Available online at www.sciencedirect.com ScienceDirect IFAC PapersOnLine 53-1 (2020) 195–200 New New New New tuning strategy for series cascade tuning strategy for series cascade tuning strategy for series cascade control structure tuning strategy for series cascade control structure control structure control structure ∗∗ Md. Imran Kalim ∗∗ Ahmad Ali ∗∗ Md. Imran Kalim ∗∗ Ahmad Ali ∗∗ Md. Imran Kalim Ahmad Ali ∗∗ Md. Imran Kalim ∗ Ahmad Ali ∗∗ of Technology ∗ of Electrical Engineering, Indian Institute ∗ Department Department of Electrical Engineering, Indian Institute of Technology ∗ ∗ Department of Electrical Engineering, Indian Institute of Technology Patna, Bihta-801106 India (e-mail: imran2025kalim@rediffmail.com). Patna, Bihta-801106 India (e-mail: imran2025kalim@rediffmail.com). ∗ ∗∗ Department of Electrical Indian Institute of Technology Department of Engineering, Indian Patna, Bihta-801106 IndiaEngineering, (e-mail: imran2025kalim@rediffmail.com). ∗∗ Department of Electrical Electrical Indian Institute Institute of of Technology Technology ∗∗ ∗∗ Patna, Bihta-801106 India Engineering, (e-mail: imran2025kalim@rediffmail.com). Patna, Bihta-801106 India (e-mail: ali@iitp.ac.in). Department of Electrical Engineering, Indian Institute of Technology Patna, Bihta-801106 India (e-mail: ali@iitp.ac.in). ∗∗ Department of Bihta-801106 Electrical Engineering, Indian Institute of Technology Patna, India (e-mail: ali@iitp.ac.in). Patna, Bihta-801106 India (e-mail: ali@iitp.ac.in). Abstract: Abstract: In In the the presence presence of of disturbances disturbances associated associated with with the the manipulated manipulated variable, variable, perforperforAbstract: In the presence of disturbances associated with with the manipulated variable, performance of a single loop control structure can be improved the help of a series cascade mance of a single loop control structure can be improved with the help of a series cascade Abstract: In the approach presence of disturbances associated with theismanipulated variable, perforcontrol. A design for a series cascade control system proposed in this paper. The mance of a single loop control structure can be improved with the help of a series cascade control. design for astructure series cascade system is the proposed in athis paper. The mance ofA acontroller singleapproach loop control cancentroid becontrol improved with help of series cascade secondary is tuned by finding the of the convex stability region to quickly control. A design approach for a series cascade control system is proposed in this paper. The secondary controller is tuned by finding the centroid of the convex stability region to quickly control. Athe design approach forbya finding series cascade control is proposed inregion this paper. The secondary controller is tuned the centroid of system the convex stability tovariable, quickly attenuate disturbances arising in the inner loop before they influence the controlled attenuate the disturbances arising in the inner loop before theyconvex influence the controlled variable, secondary controller is tuned by finding the centroid of the stability region to quickly attenuate the disturbances arising in the inner loop before they influence the controlled variable, whereas the primary controller is tuned by optimum criterion to achieve whereas the controller is in tuned by magnitude magnitude optimum criterion achieve improved improved attenuate theprimary disturbances the inner loop before they influence theto variable, servo response in the outerarising loop. Servo andmagnitude regulatory responses for the the nominal model, as whereas the primary controller is tuned by optimum criterion tocontrolled achieve model, improved servo response in the outer loop. Servo and regulatory responses for nominal as whereas the primary controller is tuned by magnitude optimum criterion to achieve improved well as for the perturbed model of the plant obtained by the proposed method are compared servo response in the outer loop. Servo and regulatory responses for the nominal model, as well asresponse for the perturbed model the plant the proposed method are model, compared servo in the outer loop.of and obtained regulatoryby responses forthat the nominal as well as for recently the perturbed model ofServo the plant obtained byresults the proposed method are compared with some reported tuning strategies. Simulation show the proposed tuning with some recently reported tuning strategies. Simulation results show that the proposed tuning well as for the perturbed model of the plant obtained by the proposed method are compared with some recently reported tuning strategies. Simulation results show that the proposed tuning strategy yields improved closed-loop performance. strategy yields improved closed-loop performance. with some recently reported tuning strategies. Simulation results show that the proposed tuning strategy yields improved closed-loop performance. © 2020, IFAC (International Federation ofperformance. Automatic Control) Hosting by Elsevier Ltd. All rights reserved. strategy yields improved closed-loop Keywords: Keywords: Cascade Cascade control, control, PID PID controller, controller, All All stabilizing stabilizing controller, controller, Magnitude Magnitude optimum optimum Keywords: Cascade control, PID controller, All stabilizing controller, Magnitude optimum criterion. criterion. Keywords: Cascade control, PID controller, All stabilizing controller, Magnitude optimum criterion. criterion. 