Application A Method for Tuning Engineering PID Control Parameters Data — Partial Model Matching Method — eeeeeeeeeeeeeeeeeeeeeeeeee renee CONTENTS: 1. Introduction 2. Conventional Optimal-Parameters Tuning Methods «+++++++ +++ +++ 3. Principle of Partial Model Matching Method «+++ ++++++> 3.1 PID control system CA 471 1P action 3.2.2 FD action 4, Partial Model Matching Method for Discrete-time Systems 4.1 PID control system 10 4.2 LPD control aystom 13 5. Practice of Partial Model Matching Method «++ ----- sees 7 1s 5.1 The denominator series expression of a process +++ +++ ++ > 15, 5.2 Determination of reference model parameters Firetree AT 5.3. Calculation of rise time parameter 9 and control parameters 18 5.3.1 PID control system ++ 0555555 Z - aR 5.3.2 IPD control system <*+++ 18 6. Concluding Remarks ++ +++ ++s+++s++ 2 1, Introduction Although self-tuning type PID controllers are generalized today, there are still many cases when ‘control parameters are necessary to be tuned manually. For obtaining the values of PID control parameters, various methods have been formulated. These conventional methods, however, may not be appropriate for today's digital PID controllers if they ace applied blindly. ‘This ABD introduces a PID parameters tuning method that is more suitable for digital PID controllers than conventional tuning methods. YOKOGAWA Written by [Link], Advoneed AEDSSA01Z S6E Technology Dpt, |ASPD ‘Apr. 25. 1996 yee ania Published by Advonoed Technology Dnt, 1aseD PT 29 ®2. Conventional Optimal-Parameters Tuning Methods Many methods for tuning optimal PID parameters are invented in the past. Among them, representative uictliods based on a process indicial response are listed in Table 1 below. Table 1. Conventional Parameter Tuning Methods Control] “ oe Integral | Derivative ‘Evaluation Boreal Type | Mode Proportional Band (PD) | time (Ti) | Time (Td) Criteria F TeaLK/T : : rogwessouss | a8 [RT| anu wan Sa =) 2% Aton ain | tame azn) 2 oat P TWOLK/ (OS) : = aan | a [a yoo.’ (St) nina Const sw oon as) ie WOOLK/ (0ST) iW a. SL a oul i danas Shiono |g [| muxoasy” [tar [| Repeen Tin, PID 100LK/ (0.6T) T OSL i F TeoL aT) i rower | A [PLL seme @en) ft Z Response Time PID 100LK/ (0.95T) 135T O47L Peer P walk 3m) i i Chientrones- i = | | stam [on | oux” ear) a TTT) Seeense Tine A eeer — po ‘0 Overshoot) mb [oui 7"toasr) rn P WeaLK/ 7H i ial gemstones |g Lp 23 7) Response “Time NOOLK,4 (1.27) Ea 0.47t. ea D O/T ; 7 , Dine tcoceitosanny | TL ishimnse my | Miret: Mi = | atin controt Nis a | PL |eersiioouertoateay | Ta Pan Bo Punts tenmacnoasatni| ern | ian Crt taoourtosttsy | A Tipe A? Toning for satpant change Type Bs Tuning for stepiorm note suppression xr Proeas gain Ts lstoreriog tne constant L Dead tne Figure 1 shows an example of control response of which Ziegler-Nichols tuning method is applied to a process expressed as 1-e*/(1+10s). AEDB3A01Z — 56E(step) Dist = PID ID fetion) Contra Bariod = 8.4. aed FBS AT eo T= 40 (see) H's Bs cane Ucinative Atie = Qi Derivative fain = @ aa Cm 110.00 Figure 1. ZieglerNichols Tuning Method for Set-point Change Conventional tuning methods, however, have several drawbacks. One is that the conventional methous can only deal with Ist-urder-lag plus dead-tinie prucesses. Ziegler-Nichols and other conventional ining methods assume a Ist-arder-lag plus dead-time process in the form of K+e“/(1+Ts), and schematically estimate its parameters from the indicial response of the process: (1) Dead-time (L): The time that takes process output to appear after an input is applied. (2) Is-order-lag Time Constant (T); The time that takes process output ty reach 63.2% of its final value after the ontpnt is appeared (3) Process Gain (K): The ratio of final output value to input value. ‘The above procedure may cause instabilities in control when applied to a process having high-order-lag time constants. Figure 2 shuws the indicia! espouse uf a secumd-uidei-lag process capressed as Tee/(1 +9.28-+28e3). ‘The process, however, may he mistaken for T+e/(1+10s) if