International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 Interaction reducer for closed-loop control of multivariable systems M.Bharathi1 and C. Selvakumar+ 1 Department of Electronics & Instrumentation Engineering, Bharath University,Selaiyur, Chennai – 600 073, India. + Department of Instrumentation & Control Engineering, St. Joseph’s College of Engineering, Chennai – 600 119, India. Abstract: In process multi-input multi-output (MIMO) systems are observed that are dominated by interactions between input and outputs. Process decouplers are synthesized to reduce these interactions before controller design. In this work analytical expressions for closed-loop interactive transfer functions are found and interaction reducers (IR) are designed (as process decouplers) for closed-loop control to minimize the interactions between input/outputs of distillation column. Multiloop PI controllers are tuned using different tuning rules and closed-loop systems with IRs are simulated. The closed-loop results reveal that undesirable responses due to interactive transfer functions are made negligible. Key words: Multivariable control, PI control, Interaction, Sequential design, Decouplers 1. Introduction Chemical and biochemical processes invariably exhibit large dead time characteristics and process non-linearity, which makes the controller design complex. Moreover the existence of noise and process interactions generally adds up to the problem. One of the objectives of multivariable control is to maintain several controlled variables at independent set points so that the process operates safely. The controller must be designed in such a way that it takes care of load changes and will be capable of rejecting disturbances effectively. Interactions between input and output cause a manipulated variable to affect more than one controlled variable and hence MIMO processes are difficult to control. The task of developing a satisfactory design procedure for multivariable PI type process controllers remains a difficult problem. Niederlinski(1971) propose a generalized heuristic method to tune PID controllers based on Zeigler-Nichols criteria. But as it is complex and yield poor performance, it is rarely use. Multiloop PI controllers were designed using biggest log ISSN: 2231-5381 http://www.ijettjournal.org Page 1 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 modulus tuning(BLT) for ten distillation ttcolumns (Luyben,1986) of different input output structures. This BLT method was refined by Basuldo and Marchetti(1990) where each individual loop was tuned with IMC rule to satisfy closedloop stability and performance.Chiu and Arkun (1992), Shen and Yu (1994) and Shiu and Hwang(1998) proposed sequential design method for controller design of multivariable processes that reduces the MIMO problem to a SISO problem. This method of tuning is an efficient method indeed. A transfer function matrix or, equivalently, a set of individual transfer function, is formulated that simplifies the design of controller. Recently, after selecting of proper input output paring, various controller structures like centralized, decentralized and decoupled were explored and tuning methods like IMC-PI and BLT methods were used by (Selvakumar et.al,2006) to evaluate performances of 2x2, 3x3 distillation columns in terms of IAE. They also studied parametric sensitivity and related robustness issues on the stability of closed loop system. Two schemes for iinverted decoupling technique using internal model control tuning were proposed (Chen & Zhang 2007) to control multivariable systems with delays and zeros.Decentralized robust PI controller was designed (Labibi et al.2009 ) using closed-loop stability criteria for industrialboiler. Control structure (configuration) was selected (He et.al., 2009) by interaction analysis using relative normalized gain array (RNGA) criteria that uses both steady state and transient perspectives. Decoupling control of electro pneumatic pressure system tank with different conditions are explained by Sompol et.al(2009). After isolating input and output pairs of a multivariable system, relative gain (ii) array (Bristol) is analyzed to find out correct input-output pairing for effective control. If ii is far