Anda di halaman 1dari 8

ISA Transactions 51 (2012) 514–521

Contents lists available at SciVerse ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Design of PID controllers in double feedback loops for SISO systems with
set-point filters
V. Vijayan a , Rames C. Panda b,∗
a
Department of E&I, St Joseph’s College of Engg, IT Highway, Chennai – 600 119, India
b
Department of Chemical Engineering, CLRI (CSIR), Near IIT Madras, Adyar, Chennai – 600 020, India

article info abstract


Article history: A PID controller is widely used to control industrial processes that are mostly open loop stable or unstable.
Received 9 November 2011 Selection of proper feedback structure and controller tuning helps to improve the performance of the
Received in revised form loop. In this paper a double-feedback loop/method is used to achieve stability and better performance of
1 March 2012
the process. The internal feedback is used for stabilizing the process and the outer loop is used for good
Accepted 14 March 2012
Available online 9 April 2012
setpoint tracking. An internal model controller (IMC) based PID method is used for tuning the outer loop
controller. Autotuning based on relay feedback or the Ziegler–Nichols method can be used for tuning an
Keywords:
inner loop controller. A tuning parameter (λ) that is used to tune IMC-PID is used as a time constant of a
Overshoot setpoint filter that is used for reducing the peak overshoot. The method has been tested successfully on
PID controller many low order processes.
Set point filter © 2012 ISA. Published by Elsevier Ltd. All rights reserved.
Tuning
IMC

1. Introduction (2) some performances are difficult to achieve for these processes.
These systems show unusual overshoot or inverse response due
Classical PID controllers are widely used in process industries to the presence of negative or positive zeros. Moreover, the
despite continued advancement in control technology. Most of presence of dead-time in the process makes the system uncertain.
the industrial loops are controlled by PID regulators due to Many researchers have synthesized the internal model controller
their simple structure, near optimal performance and robustness, to study closed-loop performances of these systems. Moreover,
applicability over wide range and ease in implementation and due to ease in implementation and maintenance, PID forms of
maintenance on analog or digital platform. There are many these controllers have received more attention. Hence, in order to
industrial processes, which produce undesirable peak overshoot achieve stable closed-loop response and safe operation, controller
that needs to be eliminated to achieve desired safety and design for such open loop unstable processes has become
economic (process input or controller output must be optimally challengeable and interesting. De’Paor and O’Malley [9], Rotstein
synthesized) norms. The problem can be addressed by synthesizing and Lewin [10], Venkatasubramaniam and Chidambaram [11], Ho
alternate loop structures and by tuning controllers properly. Many and Xu [12], Luyben [13], Huang and Chen [14], Jung et al. [15],
researchers [1–8] proposed PID tuning rules to control various Visioli [16], Yang et al. [17], Saraf et al. [18], Tan et al. [19], and
stable systems by different methods to improve closed loop Panda [20] proposed design of PID controllers for unstable systems.
performance. Open-loop unstable systems are generally observed In all the above works, attention was imparted to improve closed-
in process industries (exothermic stirred reactors with back loop performances (evaluated by error criteria) through different
domains. But these methods involve cumbersome calculations,
mixing, batch reactors, pump with liquid storage tank, combined
need user defined multiparameters and do not reduce overshoot
feed/effluent heat exchanger with adiabatic exothermic reactor
much in closed-loop responses. Many works have been reported
etc.) to operate at unstable steady states in order to achieve
to make the closed-loop response faster by designing appropriate
safety and performance. These unstable systems are difficult to
set point weight (mostly for stable systems) while they do not
control compared to open-loop stable processes mainly because
reduce overshoot satisfactorily of closed-loop responses. Later on,
(1) unstable systems are hard to stabilize for unstable poles,
loop performance was enhanced by employing setpoint filter on
the existing classical structure. Lee et al. [24] considered two-
loop control to reduce undesirable overshoot by using an ideal PID
∗ Corresponding author. Fax: +91(44)24911589. controller and setpoint weight. However, their method must be
E-mail address: panda@clri.res.in (R.C. Panda). associated with the method of design of original error feedback
0019-0578/$ – see front matter © 2012 ISA. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.isatra.2012.03.003
V. Vijayan, R.C. Panda / ISA Transactions 51 (2012) 514–521 515

and their setpoint filter needs information on process parameters.