its parameters are estimated by the above procedure, and may derive PID control parameters the same as the ones in Figure 1 when Ziegler-Nichols tuning method is used. resultant control response that clearly indicates deterioration in controllabi igure 3 shows the (pen Loost MY = 1% (Stes) Tine (o20) Figure 2. The Indicial Respones of a High orderlag Process AFD53A017 ~56FCloged Loop: SU = 1%, (Step! Dist = o% Controller Fade = PID PID set ion) Control Period = @.1 (see) Goer st) Peat =e Geo (ees coe! Derivative fet*n'= Quist Derivative Gain = ey scsan pv = vali 2.20) 1s oreo) 2.001 aa ee, 109.00 Figure 3. Control Response by Ziegler-Nichols Tuning Method Another drawback in the conventional tuning method is that the control period — an important factor for digital PID controllers is not taken into account because most of the conventional methods are invented in the era of analog PID controllers. Figure 4 shows the controllability of a control system having the same configuration as that of Figure 1 in which the control period is changed to 1 second, a typical value of DCSs. The figure indicates that the conventional tuning methods are inadequate for controllers having Jong control period compared to the process movement. Closed toons Sy = 1%, te ist = SantretVor’: Rode ='P1D BID fetions Bestest fariog = 1 ano) Cypet SID) Pan a7 00. TE WD 85 oe Dirivat ive tn = Oa Derivative Sain = & ov Seyan p= yolloe 2.9 = 2.0) nfsan its : 1s eee Process: se rd 2.0! 7 joe fs fy e 58.08 109.00 Tine (ono) Figure 4. Control Response by ZiaglerNichols Tuning Method with Long Control P In addition to the above, the conventional tuning method is inadequate for processes having extremely short dead-time because derived proportional band (correspond to reciprocal of contro! gain) becomes too narrow. ‘The conventional tuning method, therefore, can only be applied when; (1) the process can sufficiently be approximated by a Lst-order-lag plus dead-time system, (2) the process has moderate dead-time, and (3) the control period of controlier is short compared to process movement. AEDS3A01Z — 56E3, Principle of Partial Model Matching Method ‘The method is invented by T. Kitamori in 1970's". The purpose of the method is to design a controller according to the extent of knowledge on the dynamics ot concerned process even if not all the dynamics are known. For easy understanding of the concept, the method for continuous-time systems is explained in this chapter. 3.1 PID control system Consider a control system as shown in Figure 5. Disturbance d Controller + Process + Set Point Gels) Ges) Contelled Variable Figure 5 (12). PID Control System In the figure, the total transfer function from set point to controlled variable is expressed as Equation (1) below. 1 L Gals) = is Go acl asians a's ast In Equation (1), @: are parameters that determine the form of response curve, and @ is a parameter Ural represents the rise time. The smaller g represents shorter rise time, In addition, determine the coefficients in the Equation (1) so that a desirable response (i.e., controllability) can be obtained when a step input is applied to the equation. The equation, then, is called a reference model that shall satisfy; © zero positional deviation at steady-state, © appropriate attenuation charactertstic (stability), and @ shortest rise time while satisfying both (1) and (2). ‘The above conditions can be put differently with respect to Equation (1) as follows. @’ Make the coefficient of s°(ie., a0) to 1. @®’ Make the coefficients of s' in the transfer function of actual total control system (to be described later) equat t0 a@' a', in the order of = 2, 3, 4, ou as far as adjustable parameters permit. @ ’ Make the coefficient of s' to as small positive values as possible (the values become 0). Under these conditions, Kitamori recommends following values for parameters a, to obtain a step response with about 10% overshoot. [et 0y 1p 25 a hay 1 }=(1,1,0.5, 0.15, 0.08, «+--+ AGD53A01Z —58E‘A step response with about 0% overshoot can