from unity, then decentralized controller may not show good results, and it is preferred to use decoupler. By the method of decoupling, we decouple the process and then design a suitable controller for decoupled process. Thus there exist two relative gains : one for the process before decoupling (ii far away from 1) and the other for the process after decoupling (ii=1).But, in the present method of interaction reducer, interaction transfer functions (ITF) are found through sequential design and then a block is designed and added to counter-act the ITF and, thus in this method, the above mentioned problem can be avoided. The objective of the present article is to find out analytical expressions for interaction transfer functions and hence derive interaction reducer and then use PI control to study the closed-loop performance of the MIMO system. ISSN: 2231-5381 http://www.ijettjournal.org Page 2 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 Thus the rest of the paper is arranged as follows: Principle of designing IR for a 2x2 system is presented in section 2. Section 3 discusses simulation results of implementing an IR on WB column. Comparison of performances with IR and decouplers are also brought for perusal in this section. The conclusion is drawn at the end. 2. Design of Interaction Reducer’s transfer function 2.1 Closed-loop interactions Figure 1. Schematic representation of a 2 X 2 multivariable system For a 2x2 mimo system shown in Figure 1, one can easily find the sequential transfer functions as G G G22GC 2 1 as G11,CL G11 1 12 21 G11 1 1 G11G22 I G22GC 2 11 (s ) (1) G G G11GC1 1 and G22,CL G22 1 12 21 G22 1 1 G11G22 I G11GC1 22 ( s ) (2) Where suffix CL represents closed-loop. Using sequential design approach ( Shen, 1994), it is easy to synthesize controllers Gc1 and Gc2 individually in a sequential manner. In this case, the interactive transfer function can be found as ISSN: 2231-5381 http://www.ijettjournal.org Page 3 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 y G12 G12,CL 1 u2 with loop1 closed 1 GC1G11 (3) y G21 G21,CL 2 u1 with loop 2 closed 1 GC 2G22 (4) Equations (3) and (4) gives interaction measure in closed loop sense. In order to reduce the undesirable response, we need to minimize this interaction. The subsequent transfer functions are developed in latter sections. 2.2 Decoupled Structure A decoupler is a device that eliminates interaction between manipulated and controlled variables by, in effect, changing all the manipulated variables in such a manner that only the desired controlled variable changes. This structure has additional elements called decouplers to compensate for the interaction phenomenon. When RGA shows strong interaction then a decoupler is designed. But however decouplers are designed only for orders less than 3 as the design procedure becomes more complex as order increases. Different types of decoupling techniques (Deshpande and Ash, 1988) employed here are, (i) Static decoupling, in which only steady state gains are considered in designing the decouplers and are always physically realizable and easily implemented with the less process information available. A disadvantage is that control loop interactions still exist during transient conditions. (ii) Dynamic decoupling, in which the complete process information is taken into account in the decoupler design. But this may not be feasible always because of the presence of nonlinearities. (iii) Partial decoupling, in which decouplers are designed for the loops where significant interaction exists. It is preferable in control problems where one of the control variables is more important than the other or where one of the process interactions is weak or absent. But it may be less sensitive to modeling errors than complete decoupling. By principle, in a two-variable interacting system, shown in Figure 1, we can write: ISSN: 2231-5381 http://www.ijettjournal.org Page 4 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 Y1 (s) G11 (s)M1 (s) G12 (s)M 2 (s) (5) Y2 (s) G 21 (s)M1 (s) G 22 (s)M 2 (s) The decoupler inputs are two fictitious manipulated variables u 1 and u2 , and its outputs are the real manipulated variables M1 and M2 , as shown in Figure 2. We want to design the decoupler so that u1 