 2b1b3)4 + (b21 + 2b0b2)2 + (b0)0. (A.7))|2 is equal to

    |N(j)|2

    |D(j)|2= + B4

    8 + B36 + B24 + B12 + B0 + A48 + A36 + A24 + A12 + A0

    .

    (A.8)

    equal the terms of j (j = 1, 2, . . ., n) in polynomials(j)|2 so that |T(s)| 1 in the wider possible frequencyts in

    (A.9)

    = b21 2b2b0 (A.10)+ 2a4a0 = b22 2b3b1 + 2b4b0 (A.11) 2a6a0 2a4a2 = b23 + 2b1b5 2b6b0 2b4b2 (A.12)

    + 2a6a2 2a1a7 2a3a5

    0b8 + 2b6b2 2b1b7 2b3b5 (A.13)

    . Instability of the PI control conventionalthe Symmetrical Optimum criterion

    ntegrating process be dened by

    11 + sTp1 )(1 + sTp)

    , (B.1)

    p1 , Tp have been dened in Section 3. If for controllingrol of the form

    1 + sTc)

    , (B.2)

    hen the closed loop transfer function is given by

    kp

    m(1 + sTp1 )(1 + sT) + khkp(B.3)

    c 0 and T = Tp + Tc. From (B.3) it is evidentkp

    mTp1T + s3TiTm(Tp1 + T) + s2TiTm + khkp. (B.4)

    o (B.4), it is evident that T(s) is unstable since the termng. In similar fashion, if PI control of the form

    + sTn + sTc)

    (B.5)

    , then for determining controller parameter Tn via theal Symmetrical Optimum criterion, pole-zero cancella-ake place, Tn = Tp1 . Therefore, T(s) becomes

    kp

    mTp1T + s3TiTm(T + Tp1 ) + s2TiTm + khkp, (B.6)

    stable again for the same reason as stated for (B.4). control by cancelling two real or conjugate com-of G(s) cannot be applied, since it is proved that

    es unstable for the same reason as for (B.4). This is

  • 24 K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25

    justied by the Routh theorem. For a polynomial of the formD(s) = ansn + an1sn1 + + a1s + a0, necessary and sufcient condi-tion for D(s) to be stable is aj > 0, j = 0, 1, 2, . . .. Since both in (19),(16) and (B.6), a3 = 0 and a1 = 0 then according the Routh theorem,D(s) is unst

    Appendix C

    In SectioIV, III contro

    4(Tn1Tn2

    4(Tn1+Tn1Tn2T

    4(Tn1 + T+Tn1Tn

    4T2 4(Tn1By substitu

    4(n2T2 + n2

    + n3T3 +

    4(nT + nT

    4T2 4(nTor

    24n2T4 1

    12nT3 12

    4T2 8nT2Since T /=respectively

    Appendix Dcriterion

    For cont

    G(s) =(1 +

    where T2pics, the PID

    C(s) = (1 +T

    is adopted. terion paramthe widest p

    T2 = T2psystem acco

    T(s) =sTi(1

    By forcing p

    Tn1 = Tp1 ,

    By substituting (D.4) into (D.3) and calculating |T(j)| results in

    |T(j)|

    k2p

    T2T 4 + (T 2kpk T )T 2 + k2k2. (D.5)

    ore, c

    , Ti

    pkh(T

    stitu

    2T22

    rmal

    2s2

    nces

    Morar1989.

    stront Soc

    Marg. Papallers fferen. Ang,E Tran

    Ziegles of t. Cohtrol, Tourtio

    ves, Au Dwyss, 20

    stro001) oron, omatkoczo

    the ro(6) (20koges, Journ

    stroomat

    stro for P714uo, Y.

    DCD722andyode conIndustummst co328. ZuriompaDC Bnsactiim, R

    tem, Iummst co328B. Shin

    variab (2012

    strok, IEE

    Isaksstem, Aaya, C

    phasable.

    . Proof of the parameter n

    n 5 it was shown that zeros of the controller for type-V,l loops are given by

    + Tn1Tn3 + Tn1Tn4 + Tn2Tn3 + Tn2Tn4 + Tn3Tn4 )T2Tn2Tn3 + Tn1Tn2Tn4 + Tn2Tn3Tn4 + Tn1Tn3Tn4 )Tn3Tn4 = 0

    (C.1)

    n2 + Tn3 )T2 4(Tn1Tn2 + Tn2Tn3 + Tn1Tn3 )T2Tn3 = 0.