be obtained when following values are used for the parameters (at the cost of rise time). (eo, 1,12, 5,044, ***+*} = (1, 1, 0.425, 0.0975, 0.014344, +--+} ‘The transfer function of a process is gencrally expressed by a proper rational function of s. Gui) = B= betbstbss*bys"+bs'+~ eS) a astasta,s*tas* rast “@ In this discussion, however, a denominator series transfer function in which numerator is normalized to unity will be used to express the process. Gets) = 1 WT oe aiyra staista So +a st ®) Where, aulbs ar— (batt tba’) bo (= 1.2. In addition, controllers will be expressed as follows: 1 c C+C,Stes+css'+c, 8+" Gas) = Kull 5 + Tos) = “S PUTIN eT TTT VT ATTN By using Equations (3) and (4), the total transfer function of the control system from set point to controlled variable can be expressed as follows. £1 vig) = GAGs)’ 1 se WO" GO-GO 1, £ 7 s ‘The basic of partial model matching method is to determine the _@ by equating Equations (1) and (6) so that the denominator coefficients of both equations will coincide with each other, starting from lowest order to as high order as possible. ‘The following solution will be obtained by placing W(s) equal to Gm(s). AEDS3A01Z — SOEEquation (6) will be expanded as; Shih eadorf + (22 -oas 8 +o%at a0) 8+ 0°Caas-a%~ad}s! oa 8 +0"ai—aafs? seeee (7) Now, in Equation (4), the terms ¢, and after will not he nsed in PI action as welll ase, and after not be used in PID action. Therefore, the coefficients of corresponding members in Equation (7) should also be zero. However, since undetermined parameter is only @, such many coefficients cannot be made simullancously equal ty 2c1v. Thus, complying with the above mentioned condition @ *, the coefficient af lowest order is made equal ta zern, among the ones that should he zero. Accordingly, following equations will be obtained, which should be satisfied bya . Plaction: . : 7 i “® Se to%ni— ade + aa, a}-a) = 0 PIN action: 93 — pare a ae Here, find a smallest positive real root to satisfy the above mentioned condition @) * by solving Equation (8) with respect to ¢. Controller parameters necessary for respective control actiuus ae then calculated by following equations. ceeeeeeeeeeeeeees @) ‘Thus, pa 100 co - af 1-2 : eeeet et eere ta AFDS3A017 —56FIf no positive real root is existed in Equation (9), denotes that the control response that satisfies the above mentioned condition @) ’ cannot be realized by desired control action. In this case, calculate controlicr parameters, first using the value of @ two to three times the a's/a’e, and check the step response of the control system by simulation. Then, repeat the step by gradually decreasing the value of @, and adopt the value little before the one when the response becomes inappropriate. 3.2 LPP control system ‘An IPD control system can be expressed as shown in Figure 6. Because the control action of I-PD controllers is only I-action, as sown in Figure 7. P-action and D-action are considered to be feedback compensation components that are added to the process according to required comurol action. Controller Process . ry Controlled Setpoint —-O—} Gs) rls) — Variable fH Feedback Gomponent(s} Gis) Figure 6. LPD Control System : 7 : 7 Gontrotied Setpoint, —+O-—| ection Process i li Variable * Daction +} Figure 7. -PD Controller ‘The reference model and process expression of an I-PD control system are the same as that of the PID control system (1.e., Equations (1) and (2)). However, since the controller has only action, its transfer function is expressed as fallaws. Gas) = = sey Further, the transfer function of feedback compensation unit is expressed as Follows. AEDB3A01Z — 56EGAG) = fafpthisttisth ns cece : Prom Equations (3), (11) aud (12), Uke total transfer Cunct variable is expressed as follows Wis) = +s ‘The following solution will he ahtained hy placing W(s) equal to Ga(s) in the same way as the PID control system. k 7 i=—a-D-a (ha,o-a',) + (ka.0?