1. INTRODUCTION totuning totuning has has been been applied applied to to the the two two loops loops sequentially. sequentially. 1. 1. INTRODUCTION INTRODUCTION Another way of design is simultaneous tuning approach totuning has been applied to the two loops sequentially. Another way of design is simultaneous tuning approach 1. INTRODUCTION totuning has been applied to are the tuned two loops sequentially. Another way of design is simultaneous tuning approach in which the two controllers concurrently. In in which the two controllers are tuned concurrently. In In conventional single loop control structure, disturbance Another way of design is simultaneous tuning approach in which the two controllers are tuned concurrently. In Jeng (2014), the controllers are tuned using step response In conventional single loop control structure, disturbance Jeng (2014), the controllers are tuned using step response In conventional single loop control structure, disturbance rejection does not take place until the system’s output dein which thethe twocontrollers controllers are tuned concurrently. In data without considering the process model. The aim rejection does not take place until the system’s output deJeng (2014), are tuned using step response In conventional single control structure, withoutthe considering the model.step The aim is is rejection does take loop place until the system’s output de- data viates from thenot set-point. That is why why Franks disturbance and Worley Jeng (2014), controllers areprocess tuned using response to obtain the controller parameters such that the two viates from the set-point. That is Franks and Worley data without considering the process model. The aim is rejection does not take place until the system’s output deto obtain the controller parameters such that the two viates from the set-point. That is why Franks and Worley proposed aa cascade cascade control control scheme scheme (Franks (Franks and and Worley to data without the process model. Thethe aim is obtain the considering controller parameters such that two loops behave as close as possible to the assumed referproposed viates from set-point. That is why(Franks Franks and cascade behave ascontroller close as parameters possible to such the assumed referproposed a the cascade control scheme and Worley loops (1956)). Fig. 1 shows the block diagram of a series to obtain the that the two loops behave as close as possible to the assumed reference models for the two loops. In Veronesi and Visioli (1956)). Fig. 1 the block diagram of a proposed a cascade control scheme (Franks and cascade Worley ence models for the two loops. IntoVeronesi and Visioli (1956)). Fig. 1 shows shows theand block diagram of a series series cascade control scheme. Gp1 (s) G (s) the funcbehave as close asprimary possible the assumed refer(2011), parameters of and secondary processes and Gp2 (s) are are of thea transfer transfer func- loops control scheme. G ence models for the two loops. In and Visioli p1 (s) p2 (1956)). Fig. 1 shows the block diagram series cascade (2011), parameters of the the primary andVeronesi secondary processes control scheme. G p1 (s) p2processes tions of primary and secondary whereas primary and G (s) are the transfer funcence models for the two loops. In Veronesi and Visioli p1 p2 were first estimated by evaluating a step response of tions of primary and secondary processes whereas primary (2011), parameters of the primary and secondary processes control G andare Gp2 (s) are the transfer funcwere first estimatedofby evaluating a step response of the the tions of scheme. primarycontrollers and secondary processes whereas p1 (s) and secondary represented by G (s) and (2011), parameters the primary and secondary processes c1primary were first estimated by evaluating a step response ofusing the system with roughly tuned PID controllers. Then and secondary controllers are represented by G (s) and c1 tions of primary and secondary processes whereas primary system with roughly tuned PID controllers. Then using and secondary controllers are represented by G (s) and c1 G (s) respectively. y (s) denotes the output of the outer were first estimated by evaluating a step response of the c2 (s) respectively. y1 (s) denotes the output of c1 system with roughly tuned PID controllers. Then using the estimated model, the two controllers have been tuned G the outer c2 secondary controllers 1 and are isrepresented by Gyc1 (s)outer and the estimated model, the twoPID controllers have Then been tuned c2 loop and output of inner represented by SetG respectively. y11 (s)loop denotes the output of the system with roughly tuned controllers. using 2 (s). c2 (s) by