Shamsuzzoha and Lee [21] designed setpoint filters to improve
loop performances by IMC Maclaurian PID controller. These filters
are of higher order and involve complicated calculations for design.
Zhang [22] also proposed a method to design a simple set point
Fig. 1. Basic structure of proposed scheme for closed-loop control.
filter to reduce the peak overshoot. Recently Nie et al. [23] derived
PID tuning rules and implemented compensator based on gain
and phase margin specifications to reduce peak overshoot. Results So the outer controller Gc2 is given by,
from the above methods show that undesirable overshoot still
exists in closed-loop response. Moreover, many of the above 1 GDCl
Gc2 = . (4)
methods involve cumbersome calculations. Hence an easy and Gp1 (1 − GDCl )
efficient method must be looked for reducing the overshoot. An
open loop process needs a loop for stabilization at first stage. Substituting Eqs. (2) and (3) in Eq. (4),
Most of the unstable systems and processes with numerator zero
as2 + bs + c + k(fs + g )e−ds
produce overshoot mainly due to improper tuning. Hence, new Gc2 =
design procedures are proposed here to achieve desired overshoot k(fs + g )
using simple calculations. The advantage of this procedure is that 1
it avoids tedious and cost effective modelling and identification × . (5)
(ms + 1)(ns + 1) − e−ds
procedures. The objective of this research is to design PI/PID
controller for low order systems and to reduce their overshoot The time delay can be expanded in an infinite series as e−ds =
by employing set point filters. The system can be made stabilized 2 2 3 3
1 − ds + d 2s! − d 3s! + · · ·.
by restructuring loops. Several examples comprising of FOPDT,
So, Eq. (5) becomes as in Box I.
SOPDT, Integrating and unstable systems are chosen to implement
the present method and results are achieved in this study. The Simply Gc2 can be written as,
controllers are implemented in double feedback loop. The inner
φ(s)
feedback loop stabilizes the loop while the outer loop provides Gc2 = (7)
enhanced performance to the closed-loop system. This is again s
augmented with a set-point filter. Thus the entire paper is where, φ(s) is defined as in Box II.
organized as follows. Section 2 discusses principle of designing Expanding Eq. (7) according to Laurent series [8]
proposed PID controller. Results and discussions are presented
φ(s) φ ′′ (0)s2
 
in Section 3. Closed-loop results related to set-point changes 1
are discussed in this section. Results related to load disturbance, Gc2 = = · · · + φ(0) + φ (0)s + ′
+ ··· . (9)
s s 2!
measurement noise and stability analysis are also proposed in
section. Experiments on real time system is presented in Section 4. The standard form of PID controller is,
Conclusion is drawn at the end.  
1
Gc = kc 1+ + τD s . (10)
2. IMC-PID design τI s
Comparing coefficients of s terms of Eqs. (9) and (10) as ex-
The closed loop structure of the proposed scheme is shown
plained in [8,20] one obtains
in Fig. 1 where Gp is the process transfer function (either stable
or unstable or integrating type) with a, b, c , d, f , g as constant kc φ ′′ (0)
coefficients and Gc1 and Gc2 are two controllers. Here, Gc1 = kc = φ ′ (0) τI = and τD = . (11)
Kc1 (a proportional controller). The inner loop controller, Gc1 is φ(0) 2kc
proportional controller alone. It is used to stabilize the process Substituting s = 0, in Eq. (8) and in its derivatives one gets
and is tuned by either relay feedback method or Ziegler–Nichols
method. The outer loop Gc2 is obtained by IMC method that is N
explained below. Let us consider a process with general transfer
φ(0) = (12)
D
function given as
N1 ∗ D − D1 ∗ N
φ ′ (0) = (13)
kp (fs + g )e−ds D2
Gp = . (1)
as2 + bs + c D(N2 ∗ D − D2 ∗ N ) − 2D1 (N1 ∗ D − D1 ∗ N )
φ ′′ (0) = (14)
For the inner-loop, closed loop transfer function is given by, D3
where, k = kc1 ∗ kp ; N = c + kg; N1 = b − dgk + kf ; N2 =
y k(fs + g )e−ds
Gp1 = = (2) 2a + d2 kg − 2dkf
r1 as2 + bs + c + k(fs + g )e−ds
D = k(dg + gm + gn);
where, k = kc1 ∗ kp .
d2 g
 