~a')s + (kaso ast Poet (14) There are two undetermined parameters, @ and k, in Fquation (14). To obtain and k, let successive feedback compensation units be expressed as follows when fof), up tofy.. are used for f . fr = kayo"! — a's = 0 fr = kaeo"'* alr = 0 Form Equation (15), o and k can be obtained as; Therefore, f= kaa — Where, 1=1,2,3, 0-1 3.2.1 HP action In an L-P action, only fy is used for a feedback compensation unit. Consequently, by taking 4s 1 in Equations (16) through (18); 20s aia AFD53A017 —56FVG fhe a's Therefore, pp = 100 [%] ere Ty i” Isec] 3.2.2 PD action In an I-PD action, two feedback compensation units, fp an f, ate used. Consequently, by taking @ as 2 in Equations (16) through (18); as aya’ = (22)*. (24)° ee (ON fy = kaso-a's f= karo*—a', Therefore, PB = 12% [96] a T= % [sec] fi Tom F [sec] 4, Partial Model Matching Method for Discrete-time Systems This method is derived by expanding the design principle of continuous-time systems explained in Section 3, and is applicable to continuous-time systems as well as sampled-data systems. The reference model and process expressions, therefore, are the same as that of the continuous-time systems, 4.1 PID control system First, solve ¢ with following form by equating the transfer function of control system with that of reference mode! In the same way as continuous-type systeuns. AEDS3A012 — 56E" ceeeeseeeees (19) Where, Po(0) = ao plo) = a, — cata’, P:(0) = a':— 9a, a", +0%a?—as)a’o Ps (0) = aoa’, + o%(ai—as)a’\~ oa? — 2azas+a,)a’o Equation (20) is a polynomial expression of @ of which coefficients are consisted of process parameters and reference model parameters. The parameters of a PID controller, then, are given as follows: o=2® Goon, o , feveeeereaeeeees (20) Undetermined @, in here, is obtained in the same way as continuous-time systems. For instance, ¢ ¢3 and after should all be zero when the controller in PI action. However, all of cx ¢s and after can not be made equal to zero at the same time since the number of parameter o is one. Therefore, only the lowest coefficient, ez, is made equal to zero. Then, since ¢:=p.(a)/o=0 or po(a)=0,6 will be determined. Adopt a smallest o when there are more than two positive roots. Now, the pulse transfer function of a sampled value PID controller can be expressed as follows. Gr = Bete ete —% a “QL au 2 PHOT A FEE AT bad 5 a Where, F(T: Sampling period) By approximating Equation (21) with continuous-time components, the equation is transformed into following expression. 1 Hof AOR ARE egter Atos A+ | seeseeees (22) a By rearrange Equation (22) with respect to s into ascending order of powers; [ot +ets+ (ct Pet s+ (et — Tet + ET'ct s+ ~ J es From Equation (23), relations of the parameters of continuous-time PID controllers with AEDS3A01Z — S8E12 that of sampled data PID controllers are obtained as follows. ater Tel + gret By solving Equation (24) with respect to cy, and substituting them into Equation (20); + _ Palo) i o op = 2 ae pla) + F Too) sesseune-neseaiasy feteeeeeeeeerees (25) 1 pila) + Teo) + To) ae o Now, ef and after in Pl action, or ef and after in PID action, should be made equal to zero. ‘Therefore, following equations are obtained Ww deterinine the value of o for sampled data PID control systems in the same way as the continuans-time system. Plaction: ps0)+$Tpi(o) = 0 26) FID acto = px0) + Toxo) +P pW) = 0 /e root in Equation (26), the parameters for respective control By obtaining a smallest posi actions can be determined from Equation (25). Thus, PB T: ; AEDB3A01Z — 66E12 4.2 LPD control system First, solve £ with following form by equating the transfer function of control system with that of reference model in the same way as continuous-type systems. (ao D= a’ = ako) +autsals +a2(k,a)s*+ ~ 1 Q7) ai(k,o) = kas. 0"! = ++ QB) Equation (27) is a polynomial expression of k and @ of which coefficients are consisted of process parameters and reference model parameters. The parameters of feedback compensation units in continuous-time system, then, are given as follows. KO) C= OL,2,Bp0) eres + @9) Now, the transfer function of a sampled data I action controlier is