using internal model control strategy. Simultaneous loop and output of inner loop is represented by y (s). Setthe estimated model, the two controllers have been tuned 2 G respectively. y1 (s) denotes the output outer using internal model control strategy. Simultaneous c2 (s)for 2 (s). point outer is denoted by output loop output ofloop inner is represented byofythe Set- by the estimated model, the two controllers have been tuned 2 tuning based on user defined settling time and maximum pointand for the the outer loop isloop denoted by rr11 (s) (s) whereas whereas output by using internal model control strategy. Simultaneous loop and output of inner loop is represented by y (s). Settuning based on user defined settling time and maximum point for the outer loop is denoted by r (s) whereas output 2 1r2 (s)) for the inner of G (s) serves as the set-point (i.e. by using internal model control strategy. Simultaneous c1 1 tuning based on user defined settling time and maximum using genetic algorithm is reported in of Gc1for (s)the serves asloop the isset-point (i.e. for theoutput inner overshoot 2 (s)) point outer denoted by r1rr(s) whereas overshoot using genetic algorithm istime reported in Kaya Kaya 2 loop. is manipulated variable. d d of Gc1 (s) serves the set-point (i.e. for and the inner based on user defined settling and maximum c1u(s) 2 (s)) and Nalbantoğlu (2016). Internal model control approach loop. u(s) is the theas manipulated variable. d11 (s) (s) and d22 (s) (s) tuning overshoot using genetic algorithm is reported in Kaya of G (s) serves as the set-point (i.e. r (s)) for the inner and Nalbantoğlu (2016). Internal model control approach c1u(s) 2outputs 1 2 are the load disturbances entering the of G loop. is the manipulated variable. d (s) and d (s) overshoot using genetic algorithm is reported in Kaya p1 1 was used tune primary and are loadis disturbances entering the outputs of Gdp12 (s) and Nalbantoğlu (2016). Internal control approach loop.the u(s) thedisturbance manipulated variable. d1 (s) Gand (s) was used to to simultaneously simultaneously tune model primary and secondary secondary are the loadwith disturbances entering the outputs of Gp1 2and p1 and G (s) transfer functions (s) (s) and Nalbantoğlu (2016). Internal model control approach p2 d1 was used to simultaneously tune primary and secondary in Jeng Two of freedom and Gp2 (s) disturbance transferthe functions Gof (s) and are the loadwith disturbances entering outputs Gp1 (s) controllers controllers Jeng and and Lee Lee (2012). (2012). Two degree degree of freedom and Gp2 (s) with disturbance transfer functions Gd1 and p2respectively. d1 (s) G Series cascade control scheme is used was usedcontrol toin tune primary and secondary d2 (s) d1 controllers insimultaneously Jengstructures and Lee (2012). Twobeen degree of freedom cascade have also proposed in G (s) respectively. Series cascade control scheme is used d2 and G (s) with disturbance transfer functions G (s) and cascade control structures have also been proposed in p2 d1 d2 when the dynamics of the inner loop is fast so that there G (s) respectively. Series cascade control scheme is used controllers in (2008),Alfaro Jengstructures and Lee (2012). Twobeen degree of freedom d2 the dynamics of the inner loop is fast so that there Alfaro et al. et al. (2009). If there exists when cascade control have also proposed ina G (s) respectively. Series cascade control scheme is used Alfaro et al. (2008),Alfaro et al. (2009). If there exists d2 bethe will rapid attenuation of disturbances in the inner loop. when dynamics of the inner loop is fast so that there cascade control structures have also been proposed ina large time delay in the outer loop, a series cascade control will be rapid attenuation of disturbances in the inner loop. Alfaro et al. (2008),Alfaro et al. (2009). If there exists a when the dynamics of the inner loop is fast so that there large time delay in the outer loop, a series cascade control will be rapid attenuation of disturbances in the inner loop. Tuning methods reported for single loop control system, Alfaro et fails al. (2008),Alfaro al.servo (2009). Ifcascade there exists a large time delay in theacceptable outeretloop, a series control structure to give and regulatory perTuning methods reported for single loop control system, will be rapid attenuation offor disturbances the inner loop. structure fails to in give acceptable servo and cascade regulatory perTuning methods reported single(Krishnaswamy loopincontrol system, e.g. optimal integral error method et al. large time delay the outer loop, a series control structure fails to give acceptable servo and regulatory perAn cascade control structure which e.g. optimal integral error for method et al. formances. Tuning methods reported single(Krishnaswamy loop control system, Antoimproved improved cascadeservo control which (1990)), IMC based method (Lee e.g. optimal error method et al. formances. structure fails give acceptable andstructure regulatory peruses an internal model control for the inner loop a (1990)), IMCintegral based tuning