affects only Y1 and u2 affects only Y2. Figure 2. Schematic of a conventional decoupler for a 2 X 2 system From Figure 2. it can be seen that the decoupler equations can be written as M1 ( s) D11 ( s)u1 ( s) D12 ( s)u2 ( s) M 2 ( s) D21 ( s)u1 ( s) D22 ( s)u2 ( s) (6) For ease in building the decoupler, let us specify that D11(s)= D22(s) =1. In this case, the decoupler equations become: ISSN: 2231-5381 http://www.ijettjournal.org Page 5 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 M1 (s) u1 (s) D12 (s)u 2 (s) (7) M 2 (s) D 21 (s)u1 (s) u 2 (s) Substituting Equation (7) into Equation (5) gives the equations for the process-decoupler combination: Y1 G11 G12 D 21 u1 G11D12 G12 u 2 Y2 G 21 G 22 D 21 u1 G 21D12 G 22 u 2 (8) For complete decoupling, we want Y1 to be affected only by u1 and Y2 only by u2; that is, After introducing the decouplers D12 ( s) G12 ( s) G11 ( s) and D21( s) G21( s) in Equation(8) G22 ( s) becomes Y1 G11 G12 D21 u1 (9) Y2 G21 G22 D21 u2 But by introducing decouplers the interaction got reduced but the decouplers introduce another loop that affects Y1 and Y2 as indicated by Dashed and Dash-Dot line in Figure 3. By introducing the decoupler D12(s) the interaction of the loop 2 from loop1 got reduced but the effective process gain of loop1 is halved. Similarly by introducing the decoupler D21(s) the interaction of the loop1 from loop 2 got reduced but the effective process gain of loop2 is halved. So decoupler produces an inhibiting or competing effect to the normal process path. To overcome this, the proposed method introduces two more interaction reducers IR1 and IR2 along with D12(s) and D21(s) which reduces the interaction and but it does not affect its own response. With these, the output (from Eqn.9) from propose decoupled structure becomes: ISSN: 2231-5381 http://www.ijettjournal.org Page 6 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 Figure 3 2 X 2 Multivariable system with process decouplers Y1 G11IR11 G12IR 21 u1 G11IR12 G12IR 22 u 2 (10a) Y2 G 21IR11 G 22IR 21u1 G 21IR12 G 22IR 22 u 2 (10b) In order to reduce interactions, Y1 should be affected only by u1 and Y2 should be affected only by u2. Thus the modified decouplers (proposed interaction reducers) can be synthesized by making the corresponding interactive terma null (=0) from which we obtain G IR IR11 1 12 21 G11 … (11a) G12G 22 IR12 G11G 22 G 21G12 … (11b) G 21G11 IR 21 G11G 22 G 21G12 … (11c) G IR IR 22 1 21 12 G 22 … (11d) The output or the desired response from the above structure can be derived as Y1 G11 IR1 u1 Y2 G21 IR2 u2 Y1 (G11)u1 … (12a) Y2 (G 22 )u 2 … (12b) ISSN: 2231-5381 http://www.ijettjournal.org Page 7 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 Thus, one can observe from Eqn (4) that, output can be retained to its desired value if denominator of RHS of Eqn(4) were larger, i.e., y y Y1 1 IR 21 2 u1 CL u1 (13) In the above equation, LHS represents totaloutput and RHS is the sum of desirable (1st part) and undesirable (2nd part) parts. This reveals that IR21 should have a very high gain for minimizing interactions. Thus by multiplying IR12 with G12 and IR21 with G21 we get desired interaction reducers. Thus the schematic diagram with interaction reducer becomes (Figure 3(b)). A close look to Equation (13) and Figure 3(b) reveals that IR1 is proportional to D21 and IR2 is proportional to D12. Thus it is possible to design these interaction reducers and implement them as shown in Figure 3(c). G11 U1 Y1 IR1 IR2 U2 Y2 G22 Figure 3.(b) Schematic representation of proposed Interaction reducers for multivariable systems ISSN: 2231-5381 http://www.ijettjournal.org Page 8 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 Y1 - R1 Gp11 Gc1 IR1 Gp21 IR2 Gp12 R2 Y2 Gp22 Gc2 Figure 3.