Let the desired closed loop transfer function for the entire block
D1 = k df + fm + fn − + gmn ;
diagram become, 2
3
 
e−ds d g
GDCl = . (3) D2 = k −d2 f + + 2fmn .
(ms + 1)(ns + 1) 3

Here, m = λ and n = 0 for first order system and m = n = λ Eqs. (11)–(14) can be used to compute PID parameters. Here λ
for the second order system. is used to tune the PID parameter which also used as filter time
λ is a tuning parameter used for tuning kc , τI and τD . It is also constant in setpoint filter. Same value of λ is to be used both in PID
considered as a time constant of simple set point filter [22]. parameters and setpoint filter to get less ITAE value.
516 V. Vijayan, R.C. Panda / ISA Transactions 51 (2012) 514–521

as2 + bs + c + k(fs + g )e−ds


Gc2 =     2  . (6)
d2 g d3 g
sk dg + gm + gn + s df + fm + fn − 2
+ gmn + s2 − d2f + 6
+ fmn

Box I.

as2 + bs + c + k(fs + g )e−ds


φ(s) =     2  . (8)
d2 g d3 g
k dg + gm + gn + s df + fm + fn − 2
+ gmn + s2 − d2f + 6
+ fmn

Box II.

3. Results and discussion overshoot (PO) reported was 1.009. (For the proposed method,
the process is converted into the form similar to Eq. (1) with
The examples considered for simulation are selected from kp = 2, d = 1, a = 50, b = 15, c = 1, f = 0, g = 1).
stable, unstable and integrating type processes and are provided Using present method, a setpoint filter is designed with λ = 0.7.
in Table 1. Process types or examples, methods of tuning, PID Closed loop simulation resulted as PO of lesser value, 1.0067, and
parameters, filters and performance values are provided in the better performance values (Table 1) are obtained compared to
Table 1. Fig. 2 contains set-point responses (closed-loop) for Shamsuzzoha and Lee [21] who used a second order filter.
different examples considered in Table 1. Detailed results with
different examples are given as follows: Example 3 (SODUP (One Unstable Pole)). As an unstable SODUP
process [21], let us take Example 3 as mentioned in Table 1. The
Example 1a (Stable FOPDT Process). An example of a stable FOPDT PID controller settings used are kc = 6.7051, τI = 5.4738,
process [22] having transfer function (Table 1), G(s) = s+1 1 e−2s τD = 1.333 and a second order filter with transfer function
1.6421s+1
is chosen for study. Comparing this transfer function with that of GF = 7.2966s 2 +5.4738s+1 was used that yielded a PO of 1.03 (Table 1).

Eq. (1), we find, kp = 1, d = 2, f = 0, g = 1, a = 0, b = Whereas by using the present method, a first order filter was
1c = 1. Naturally, in this case, m = λ and n = 0. Using relay designed whose λ = 0.6. (For the proposed method, the process
feedback method, a proportional controller is designed with kc1 = is converted into the form similar to Eq. (1) with kp = 1, d =
0.76 for the inner loop. This will make the loop to respond faster. 0.939, a = 10.35, b = 2.93, c = −1, f = 0, g = 1). After the
Using all the above values in Eqs. (11)–(14), the PID parameters closed-loop simulation with same PID settings, better performance
will be obtained. The λ value is adjusted simultaneously both in values are obtained with less PO value of 1.0065.
PID equation (11) and in setpoint filter until it produces less ITAE
Example 4 (SODUP (One Unstable Pole)). As an unstable SODUP
values. Thus, PID parameters are obtained as kc = 0.3147, τI =
process [23], let us take Example 4 as mentioned in Table 1. The
0.3942, τD = 1.7141. Using present method, a setpoint filter is
PID controller settings used are c0 = 0.155 + 0.314 and the
designed with λ = 0.9. The overshoot obtained with the present s