tuning method(Krishnaswamy (Lee et et al. al. (2002)), (2002)), formances. An improved cascade control structure e.g. optimal integral error method (Krishnaswamy et al. uses an internal model control for the inner loop and andwhich a PIPIdirect synthesis approach (Raja and Ali (2015)) etc. have (1990)), IMC based tuning method (Lee et al. (2002)), formances. An improved cascade control structure which PD Smith predictor in the outer loop has been proposed direct synthesis approach (Raja and Ali (2015)) etc. have uses an internal model control for the inner loop and a PI(1990)), IMC based tuning method (Lee et al. (2002)), PD Smith predictor in the outer loop has been proposed direct synthesis approach (Raja and Ali (2015)) etc. have been extended to cascade control scheme. There exists two uses an internal model control for loop the inner loop and a PIPD Smith predictor in the outer has been proposed in Kaya et al. (2007) for such class of processes. been extended to cascade control scheme. There exists two direct synthesis (Raja and Ali (2015)) etc. have Kaya et predictor al. (2007) in forthe such classloop of processes. been extended toapproach cascade control scheme. There exists two in different ways design cascade control structure. The PD Smith outer has been controller proposed in Kaya et al. (2007) for such class of processes. Determination of all stabilizing PI, PD and different ways to to cascade design a acontrol cascade controlThere structure. been extended scheme. exists The two Determination of all stabilizing PI, PD and PID PID controller usual approach is to first tune the secondary controller by different ways to design a cascade control structure. The in Kaya et al. (2007) for such class of processes. parameters is reported in Ho et al. (1997); Tan et al. usual approach is to first tune the secondary controller by Determination of all stabilizing PI, PD and PID controller different ways tois design atune cascade control structure. The is reported in Ho et al. Tan et controller al. (2003, (2003, assuming the primary controller open state. The next usual approach to first thein secondary controller by parameters Determination of all(2013, stabilizing PI,(1997); PDusing and PID 2005a, 2006); Onat 2019). By the above reassuming the primary controller in open state. The next parameters is reported in Ho et al. (1997); Tan et al. (2003, usual approach is to first tune the secondary controller by 2005a, 2006); Onat (2013, 2019). By using the above reassuming the primary controller in open state. The next step is to tune the primary controller with the application parameters is reported in Ho2019). et al. (1997); Tan et above al. (2003, 2005a, 2006); Onata (2013, Byregion using the reported techniques, convex stability is obtained instep is to tune the primary controller with the application assuming the primary controller incontroller open The next ported techniques, a convex stability region is obtained instep is previously to tune thetuned primary controller withstate. thetoapplication of the secondary the inner 2005a, 2006); Onat (2013, 2019). By using the above reported techniques, a convex stability region is obtained inside the stability boundary locus in the plane of controller of the previously tuned secondary controller to the inner step isThis to tune thetuned primary controller with thetoapplication side the stability boundary locus in the plane of controller loop. approach is known as sequential approach. of the previously secondary controller the inner ported techniques, a convex stability region is of obtained inparameters. Weighted geometrical center (WGC) method loop. This approach is known as sequential approach. side the stability boundary locus in the plane controller of the This previously tunedis secondary todesign the inner parameters. Weighted geometrical center (WGC) method Direct synthesis approach has been used to the loop. approach known ascontroller sequential approach. side the stability boundary locus inthe theconvex plane of controller and calculation of centroid from stability reDirect synthesis approach has been used to design the parameters. Weighted geometrical center (WGC) method loop. This approach is known as (2015). sequential approach. calculation of centroid from the convex stability reDirect synthesis approach has Ali been used to design the and controller settings in In Vivek parameters. Weighted geometrical center (WGC) method and calculation of centroid from (2013) the convex stability rehas been reported in Onat and Onat (2019) controller settingsapproach in Raja Raja and and (2015). In design Vivek and and Direct synthesis has Ali been used to the gion gion has been reported in Onat (2013) and Onat (2019) controller settings in Raja and Ali (2015). In Vivek and Chidambaram (2013), a typical relay feedback based auand of centroid from (2013) the convex stability region calculation has been reported in Onat and Onat (2019) Chidambaram (2013), a typical relay feedback based aucontroller settings in Raja and Ali (2015). In Vivek Chidambaram (2013), a typical relay feedback based and au- gion has been reported in Onat (2013) and Onat (2019) Chidambaram a typical relay feedback basedControl) au- Hosting by Elsevier Ltd. All rights reserved. 