    (C.2)

    + Tn2 )T + Tn1Tn2 = 0. (C.3)ting (95) into (C.1)(C.3) results in

    T2 + n2T2 + n2T2 + n2T2 + n2T2)T2 4(n3T3 n3T3 + n3T3)T + n4T4 = 0, (C.4)

    + nT)T2 4(n2T2 + n2T2 + n2T2)T + n3T3 = 0,(C.5)

    + nT)T + n2T2 = 0 (C.6)

    6n3T4 + n4T4 = 0, (C.7)

    n2T3 + n3T3 = 0, (C.8)

    + n2T2 = 0. (C.9) 0, from (C.6)(C.9) we obtain (98), (99) and (100).

    . The conventional Magnitude Optimum

    rolling the process dened by

    1 sTp1 )(1 + sTp2 )(1 + sT2p )

    , (D.1)

    =n

    i=3Tpi stands for the process unmodelled dynam-controller of the form

    Tn1s)(1 + Tn2s)is(1 + Tc s)

    (D.2)

    According to the conventional Magnitude Optimum cri-eters Tn1 , Tn2 , Ti will be determined so that |T(j)| 1 in

    ossible frequency range. Assuming that Tc T2p and+ Tc , the transfer function of the closed loop controlrding to Fig. 1 is equal to

    kp(1 + sTn1 )(1 + sTn2 ) + sTp1 )(1 + sTp2 )(1 + sT2 ) + khkp(1 + sTn1 )(1 + sTn2 )

    .

    (D.3)

    ole zero cancellation according to

    Tn2 = Tp2 . (D.4)

    Theref

    kh = 1= 2k

    By sub

    T(s) =

    and no

    T(s) =

    Refere

    [1] M. NJ,

    [2] K.J.me

    [3] N.I.[4] K.G

    troCon

    [5] K.HIEE

    [6] J.G.tion

    [7] G.Hcon

    [8] B. Cdri

    [9] A. OPre

    [10] K.J.9 (2

    [11] L. LAut

    [12] S. Sing52

    [13] S. Sing

    [14] K.J.Aut

    [15] K.J.tion699

    [16] L. Gfor223

    [17] B. Bmoon

    [18] V. Mboo288

    [19] E.WA cDCTra

    [20] K. Ksys

    [21] V. Mboo288

    [22] H.-for(3)

    [23] K.J.boo

    [24] A.J.sys

    [25] I. Kandi 2 i h 2 i p h

    ondition |T(j)| 1 is satised when= 2kpkhT2 = 2kpkh(T Tp1 Tp2 )

    Tn1 Tn2 ). (D.6)ting (D.4) and (D.6) into (D.3) results nally

    1

    s2 + 2T2s + 1(D.7)

    izing the time by setting s = sT2 leads to1

    + 2s + 1 . (D.8)

    i, E. Zariou, Robust Process Control, Prentice-Hall, Englewood Cliffs,

    m, T. Hgglund, PID Controllers: Theory, Design and Tuning, Instru-iety of America, 1995.aris, Lectures in Applied Automatic Control, Tziolas, 2003 (in Greek).dopoulos, E. Papastefanaki, N. Margaris, Optimal tuning of PID con-or type-III control loops, in: Proceedings of 19th Mediterraneance on Control and Automation, MED11, 2011.

    G. Chong, Y. Li, PID control system analysis design and technology,sactions on Control Systems Technology 13 (4) (2005) 559576.r, N.E. Nichols, Optimum settings for automatic controllers, Transac-

    he ASME 64 (1942) 759768.en, G.A. Coon, N.Y. Rochester, Theoretical considerations of retardedransactions of the ASME (1953) 175185.l, I.D. Landau, High speed adaptation system for controlled electricaltomatica 11 (1975) 4153.er, Handbook of PI and PID Controller Tuning Rules, Imperial College

    03.m, T. Hgglund, The future of PID control, Control Engineering Practice11631175.Tuning of PID controllers by the non-symmetrical optimum method,ica 33 (1997) 103107.wski, S. Domek, K. Pietrusewicz, B. Broel-Plater, A method for improv-bustness of PID control, IEEE Transactions on Industrial Electronics05 December) 16691676.tad, Simple analytic rules for model reduction and PID controller tun-al of Process Control 13 (2003) 291309.m, C.C. Hang, P. Persson, W.K. Ho, Towards intelligent PID control,ica, IFAC 28 (1) (1992) 19.m, T. Hgglund, C.C. Hang, W.K. Ho, Automatic tuning and adapta-ID controllersa survey, Control Engineering Practice 1 (4) (1993)