method is 1.023. Zhang [22] reported PID controller parameters as compensator is c (s) = 2 00..7462s+1
143s+1
that yielded a PO of 1.0505
kc = 0.91, τI = 2.75, τD = 0.68 and obtained a peak overshoot of (Table 1). Whereas by using the present method, a first order filter
1.022. was designed whose λ = 0.25. (For the proposed method, the
process is converted into the form of Eq. (1) as kp = 1, d =
Example 1b (Unstable FOPDT Process). We consider an unstable 0.5, a = 1, b = 1.5, c = −1, f = 0, g = 1.) After the closed-
FOPDT process [15,22] having transfer function (Table 1), Gp = loop simulation with same PID settings, better performance values
e−0.5s are obtained with less PO value of 1.0269.
s−1
. Comparing this transfer function with that of Eq. (1), we get,
kp = , = . , f = 0, g = 1, a = 0, b = 1, c = −1. Since this
1 d 05 Example 5 (SOPDT with Inverse Response). Next example is chosen
is first order system m = λ and n = 0 are assumed. The value of as a SOPDT process with a zero in numerator that often shows
proportional controller constant, kc1 = 1.268 of inner is obtained inverse response as also was considered by Shamsuzzoha and
by Ziegler–Nichols method. It is a simple proportional controller Lee [21]: The PID parameter was set to be kc = 3.0819, τI =
used in the inner loop to make the system stable and at the same 1.6399, τD = 0.4295 for this Example 5 and a second order filter
time to achieve faster response. By substituting all the above values with transfer function GF = 0.7044s2 +11.6399s+1 was used. They
in Eqs. (11)–(14), then adjusting the tuning parameter λ the PID obtained a PO of 1.0127 and ITAE of 1.135. (For the proposed
parameters will be obtained. The λ is adjusted simultaneously both method, the process is converted into the form of Eq. (1) as kp =
in PID equation (11) and in setpoint filter until it produces less ITAE 1, d = 0.2, a = 1, b = 2, c = 1, f = −0.2, g = 1.) By using the
values. Thus, the obtained PID parameters are kc = 0.3533, τI = present method, a setpoint filter with time constant λ = 0.29 is
1.5046, τD = 0.5166. Using present method, a setpoint filter is obtained and after simulation, an ITAE value of 1.071 is obtained.
designed with λ = 0.4. The overshoot is found to be 1.0014 Thus the performance of the system is improved by the present
with the present method. Jung et al. [15] reported PID controller method. The performance values for this example are given in
parameters as kc = 1.5353, τI = 7.5753 and obtained a peak Table 1.
overshoot of 1.0044 which is slightly more than the present value.
Zhang [22] reported PID controller parameters as kc = 2.6, τI = Example 6 (First Order Delay Integrating Process (FODIP)). Another
1.7826, τD = 0.2473 and obtained a peak overshoot of 1.48895 type of model structure (FODIP) is generally observed in process
which is more than the present value. industries. Due to presence of an integrator, the step response
becomes unstable for these systems. Shamsuzzoha and Lee [21]
Example 2 (Stable SOPDT Process). Consider a stable SOPDT used PID settings kc = 0.3593, τI = 12.13, τD = 2.704 and
process [21] as Example 2 (Table 1) with the PID controller designed a second order setpoint filter GF = 32.8106s2 +112.1304s+1
parameters as kc = 9.8092, τI = 5.4502, τD = 1.6898. The peak that yielded a PO of 1.012 and ITAE value of 87.31 whereas using
V. Vijayan, R.C. Panda / ISA Transactions 51 (2012) 514–521 517

Table 1
Resulting performance for examples.
Process type Method PID parameters Filter parameters Performance

Stable FOPDT kc1 = 0.76, IAE = 3.816


Example 1a kc = 0.3147, ISE = 3.071
Proposed λ = 0.9, GF = 1
0.9s+1
Gp = s+1 1 e−2s τI = 0.3942, ITAE = 8.778
τD = 1.7141 PO = 1.023

kc = 0.91, IAE = 3.51


τI = 2.75, 1.26s2 +1.47s+1 ISE = 2.634
Zhang [22]
τD = 0.68 1.26s2 +1.86s+1 ITAE = 9.477
PO = 1.022