2405-8963 © 2020,(2013), IFAC (International Federation of Automatic Peer review under responsibility of International Federation of Automatic Control. 10.1016/j.ifacol.2020.06.033 Md. Imran Kalim et al. / IFAC PapersOnLine 53-1 (2020) 195–200 196 Fig. 1. Series cascade control scheme respectively. Calculation of WGC is very tedious as it uses all frequency points forming the stability boundary locus, whereas calculation of centroid requires only few frequency dependent points, namely cusp and corner points. It is also reported in Onat (2019) that the controller parameters corresponding to the centroid yields satisfactory closedloop performance. Furthermore, the centroid is located at a safe distance from all the points on the stability boundary locus. Hence, it will give improved set-point tracking and yields faster disturbance rejection. Also, it will maintain excellent robustness in case of process parameter variations. Hence, the controller settings for the inner loop is obtained by finding the centroid of the convex stability region. For the output to exactly follow the set point, the closed loop response should have infinite bandwidth and zero phase shift which is practically not possible (Vrancic et al. (2001)). The magnitude optimum criterion tries to maintain the closed loop magnitude frequency response curve flat and close to unity over the widest possible frequency range for any given plant and controller structure (Umland and Safiuddin (1988)). This method is called as modulus optimum (Åström and Hägglund (1995)), Betragsoptimum (Kessler (1955)), magnitude optimum (Umland and Safiuddin (1988)). Actually, this criterion is based on optimizing the closed-loop amplitude response. Furthermore, it is reported in the literature that the magnitude optimum criterion results in superior tracking performance for a large number of process models (Kessler (1955); Vrancic et al. (2001); Vrančić et al. (1999)). Hence, in this proposed work, the primary controller is tuned with the help of magnitude optimum criterion to achieve improved set-point tracking performance and faster disturbance rejection in the outer loop. The paper is organized as follows: Tuning rules of primary and secondary controllers are derived in Section 2. Simulation examples, results and comparisons are given in Section 3 while Section 4 provides the concluding remarks about the proposed work. 2. TUNING OF GC1 AND GC2 2.1 All stabilizing PI controller for Gc2 In the proposed work, Gc2 is considered as a PI controller whose transfer function is given below, Gc2 (s) = Nc2 (s) Ki2 = Kp2 + Dc2 (s) s (1) where Kp2 and Ki2 are the proportional and integral gains of the secondary controller respectively. Gc2 (s) is tuned by finding the centroid of the convex stability region obtained by plotting the stability boundary locus in Kp2 and Ki2 plane (Tan et al. (2003), Onat (2019)). Let, the transfer function of the secondary process be, Gp2 (s) = Np2 (s) −τ2 s e Dp2 (s) (2) The characteristic polynomial of the inner loop is given by, ∆P (s) = 1 + Gc2 (s)Gp2 (s) = Dp2 (s)Dc2 (s) + Np2 (s)Nc2 (s)e−τ2 s Substituting s = jw and e jsin(τ2 w)) in (3), we get −τ2 s = e −jτ2 w (3) = (cos(τ2 w) − ∆P (jw) = Dp2 (jw)Dc2 (jw) + Np2 (jw)Nc2 (jw)(cos(τ2 w) − jsin(τ2 w)) (4) Let R∆P and I∆P represent the real and imaginary parts of ∆P (jw). Following are the three different ways in which a root of any stable polynomial can cross the imaginary axis (Tan (2009)). (i) Real Root Boundary (RRB): A real root can cross the imaginary axis at s = 0. RRB is obtained by substituting s = 0 in (3) which gives Ki2 = 0 as the RRB line. (ii) Infinite Root Boundary (IRB): A real root can cross the imaginary axis at s = ∞. IRB is obtained by substituting s = ∞ in (3). There will not be any IRB line for proper transfer function. (iii) Complex Root Boundary (CRB): Roots on the imaginary axis can become unstable which implies that both the real and imaginary parts of (4) will become zero simultaneously. The procedure to obtain complex root boundary is given in Tan (2009) and is discussed below in brief. Kp2 and Ki2 are obtained as functions of w by solving R∆P = 0 and I∆P = 0. The complex root boundary in Kp2 and Ki2 plane is obtained by varying w from 0 to wc , where wc is the lowest value of frequency other than w = 0 at which Ki2 = 0. Convex stability region in Kp2 and Ki2 plane is the area bounded by CRB and RRB which contains all the stabilizing values of Kp2 and Ki2 for the considered plant. The cusp and corner points of the convex stability region are identified as (Kp1 ,Ki1 ),(Kp2 ,Ki2 ).......(Kpm ,Kim ) and (Kp1 ,Ki1 ),(Kp2 ,Ki2 )......