    .J. Hung, R.M. Nelms, Evaluation of DSP-based PID and fuzzy controllersC converters, IEEE Transactions on Industrial Electronics 56 (6) (2009)48.padhyay, P.S. Gandhi, S. Kurode, Sliding mode observer based slidingtroller for slosh-free motion through PID scheme, IEEE Transactionsrial Electronics 56 (9) (2009) 34323442.adi, Design of robust digital PID controller for H-bridge soft-switching

    nverter, IEEE Transactions on Industrial Electronics 58 (7) (2011)97.ta-Bustamante, J. Linares-Flores, E. Guzman-Ramirez, H. Sira-Ramirez,rison between the GPI and PID controllers for the stabilization of auck converter: a eld programmable gate array implementation, IEEEons on Industrial Electronics 58 (11) (2011) 52515262..C. Schaefer, Tuning a PID controller for a digital excitation controlEEE Transactions on Industry Applications 41 (2) (2005) 485492.adi, Design of robust digital PID controller for H-bridge soft-switching

    nverter, IEEE Transactions on Industrial Electronics 58 (7) (2011)97., J.-G. Park, Anti-windup PID controller with integral state predictorle-speed motor drives, IEEE Transactions on Industrial Electronics 59) 15091516.m, T. Hgglund, PID Control, in: W.S. Levine (Ed.), The Control Hand-E Press, Piscataway, NJ, 1996, pp. 198209.on, S.F. Graebe, Analytical PID parameter expressions for higher orderutomatica 35 (1999) 11211130.ontroller design for integrating processes using user-specied gaine margin specications and two degree-of-freedom IMC structure,

  • K.G. Papadopoulos, N.I. Margaris / Journal of Process Control 22 (2012) 11 25 25

    in: Proceedings of 2003 IEEE Conference on Control Applications, 2003, pp.898902.

    [26] I. Kaya, Improved cascade control structure for controlling unstable and inte-grating processes, in: Proceedings of the 44th IEEE Conference on Decisionand Control, and the European Control Conference, Seville, Spain, 2005, pp.71337138.

    [27] M.R. Matausek, A.D. Micic, A modied smith predictor for controlling a processwith an integrator and long dead-time, IEEE Transactions on Automatic Control41 (1996) 11991203.

    [28] R.C. Oldenbourg, H. Sartorius, A uniform approach to the optimum adjustmentof control loops, Transactions of the ASME (1954) 12651279.

    [29] C. Kessler, Das Symmetrische Optimum, Regelungstechnik, 1958 (395400 und432426).

    [30] W.J. Umland, M. Sauddin, Magnitude and symmetric optimum criterion forthe design of linear control systems: what is it and how does it compare

    with the others? IEEE Transactions on Industry Applications 26 (3) (1990)489497.

    [31] S. Preitl, R.E. Precup, An extension of tuning relation after symmetrical optimummethod for PI and PID controllers, Automatica 35 (1999) 17311736.

    [32] E. Poulin, A. Pomerleau, PI settings for integrating processes based on ultimatecycle information, IEEE Transactions on Control Systems Technology 7 (1999)509511.

    [33] F. Frhr, F. Orttenburger, Introduction to Electronic Control Engineering,Siemens, Berlin, 1982.

    [34] A. Buxbaum, K. Schierau, A. Straughen, Design of Control Systems for DC Drives,Springer-Verlag, Berlin, 1990.

    [35] O. Fllinger, Regelungstechnik, Hthig, Berlin, Heidelberg, 1994.[36] H. Lutz, W. Wendt, Taschenbuch der Regelungstechnik, Verlag, Harri Deutsch,

    Frankfurt am Main, 1998.

    Extending the Symmetrical Optimum criterion to the design of PID type-p control loops1 Introduction2 Definitions and preliminaries3 The conventional Symmetrical Optimum design criterion4 Extending the Symmetrical Optimum design criterion to type-III control loops5 Extending the Symmetrical Optimum design criterion to type-p control loops6 Simulation results6.1 Process with dominant time constants6.2 Process with time delay6.3 A non-minimum phase process6.4 Controller tuning without pole zero cancellation6.5 Comparison between a type-I and a type-III control loop6.6 A type-IV and a type-V control loop6.7 Effect of the process unmodelled dynamics to the control performance

    7 Conclusions and discussionAcknowledgementsAppendix A Optimization conditionsAppendix B Instability of the PI control conventional tuning via the Symmetrical Optimum criterionAppendix C Proof of the parameter nAppendix D The conventional Magnitude Optimum criterionReferences