(c) Schematic of proposed control scheme with interaction reducers 2.3 Controller tuning The PI controller is of multiloop structure and based on first order plus dead time (FOPDT) model (K p e Ds / s 1) structure the parameters are tuned using following formula for IMCLaurent tuning Kc I K P Dp (14) D 2p 2 D p I p 2 p D D 3p 3 Dp ISSN: 2231-5381 2 I (15) D 2p I D p http://www.ijettjournal.org (16) Page 9 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 For BLT tuning, K ci K ZNi and Ii F ZNi where F varies from 2 to 5 and K ZNi and ZNi F are Ziegler-Nichols tuning parameters. Sequential tuning is based on Shen & Yu (1994) method and IMC tuning is based on following formula Kc p K p and i p where max 1.7,0.2 p Results and discussion: The present scheme was implemented on a WB to study closed-loop performance through simulation. A unit step change was first applied to the set point of the first loop, and the process responses were recorded, then another step change was applied to the second loop. The PI controller parameters designed using BLT criteria are shown in table 1. A comparison of the closed loop results (Figure 2 to Figure 5 as well as IAE values in Table 2) reveals that the proposed method produces significantly less interactions when compared to those of decentralized PI controllers and comparable with decoupled method and also the proposed scheme for setpoint changes in product size, the set points are reached quickly and smoothly with less overshoot and offset. It seen from Fig 2 (Y1) that closed-loop (with BLT tuning) response with IR (Modified decoupler) scheme yields smoother response both in set point as well as load change cases. This can be also observed from Table 2 where we see that IAE value with modified decoupler (IR) scheme gives IAER=1.736 and IAEL=0.0302 that are least when compared to that with other schemes , decoupler and decentralized . Table.2 gives the comparison of the integral absolute error (IAE) value which is computed as the sum of the absolute difference between the closed-loop process response and the final steady state value. The performance of modified decoupler (using IR) gives a better performance. Figure 3 (with sequential tuning) shows with decoupler scheme, the setpoint (Y2) response oscillates and becomes unstable whereas with IR schemes. It is stable and takes less time settle as it is evident from load response also. In Figure 4, with IMC tuning an IR scheme, least oscillations and fast settling compared to other strategies, like decoupler and decentralized. At the same time if we see Table 2, we find modified decoupler scheme (IR) produces least values of IAEs (both during set point and load change).When a step change in Y2 is given, the responses obtained are shown in Figure 5. This figure also reveals that with IR scheme, a better performance can be achieved. ISSN: 2231-5381 http://www.ijettjournal.org Page 10 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 Figure 2 .Closed Loop response for a step change in Y1,set (with BLT Tuning on WB Column) Decoupler(Solid line),Modified decoupler (Dashed line),Decentralised (Dash-dot line) Figure.3 Closed Loop response for a step change in Y2,set (with BLT Tuning on WB Column ) Decoupler(Solid line),Modified decoupler (Dashed line),Decentralised (Dash-dot line) ISSN: 2231-5381 http://www.ijettjournal.org Page 11 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 Figure 4. Closed Loop response for a step change in Y1,set (with Sequential Tuning on WB Column) Decoupler(Solid line),Modified decoupler (Dashed line),Decentralised (Dash-dot line) Figure 5 Closed Loop response for a step change in Y2,set (with Sequential Tuning on WB Column ) Decoupler(Solid line),Modified decoupler (Dashed line),Decentralised (Dash-dot line) ISSN: 2231-5381 http://www.ijettjournal.org Page 12 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 WB Table 1: Tuning results for various distillation column examples BLT tuning Sequential Independent IMC autotuning tuning 0.375 0.868 0.737 K c I -0.075 8.29 -0.0868 3.246 -0.103 17.2 Table 2: IAE value comparisons for a Wood and Berry(WB) column Method For step input Y1,set For step input Y2,set BLT Tuning Decoupler 3.477 0.1249 0.0001185 28.67 Sequential Autotuning Modified Decoupler 1.736 0.0302 0.0006164 15.82 Decentralised 3.31 14.67 3.049 28.41 Decoupler 6.231 0.04321 0.1413 12.36 Modified Decoupler 0.2922 0.0009452 0.0001073 6.177 Decentralised 0.5851 6.368 0.5676 12.34 Decoupler 3.634 0.06555 0.003942 15.71 1.823 0.006035 0.0007742 7.957 3.591 8.058 3.418 15.59 IMC Tuning Modified Decoupler Decentralised The present IR control scheme is used in closed-loop (with different PI tuning schemes, BLT, IMC and Sequential autotuning) with WB column for robustness studies. There