(Kpn , Kin ), where m and n are the number of cusp and corner points respectively. The parameters of inner controller are obtained by finding the centroid of the convex stability region as given below (Onat (2019)), n m ) + ( l=1 (Kpl ) ( l=1 Kpl (5) Kp2 = m + n n m ( l=1 Kil ) + ( l=1 (Kil ) Ki2 = (6) m+n Md. Imran Kalim et al. / IFAC PapersOnLine 53-1 (2020) 195–200 2.2 Magnitude optimum criterion for Gc1 197 120 Cusp point of the convex stability region 100 G1 (s) = Gc2 (s)Gp2 (s) Gp1 (s) 1 + Gc2 (s)Gp2 (s) (8) .Let Gcl (s) be the closed loop transfer function from reference to output. To achieve the magnitude optimum criterion, the controller should satisfy the following two conditions for as many q as possible (Vrancic et al. (2001)): Gcl (0) = 1 dq |Gcl (jw)| = 0, q = 1, 2, ..., qmax dwq w=0 (9) (10) First equation can be satisfied by using a controller structure having an integral term. qmax is taken as 2 for a PI controller and 3 for a PID controller. Assuming that the plant model for the primary controller is given as follows: 1 + a1 s + a2 s2 + .... + am sm −τ1 s e (11) 1 + b1 s + b2 s2 + .... + bn sn where Kr is the process steady state gain, τ1 is the process time delay. (a1 ,..., am ) and (b1 ,..., bm ) are the process parameters. Time delay term is replaced by its Taylor’s series approximation. Multiplying (7) and (11), the open loop transfer function becomes, G1 (s) = Kr c0 + c1 s + c2 s2 + ....... (12) G1 (s)Gc1 (s) = d0 s + d1 s2 + d2 s3 + ....... where, the coefficients ci and di (i = 0, 1, 2, 3...) are functions of the plant and controller parameters which can be obtained using the expressions given below (Vrancic et al. (2001)): di = bi ci = Kr [Ki1 φi + Kp1 φi−1 + Kd1 φi−2 ] φi = i (−1)i−l al l=0 τ1i−l (i − l)! a0 = b0 = 1 ai = bi = 0, i < 0. (13) Necessary and sufficient condition for stability reported in Hanus (1975) is as given below: c0 >0 (14) d0 The PID controller parameters can be obtained by solving the following equations for n = 0 − 2 (Hanus (1975)). 2n+1 i=0 (−1)i ci d2n+1−i = 0.5 2n (−1)i di d2n−i (15) i=0 By substituting (13) into (15), following expressions are obtained for Kp1 , Ki1 and Kd1 (Vrancic et al. (2001)). Stability boundary locus i2 80 K In the proposed work, Gc1 (s) is assumed as a PID controller with following transfer function: Ki1 Gc1 (s) = Kp1 + + Kd1 s (7) s where Kp1 , Ki1 and Kd1 are the proportional, integral and derivative gains of the primary controller respectively. From Fig. 1, G1 (s) can be calculated as, CONVEX STABILITY REGION 60 (15.89,38.03) Centroid of the convex stability region 40 20 Corner points of the convex stability region 0 −20 −5 0 5 10 15 Kp2 20 25 30 35 Fig. 2. Stability boundary locus of inner loop of example 1 A23 − A1 A5 (16) 2(A1 A2 A3 + A0 A1 A5 − A21 A4 − A0 A23 ) A2 A3 − A1 A4 Ki1 = (17) 2(A1 A2 A3 + A0 A1 A5 − A21 A4 − A0 A23 ) A3 A4 − A2 A5 Kd1 = (18) 2(A1 A2 A3 + A0 A1 A5 − A21 A4 − A0 A23 ) where symbols A0 , A1 , A2 , A3 , A4 and A5 are known as characteristic areas of the process. These areas can be calculated using the following expressions. A0 = Kr j l τ b j−l ) Aj = Kr (−1)j+1 (bj − aj ) + ( (−1)j+l 1 l! Kp1 = l=1 + j−1 (−1)j+l−1 Al aj−l (19) l=1 3. SIMULATION EXAMPLES In this section, three simulation examples are provided to demonstrate the effectiveness of the proposed method. Following commonly used parallel form of PID controller is used in simulation. Kd1 s Ki1 + (20) Gc1 (s) = Kp1 + s tf 1 s + 1 where, tf 1 = 0.1Td1 , and Td1 = Kd1 /Kp1 . Example 1. In Jeng (2014), following higher order process has been studied. 1 Gp1 (s) = Gd1 (s) = (10s + 1)(4s + 1)(s + 1)2 1 e−0.1s Gp2 (s) = Gd2 (s) = (21) 1.9s + 1 For tuning of secondary controller, convex stability region of the inner loop is obtained in Kp2 and Ki2 plane using the procedure given in section 2.1, and is shown in Fig. 2. It is having one cusp point and two corner points as shown. Centroid of the convex stability region is calculated using (5) and (6). By applying (16) to (18), controller settings for outer loop are obtained. Table 1 shows the controller tuning parameters for the method of Jeng (2014) and the proposed technique. Closed loop response to unit step setpoint change at t = 0s, step input disturbances d2 = 30 at t = 80s and d1 = 1 at t = 160s is shown in Fig. 3 while control signal is provided in Fig. 4. Performance measures are given in Table 2. To show the robustness to parameter Md. Imran Kalim et al. / IFAC PapersOnLine 53-1 (2020) 195–200 198 Table 1. Controller tuning parameters for various examples. Example 1 Example 2 Example 3 Tuning method J.C.Jeng Proposed J.C.Jeng Proposed Kaya and Nalbantoglu Proposed Inner Kp2 3.633 15.89 0.617 1.218 1.2214 0.79 loop Ki2 3.043 38.03 0.0595 0.128 1.2214 5.73 Outer Kp1 1.974 3.8113 0.088 0.1016 1 1.0548 loop Ki1 0.1376 0.2690 0.0010136 0.0012 0.32106 0.4897 Kd1 6.025 11.219 2.27304 2.2703 0.6090 0.5899 Table 2. Performance measures for various examples. Example 1 Example 2 Example 3 Tr (sec.) Ts (sec.) %Mp IAE ISE ITAE Tr (sec.) Ts (sec.) %Mp IAE ISE ITAE Tr (sec.) Ts (sec.) %Mp IAE ISE ITAE Servo J.C.Jeng 10.5 21.6 0 7.268 4.807 38.46 390 201 0.12 98.95 75.4 6042 Kaya and Nalbantoglu 4.64 5.587 1.006 3.1298 2.485 5.687 Performance Proposed 7.7 13.7 6 4.272 2.993 13.01 146.3 231 5.84 91.98 69.39 5364 Proposed 2.94 7.1 8.1 2.3 1.82 3.318 Regulatory(d2) J.C.Jeng 51 38.3 6.064 1.164 121.3 425 53.1 80.82 