Unstable FOPDT kc1 = 1.268, IAE = 1.202


Example 1b kc = 0.3533, ISE = 0.926
e−0.5s
Proposed λ = 0.4, GF = 1
0.4s+1
Gp = s−1
τI = 1.5046, ITAE = 0.8572
τD = 0.5166 PO = 1.0014

kc = 1.5353, IAE = 2.654


τI = 7.5753 1 ISE = 2.002
Jung et al. [15] GF = 7.5753s +1 ITAE = 4.645
PO = 1.0044

kc = 2.6, IAE = 1.355


τI = 1.7826, ISE = 0.8738
GF = 00..4409s +0.6906s+1
2
Zhang [22]
τD = 0.2473 4409s2 +1.7826s+1 ITAE = 1.619
PO = 1.48895

SOPDT (stable) kc1 = 3.6568, IAE = 3.0582


Example 2 kc = 0.4399, ISE = 2.3466
−1s Proposed λ = 0.7, GF = 1
0.7s+1
Gp = (10s+2e1)(5s+1) τI = 0.9287 ITAE = 5.77
τD = 6.8788, PO = 1.0067

kc = 9.8092, IAE = 4.174


τI = 5.4502, 1.6351s+1 ISE = 3.058
Shamsuzzoha and Lee [21] GF = 9.2099s
τD = 1.6898 2 +5.4502s+1
ITAE = 10.89
PO = 1.009

SODUP kc1 = 1.3965, IAE = 2.6663


Example 3 kc = 0.5469, ISE = 2.0924
e−0.9 3 9 s
Proposed λ = 0.6, GF = 1
0.6s+1
Gp = (5s−1)(2.07s+1)
τI = 4.1202, ITAE = 4.2817
τD = 6.7344 PO = 1.0065

kc = 6.7051, IAE = 3.244


τI = 5.4738, 1.6421s+1 ISE = 2.389
Shamsuzzoha and Lee [21] GF = 7.2966s
τD = 1.333 2 +5.4738s+1
ITAE = 6.846
PO = 1.03

SODUP kc1 = 1.3977, IAE = 1.2


Example 4 kc = 0.591, ISE = 0.9456
e−0.5s
Proposed λ = 0.25, GF = 1
0.25s+1
Gp = (0.5s+1)(2s−1)
τI = 2.077, ITAE = 0.9523
τD = 1.4746 PO = 1.0269

IAE=1.955
ISE = 1.418
Nie et al. [23] c0 = 0.155 + 0.314 c (s) = 2 00..7462s +1
143s+1
s ITAE = 2.893
PO = 1.0505

SOPDTZ kc1 = 2.4438, IAE = 1.1532


Example 5 kc = 0.7493, ISE = 0.8606
(−0.2s+1)e−0.2s
Proposed λ = 0.29, GF = 1
0.29s+1
Gp = (1s+1)(1s+1)
τI = 0.4147, ITAE = 1.071
τD = 0.9562 PO = 1.0359

kc = 3.0819, IAE = 1.395


τI = 1.6399, ISE = 1.106
Shamsuzzoha and Lee [21] GF = 0.7044s2 +11.6399s+1
τD = 0.4295 ITAE = 1.135
PO = 1.0127

FOIPDT kc1 = 0.1318, IAE = 10.5646


Example 6 kc = 0.4909, ISE = 8.3239
−4s Proposed λ = 2.1, GF = 1
2.1s+1
Gp = s(1e4s+1) τI = 4.0251, ITAE = 69.907
τD = 9.6422 PO = 1.0135

kc = 0.3593, IAE = 12.33


τI = 12.13, 1 ISE = 9.897
Shamsuzzoha and Lee [21] GF =
τD = 2.704 32.8106s2 +12.1304s+1 ITAE = 87.04
PO = 1.012
518 V. Vijayan, R.C. Panda / ISA Transactions 51 (2012) 514–521

Fig. 2. Closed loop responses of various process (examples are taken from Table 1).

present procedure, the filter is designed with λ = 2.1. (For After simulating the performance is obtained as ITAE = 69.907.
the proposed method, the process is converted into the form of Thus improved performance values are obtained and are reported
Eq. (1) as kp = 1, d = 4, a = 4, b = 1, c = 0, f = 0, g = 1.) in Table 1.
V. Vijayan, R.C. Panda / ISA Transactions 51 (2012) 514–521 519

Fig. 3. Load rejection response of the process Example 5. Fig. 4. Closed loop response of the process Example 5 under measurement noise.