exists trade-off between robustness and fast response. If anyone wants to have faster response then he has to sacrifice on sacrifice on stability. Similarly, if anyone needs to have better stability then he may to be satisfied with a sluggish response . In order to achieve desired performance under model uncertainties, a good controller has to be robust. We analyze robust stability and compute stability margins for the case in which parameters linearly in the state space matrices. With variation on process parameters, the sensitivity of the closed-loop transfer function is observed and results obtained. The sensitivity of the closed loop system is determined by ISSN: 2231-5381 http://www.ijettjournal.org Page 13 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 perturbing parameters of state space system to +10%. Process for study is selected as WB column. The controller is multiloop PI controller tuned with three different schemes, namely, BLT, IMC and sequential autotuning. In the IMC case, tuning parameters are calculated based on independent diagonal process transfer functions. With the IR scheme under sequential autotuning case, the results show that there are enough scopes to increase the peak gain to make the system response faster, or in other words, the present controller can accept more degree of perturbation compared to other strategies. Thus it provides a wider stability margin. Conclusion: Analytical expressions for closed loop transfer function representing response with direct input-output relation and with the interaction phenomena are achieved using sequential identification techniques/method. The closed loop interaction transfer function (gij CL , with ij) gives information regarding interactions present between input/output of the system. The present method illustrates synthesizing a decoupler like block (interaction reducer, IR) to reduce the interaction between input-output for MIMO systems. Internal model control based PID tuning rules are developed an implemented on MIMO systems using sequential tuning criteria. The present IR scheme with sequential autotuning is suggested to be roust and can be implemented even for multivariable systems with strong interactions. REFERENCES [1] Basualdo,M.S,Marchetti,J.L.Tuning Method for interactive multiloop IMC,PI and PID controllers, Chem.Eng.Commun,1990,97,47. [2] Bristol, E.H., On a new measure of interaction for multivariable process control, IEEE Trans Automatic Control, AC-11, 133-134, 1966 [3] Chen P. and Zhang W, “Improvement on an inverted decoupling technique for a class of stable linear multivariable processes, ” ISA Trans., vol.no.2,pp.199-210,2007. [4] Chiu,M.S.; Arkun,Y. A Methodology for Sequential design of Robust Decentralized Control Systems. Automatica, 28,997,1992. [5] Deshpande, P.B., Ash, R.H., Computer Process Control with Advanced Control Applications, 2nd Ed., ISA, North Carolina, 1988. ISSN: 2231-5381 http://www.ijettjournal.org Page 14 International Journal of Engineering Trends and Technology (IJETT) – Volume3 Issue 4 Number2–Aug 2012 [6] Labibi B,Marquez H.J.and Chen T., “Decentralized robust PI controller design for an industrial boiler ,” J.Proc. Control., vol 19 no.2,pp.216-230,2009. [7] Luyben, W.L., Process Modeling, Simulation and Control for Chemical Engineers, McGrawHill, New York, 1990. [8] Luyben, W.L., Simple Method for Tuning SISO controllers in Multivariable Systems, Ind. Eng.Chem.Process Des. Dev. 25, 654-660, 1986. [9] Selvakumar, C., Bharathi, M., Panda R. C., A comparative test study of simple PI controllers for multivariable systems.AMSE, 61(4),31-50,2006. [10] Shen S.H. and Yu, Cheng-Ching, Use of relay-feedback test for automatic tuning of multivariable systems, AIChE Journal, 40(4), 627-46, 1994. [11] Shiu S.J.and Hwang, S.H, Sequential design method for multivariable decoupling and multiloop PID controller ,Ind.Eng. Chem.Res., 37 107-19,1998. [12] sompol,A. Chitwong,S and Nilas p., “Decoupling control of Electro-pneumatic pressure tank system”, Proc. Internatinal Multi Conf of Engrs & Comp Scientists (IMECS),Vol. II, Mar 18-20,Hong Kong ,2009. [13] Ziegler,J.G.; Nichols,N.B. Optimum Setting for Automatic Controller. Trans.ASME., 64,759,1942 ISSN: 2231-5381 http://www.ijettjournal.org Page 15