24.98 14990 Kaya and Nalbantoglu 12.37 16.2 0.7874 0.06688 5.054 Performance Proposed 7.8 2.7 0.3149 0.00485 4.827 334 24.5 36.01 5.478 6160 Proposed 6.72 16.3 0.2092 0.00817 0.8688 Regulatory (d1) J.C.Jeng 50 31 7.257 1.517 153.8 408 60.7 98.77 41.67 17810 Kaya and Nalbantoglu 14.27 63.5 3.115 1.288 16.85 Performance Proposed 38 20 3.714 0.4766 66.12 317.5 57.6 83.36 34.51 13790 Proposed 8.6 60.2 2.047 0.8686 8.273 1.5 1.5 y 1 1 0 0 y1 1 0.5 J.C.Jeng Proposed 50 100 150 Time (seconds) 0.5 0 0 Control Signal Magnitude Fig. 3. Closed-loop response of example 1 4 2 0 0 50 100 150 200 50 100 150 Time(seconds) 200 Fig. 5. Closed-loop response for the perturbed plant of example 1 J.C.Jeng Proposed 6 J.C.Jeng Proposed 200 250 Time (seconds) Fig. 4. Control signal of example 1 variations of the plant, the controller is applied to following perturbed model of the plant. 1.2 Gp1 (s) = (8s + 1)(3.2s + 1)(0.8s + 1)2 1.2 e−0.12s Gp2 (s) = (22) 1.65s + 1 Closed loop response for the perturbed plant is shown in Fig. 5. So, the proposed method provides excellent robustness with improved closed loop response. It can be concluded from Fig. 3, Fig. 4, Fig. 5, and Table 2 that the proposed method performs better than the strategy reported in Jeng (2014). Example 2. Let us consider the following higher order process which exhibits nonminimum-phase dynamics (Jeng (2014)). 10(−5s + 1) Gp1 (s) = Gd1 (s) = e−5s (30s + 1)3 (10s + 1)2 3 e−3s (23) Gp2 (s) = Gd2 (s) = 13.3s + 1 The computed convex stability region is shown in Fig. 6. Table 1 summarizes the controller settings for the inner and outer loops for the design method of Jeng (2014) and the proposed technique. Fig. 7 shows the closed loop response to unit step set point change at t = 0s, step input disturbances d2 = 0.4 at t = 700s and d1 = 0.1 at t = 1400s. Control signal is shown in Fig. 8. For robustness consideration, following perturbed model of the plant has been considered. 12(−6s + 1) Gp1 (s) = e−6s (24s + 1)3 (8s + 1)2 3.6 e−3.6s Gp2 (s) = (24) 10.64s + 1 It is observed from Fig. 7, Fig. 8, Fig. 9, and Table 2 that the proposed method yields improved closed loop Md. Imran Kalim et al. / IFAC PapersOnLine 53-1 (2020) 195–200 0.4 Cusp point of the convex stability region Stability boundary CONVEX 0.3 locus STABILITY 0.25 REGION Cusp point of the convex stability region 16 0.2 (1.218,0.128) Centroid of the convex stability region 0.1 CONVEX STABILITY REGION 8 (0.79,5.73) Centroid of the convex stability region 4 Corner points of the convex stability region (Corner points of the convex stability region) 0.05 0 −0.05 −0.5 Stability boundary locus 12 Ki2 Ki2 0.35 0.15 199 0 0 0.5 1 K 1.5 2 2.5 −2 3 −1 −0.5 0 p2 0.5 K 1 1.5 2 2.5 p2 Fig. 6. Stability boundary locus of inner loop of example 2 Fig. 10. Stability boundary locus of inner loop of example 3 1.5 J.C.Jeng Proposed 0.5 0 0 500 1000 1500 Time (seconds) y1 y 1 1.5 1 0.5 0 0 20 40 60 80 Time(seconds) 100 120 Fig. 11. Closed-loop response of example 3 0.6 J.C.Jeng Proposed 0.4 0.2 500 1000 1500 2000 Time(seconds) Control Signal Magnitude Control Signal Magnitude Kaya and Nalbantoglu Proposed 2000 Fig. 7. Closed-loop response of example 2 0 0 1 Fig. 8. Control signal of example 2 1.5 1 0.5 0 0 Kaya and Nalbantoglu Proposed 20 40 60 80 100 120 Time (seconds) 2 Fig. 12. Control signal of example 3 1 0 0 1.5 J.C.Jeng Proposed 0.5 500 1000 1500 Time(seconds) 2000 1 y1 y1 1.5 0.5 Fig. 9. Closed-loop response for the perturbed plant of example 2 performance for nominal as well as perturbed model of the plant. Example 3. As a third example, Following self regulating process has been considered (Kaya and Nalbantoğlu (2016)). 1 Gp1 (s) = Gd1 (s) = e−s (s + 1)2 1 e−0.1s Gp2 (s) = Gd2 (s) = (25) 0.1s + 1 The convex stability region is shown in Fig. 10. Obtained values of the controller parameters for the two loops are given in Table 1 for the proposed method and the design approach of Kaya and Nalbantoğlu (2016) . Fig. 11 shows the closed loop response to unit step set-point change at t = 0s, step input disturbances d2 = 1 at t = 40s and 0 0 Kaya and Nalbantoglu Proposed 20 40 60 80 Time (seconds) 100 120 Fig. 13. Closed-loop response under noise for example 3 d1 = 1 at t = 80s. Control signal is provided in Fig. 12. As can be seen from the graph that the proposed method yields improved servo as well as regulatory responses as compared to the method of Kaya and Nalbantoğlu (2016). To demonstrate the effects of the noise constraint, the system is subjected to white noise of standard deviation σn = 0.005, band-limited by the Nyquist frequency corresponding to the controller sampling period h = 0.01. Closed-loop response and corresponding control signal under noise are shown in Fig. 13 and Fig. 14 respectively. As shown in the figures, the response by the proposed method shows still better performance than that by the method of Kaya and Nalbantoğlu (2016). Md. Imran Kalim et al. / IFAC PapersOnLine 53-1 (2020) 195–200 200 Control Signal Magnitude 3 2 1 0 −1 −2 0 Kaya and Nalbantoglu Proposed 20 40 60 80 100 120 Time (seconds) Fig. 14. Control signal