Table 2 3.3. Stability analysis


Performance for closed-loop scheme with present controller for load disturbance.
Method IAE ISE ITAE For an open loop unstable process, the output (y) keeps on
increasing/decreasing for a change in the input variable (u). An
Proposed 0.3816 0.05722 0.8023
Shamsuzzoha and Lee [21] 0.5402 0.09955 1.174
IMC controller can make the closed-loop stable if the following
conditions are satisfied:
GP GC
(1) GP1 = 1+G G1 must be stable, (2) (1 − GP1 GIMC
C ) GP1 should
P C1
3.1. Load disturbance be stable.
Thus according to condition (2) above, the unstable poles of
Load changes frequently occur in process industries. Distur- GP must be cancelled by zeros of GIMC
C and according to condition
bance rejection is a major criterion in chemical process control. (3), the unstable poles of GP1 must be cancelled by zeros of (1 −
Evaluating the performance of the controller under load changes GP1 GIMC
C ). As the proposed controller is of PID type, the closed-loop
is an important consideration in judging the suitability of con- characteristic equation,
troller. Process model (Example 5) is subjected to load disturbance 1 + GP (s)GC (s) = 0 (15)
of unity with controller in the closed loop. First Gc1 is designed
(KC1 = 2.4438) by ZN method to stabilize inner loop. Second, Gc2 is will hold good for analysing the stability of the closed loop system
KP e−DP s
designed by IMC-PID method. Simultaneously, both setpoint filter (while the open loop process is unstable). Let, GP (s) = τP s−1
,
and Gc2 are tuned by varying the lambda. Lambda is being used as rewriting in the form of Eq. (1)
time constant of setpoint filter and tuning parameter of IMC-PID
KP e−ds y ke−ds
filter. The filter (λ) and PID parameters (kC = 0.7493, τI = 0.4147 GP (s) = , Gp1 = =
and τD = 0.9562) are given in Table 1. Fig. 3 shows the closed-loop bs − 1 r1 bs − 1 + ke−ds
performance for this disturbance rejection case. It can be seen that where, k = kc1 ∗ kp
the present controller is able to eliminate the effect due to the said
KC τD s2 + KC s + τC
K
disturbance much faster compared to the same by using controller r 1 s2 + r 2 s + r 3
proposed by Shamsuzzoha and Lee [21]. The present method gives GC (s) = I
= (16)
s s
an IAE (Table 2) of value = 0.3816 against the same (IAE = 0.5402)
where, r1 = KC τD , r2 = KC , r3 = τC .
K
obtained by Shamsuzzoha and Lee [21] proving the superiority of I

the method over others. The performance values show that present The closed loop characteristics equation is given by, 1 +
method works better even for load disturbance which is shown in GC (s)Gp (s) = 0, expanding exponential delay term up to first five
terms (neglecting remaining higher order terms), we get
Table 2.
p6 s6 + p5 s5 + p4 s4 + p3 s3 + p2 s2 + p1 s + p0 = 0 (17)
3.2. Presence of measurement noise where,
p0 = kr3 , p1 = k + kr2 − dkr3 − 1,
Measurement noise is a common problem in almost all process
industries. Ability of tracking the set point under measurement d2 kr3
p2 = b − dk + kr1 + − dkr2
noise is also a test to evaluate the goodness/suitability of a 2
controller for controlling the process. A process model (Example 5) d2 k d2 kr2 d3 kr3
is selected for this purpose. Accuracy of the presently proposed p3 = + − − dkr1 ,
2 2 6
method was rationalized by employing a white random noise
d2 kr1 d3 k d3 kr2 d4 kr3
with a noise to signal ratio (NSR) of 0.001 at the output level. p4 = − − +
The noisy measured output data were feedback (negative) to the 2 6 6 24
comparator and error between measured data and set point were d4 k d3 kr1 d4 kr2 d5 kr3
p5 = − + − ,
fed to the controller (Gc2 ). The process output is shown in Fig. 4 24 6 24 120
which reveals that the present method is able to design controller d4 kr1 d5 k d5 kr2
that can efficiently overcome the effects of measurement noise. p6 = − −
24 120 120
520 V. Vijayan, R.C. Panda / ISA Transactions 51 (2012) 514–521