expressed by following equation. settee tteteeeeeeeeeeereees (30) By approximating Equation (30) with continuous-time components, the equation is transformed into following expression. as s G1) In addition, sampled data PD action compensation units are expressed by Equation (32) that is also transformed into Equation (33) with continuous-time components. ON ee Fea Re eae ed a eet tleen IC f= S[adretasem9), on =f+cr deine sar ters Lerner StF Potted se Sone Sra Lariat 3) From the above equations, relations of the parameters of continuous-time I-PD controllers with that of sampled data I-PD controllers are obtained as follows. AEDS3A01Z — 56Enan pre Following equations will be obtained by solving Equation (34) with respect to ff, and substituting them into Equation (29). fi = aulk,0) J 1 ff = qk,o) +3 Tadk,o) fF = ako) +Tad.o)+ ZT adko) pf areeeeneeny (35) = a.o)+ Stalk.o) +P Tako+ Palko) There are two undetermined parameters, k and o, in Equation (55). ‘These parameters can be determined by making ff and ff to zero for LP action, or ff and ff for I PD action, in the same way as the continuous-time system. By substituting Equation (28) for ai(k.a) in Equation (35), the equation concerning @ is obtained as follows: 1-P action: (a) + Fteoa 20° +(-ae+ GP anew +c 4e,—drair=0 LED action: (als #Tah4 FT aaa + (a's +50 + LP a aot} + (36) +Ca’s — ite ' [phawTaxo 1 dae - Ley Te +Cgas ate gh #)T =0 By obtaining smallest positive root for a, k"=k becomes as follows: AEDB3A01Z ~ 56E18 ait sta, LP action T oat FT) eveta Le, TIE easy 1-PD action olao*+ Ta+ 27) Krom Equation (35) fi" is obtained as; seeeeees (38) Thus, PB 100 qT Ke net 5. Practice of Partial Model Matching Method Each parameter calculation formula in the discrete-time control system explained in the above chapter coincides with that in the continuous-time control system when T is made equal to zero, Accordingly, the procedure for determining parameters in discrete-time control systems is explained in this chapter. Although these calculations can be done manually with a calculator, the use of a persanal computer is preferable. Because, the calculations are somewhat complicated, and there may be a case when it is necessary to modify parameters while monitoring simulated response curves. 5.1 The denominator series expression of a process A process transfer function is generally expressed by a rational polynomial expression of s. At first, it is necessary to transform the transfer function into a denominator series expression with its numerator normalized to unity, as shown in Equation (3). In addition, 1ce dead time exists in actual processes, the denominator series expression AEDS3A017 — RE16 should include the dead time. Because dead time cannot be expressed directly as a rational polynomial expression of 8, Padés approximation is used instead. Following expression shows the second-order approximation of dead time, and will suffice for the purpose. 12-6Ls+L*st "124 6Ls+L?s? Lis) = < Example > (1) Transfer function of process without dead time b(U) + b(1)s + b(2)s*+ al) + a(I)s+a(2)s? + “+b(n)s" +a(n)s™ G(s) = where,m @ =n (2) Approximated transfer function of dead time — 4(0) | d()s | d(@)s* HS) = CO +ell)s+e)s* (3) Combined process transfer function bb(0) + bb(1)s + bb(2)s* + aa(0)+ aa(1)s + aa(2)s* +bb(q)s* +aa(p)e” Gals) The coefficients of denominator and numerator of Gas) are obtained by multi of G(s) and L(6) respectively. ying those Calculation of denominator: for(i=0;i<=pjit+) fo k= 0; k <= 1; k++) if j + k == i) aa[i] += alj]*e(k]; where, 1: Order of Pade’s approximation (= 2) Pp: Order of denominator of Ga(s) AED5S3A01Z — 56ECalculation of numerator: for(i=0si<=q;itt) { pli] - 0; for( j= 0; j (1). Kitamori, "Control system design based on the partial knowledge of a control objective”, SICE Collected Papers 15 — 4, pp. 549/555, 1979 (JAPANESE) (2)T. Kitamori, “Sampled data control system design based on the partial knowledge ot a control objective”, SICE Collected Papors 15 — 5, pp. 695/700, 1979 (JAPANESE) ACDS3A01Z 56C,