under noise for example 3 4. CONCLUSIONS A tuning strategy for a series cascade control structure has been presented in this paper. Simulation examples have been provided to show the benefits of the proposed technique. Closed-loop responses illustrate that the proposed tuning method yields better servo and regulatory performances as compared to the recently reported tuning strategies for the nominal as well as the perturbed model of the plant. The improved results of the procedure are demonstrated with respect to measurement noise. An extension of the present work can include the simultaneous tuning of series cascade control system and also the incorporation of performance-robustness trade-off into the parameter selection criterion. REFERENCES Alfaro, V.M., Vilanova, R., and Arrieta, O. (2008). Twodegree-of-freedom PI/PID tuning approach for smooth control on cascade control systems. In 47th IEEE Conference on Decision and Control, 5680–5685. IEEE. Alfaro, V., Vilanova, R., and Arrieta, O. (2009). Robust tuning of two-degree-of-freedom (2-dof) PI/PID based cascade control systems. Journal of process control, 19(10), 1658–1670. Åström, K.J. and Hägglund, T. (1995). PID controllers: theory, design, and tuning, volume 2. Instrument society of America Research Triangle Park, NC. Franks, R. and Worley, C. (1956). Quantitative analysis of cascade control. Industrial & Engineering Chemistry, 48(6), 1074–1079. Hanus, R. (1975). Determination of controllers parameters in the frequency domain. Journal A, 16(3), 128. Ho, M.T., Datta, A., and Bhattacharyya, S. (1997). A linear programming characterization of all stabilizing PID controllers. In Proceedings of the American Control Conference, volume 6, 3922–3928. IEEE. Jeng, J.C. (2014). Simultaneous closed-loop tuning of cascade controllers based directly on set-point stepresponse data. Journal of Process Control, 24(5), 652– 662. Jeng, J.C. and Lee, M.W. (2012). Simultaneous automatic tuning of cascade control systems from closed-loop step response data. Journal of Process Control, 22(6), 1020– 1033. Kaya, İ. and Nalbantoğlu, M. (2016). Simultaneous tuning of cascaded controller design using genetic algorithm. Electrical Engineering, 98(3), 299–305. Kaya, I., Tan, N., and Atherton, D.P. (2007). Improved cascade control structure for enhanced performance. Journal of Process Control, 17(1), 3–16. Kessler, C. (1955). Über die vorausberechnung optimal abgestimmter regelkreise teil 111. die optimale einstellung des reglers nach dem betragsoptimum. Regelmgstechnik, 40–49. Krishnaswamy, P.R., Rangaiah, G.P., Jha, R.K., and Deshpande, P.B. (1990). When to use cascade control. Industrial & engineering chemistry research, 29(10), 2163–2166. Lee, Y., Oh, S., and Park, S. (2002). Enhanced control with a general cascade control structure. Industrial & engineering chemistry research, 41(11), 2679–2688. Onat, C. (2013). A new concept on PI design for time delay systems: weighted geometrical center. International Journal of Innovative Computing, Information and Control, 9(4), 1539–1556. Onat, C. (2019). A new design method for PI-PD control of unstable processes with dead time. ISA Transactions, 84, 69–81. Raja, G.L. and Ali, A. (2015). Improved tuning of cascade controllers for stable and integrating processes with time delay. Michael Faraday IET International Summit, Kolkata, India. Tan, N. (2009). Computation of stabilizing PI-PD controllers. International Journal of Control, Automation and Systems, 7(2), 175–184. Tan, N., Kaya, I., and Atherton, D.P. (2003). Computation of stabilizing PI and PID controllers. In Proceedings of IEEE Conference on Control Applications, volume 2, 876–881. IEEE. Tan, N., Kaya, I., and Atherton, D.P. (2005a). A graphical method for computation of all stabilizing PI controllers. IFAC Proceedings Volumes, 38(1), 349–354. Tan, N., Kaya, I., Yeroglu, C., and Atherton, D.P. (2006). Computation of stabilizing PI and PID controllers using the stability boundary locus. Energy Conversion and Management, 47(18-19), 3045–3058. Tan, W., Liu, J., Chen, T., and Marquez, H.J. (2005b). Robust analysis and pid tuning of cascade control systems. Chemical Engineering Communications, 192(9), 1204–1220. Umland, J.W. and Safiuddin, M. (1988). Magnitude and symmetric optimum criterion for the design of linear control systems-what is it and does it compare with the others? In Conference Record of the IEEE Industry Applications Society Annual Meeting, 1796–1802. IEEE. Veronesi, M. and Visioli, A. (2011). Simultaneous closedloop automatic tuning method for cascade controllers. IET Control Theory & Applications, 5(2), 263–270. Vivek, S. and Chidambaram, M. (2013). Cascade controller tuning by relay auto tune method. Journal of the Indian Institute of Science, 84(3 & 4), 89. Vrančić, D., Peng, Y., and Strmčnik, S. (1999). A new PID controller tuning method based on multiple integrations. Control Engineering Practice, 7(5), 623–633. Vrancic, D., Strmcnik, S., and Jurcic, D. (2001). A magnitude optimum multiple integration tuning method for filtered PID controller. Automatica, 37(9), 1473– 1479.