Table 3
Values of perturbations on parameters of system for testing stability.
System Process Gain Time constant Time delay All process parameters simultaneously
e−0.5s
G(s) = s− 1
±98% ±100% ±10% ±90%, ±90%, ±10%
Frequency range (r/s) 674 832 9054.8 4199 148 181

Fig. 6a. Experimental setup for two interacting tank level control system.
Fig. 5. Nyquist image for four Kharitonov’s polynomial.

are defined in interval pLi ≤ pi ≤ pUi with pLi as lower bound


and pUi as upper bound of parameters pi where i = 0, 1, 2, . . . , 6.
According to Kharitonov’s theorem, every polynomial (character-
istic equation) in the interval family will be stable (Hurwitz) if the
following four Kharitonov polynomials are Hurwitz.

K11 (s) = pL0 + pL1 s + pU2 s2 + pU3 s3 + pL4 s4 + pL5 s5 + · · · (18)

K12 (s) = pL0 + pU1 s + pU2 s2 + pL3 s3 + pL4 s4 + pU5 s5 + ··· (19)

K21 (s) = pU0 + pL1 s + pL2 s2 + pU3 s3 + pU4 s4 + pL5 s5 + · · · (20)

K22 (s) = pU0 + pU1 s + pL2 s2 + pL3 s3 + pU4 s4 + pU5 s5 + ···. (21)
As the process parameters (KP , τP , DP ) are perturbed, closed- Fig. 6b. Schematic of experimental setup for two interacting tank level control
loop stability changes accordingly the coefficients of Eq. (17) system.
changes. Since the characteristic equation is of 6th order, the value
set will move through 3 quadrants in a counter-clockwise direction fluid. The experiment set-up of coupled tanks is designed so that
and exclude the origin if the family is robustly stable. Fig. 5 shows the system can be configured. The 33-041 coupled tank system
that the value sets of characteristic polynomial (9) that is Hurwitz. is interfaced with computer through MATLAB/SIMULINK and an
It describes that Nyquist image of set of all polynomials of Eq. (17) Advantech PCI 1711 data acquisition interface card. The set-up has
is always bounded by a rectangle whose corners are constructed four translucent tanks each with a pressure sensor to measure the
from the Nyquist images of the four Kharitonov’s polynomials water levels. Two of the liquid level tanks of 3.7 litres capacity
(Eqs. (18)–(21)). The four corner points denote 4 equations (upper each are connected in series (Fig. 6b) where the exit flow from
left (Eq. (19)), upper right (Eq. (21)), lower left (Eq. (18)), and 1st tank enters 2nd tank and level (h2 ) is measured and controlled
lower right (Eq. (20))). It can be easily seen that the Nyquist in 2nd tank by manipulating flow rate of liquid in 1st tank. The
image of the vertex polynomials starts at 3rd quadrant and slowly input is the voltage given to the pump through pump control
moves without passing through origin. With increase in level of unit. The Differential pressure type level transmitter is used to
perturbations (on kp ± 90%, on τp ± 90% and on time delay ±10%) measure the level in the tank. To measure the level in both tanks,
on parameters of the chosen system. These results on robustness two DPT type level transmitters are implemented at the bottom
ability of the controller to parametric changes are shown in Table 3. of the tanks. The hand valve HV1 and HV3 are introduced in the
The image (Fig. 5) tries to move towards 3rd quadrant, ensuring the pipe to adjust the inflow and outflow respectively. The card gives
polynomial as Hurwitz till a maximum frequency of 148 181 rad/s. 0–5 V output corresponding to the level 0–25 cm in the tank. The
sampling time used here is 0.1 s. The controlled inputs are voltage
4. Real-time test applied to pumps so as to adjust the feed flow rate in 1st tank.
The maximum excitation for 100% actuator output is restricted to
The coupled tanks set-up (Fig. 6a) is a model of a liquid storage 3.5 V for corresponding maximum flow rate of 4907.8 cm3 /min.
system in process industries. Often tanks are coupled through For open-loop studies, the level is set at 12 cm in the second tank.
connecting pipes. These storage facilities contain fluids where the The nominal flow for this 12 cm is maintained at 1654.5 cm3 /min
reactant level and flow are to be controlled. Water is chosen as the (equivalent to an excitation of 2.2 V to the pump). Step test is
V. Vijayan, R.C. Panda / ISA Transactions 51 (2012) 514–521 521

5. Conclusion

A simple first order set-point filter and a new IMC based


PID tuning rule is proposed for open loop low order processes.
The tuning parameter lambda (λ) that is used for tuning the
PID controller is used as filter time constant. According to
design objectives, the peak overshoot is reduced and better
closed loop performances like IAE, ISE, and ITAE were obtained.
Several practical processes are selected to implement the proposed
strategy and it has been found that the method works satisfactorily.
The controller efficiently tracks set point and gives desired load
disturbance properties. Results of the stability analysis show that
the present method is able to provide overall closed-loop stability
for an open loop system. The set point filter offers a value addition
in yielding better performances over other controllers available in
the literature. This design method is very simple to implement.
The controller has been tested for disturbance rejection and
Fig. 7. Comparison of closed-loop real time control of liquid-level in two performance under the presence of measurement noise has been
interacting tanks (in series) with setpoint filter (present method with solid line;
Shamsuzzoha and Lee [21], dash–dot line).
found satisfactory.

Table 4 References
Comparison of real-time closed-loop performance details for the control of liquid
level in two-interacting tanks.
Shamsuzzoha and Lee [21] Proposed [1] Cohen GH, Coon GA. Theoretical considerations of retarded control. Trans
ASME 1953;75:827.
ITAE 1.294 × 10 5
8.652 × 104
[2] Astrom KarlJ, Hagglund Tore. PID controller: theory, design and tuning. 2nd
ISE 7382 6967 ed. Research Triangle Park (NC): ISA; 1995.
IAE 1062 943.3 [3] Chen D, Seborg DE. PI/PID controller design based on direct synthesis and
Overshoot 2.4592 0.7130 disturbance rejection. Ind Eng Chem Res 2002;41:4807–22.
of PV in % [4] Chien IL, Fruehauf PS. Consider IMC tuning to improve controller performance.
1.17s+1 Chem Eng Prog 1990;86:33–41.
Filter = 10.37s2 +5.85s+1
KC = 0.5435, τi = 25.118,
Controller [5] Kavdia Mahendra, Chidambaram M. On-line controller tuning for unstable
τD = 2.1008, λ = 0.5
details systems. Comput Chem Eng 1996;20:301–5.
Compensator = 00..4261s
0333s+1
+1 Filter = 0.5s1+1 [6] Skogestad Sigurd. Simple analytic rules for model reduction and PID controller
KC = 0.7678, τi = 5.85,
tuning. J Process Control 2003;13:291–309.
τD = 1.773, λ = 1.5
[7] Dwyer AO. Handbook of PI and PID controller tuning rules. Imperial College
Press; 2006. pp. 392–608.
[8] Panda RC. Synthesis of PID tuning rule using the desired closed-loop response.
carried-out and the input–output data is stored in the computer. Ind Eng Chem Res 2008;47(22):8684–92.
Then the response graph is drawn using MATLAB software. From [9] De Paor AnnraoiM, O’Malley Mark. Controllers of Ziegler–Nichols type for
unstable processes. Internat J Control 1989;49:1273–84.
the graph the time constant of the two tank is calculated. Thus the [10] Rotstein GE, Lewin DR. Simple PI and PID type controllers for unstable Systems.
transfer function of the entire system is derived. Using step test, Ind Eng Chem Res 1991;30:1964.
the open loop transfer function, relating height of liquid in 2nd [11] Venkatasubramaniam V, Chidambaram M. Design of P and PI controllers for
unstable FOPTD model. Internat J Control 1994;60:137.
tank to feed flow rate of 1st tank of the system, is identified as [12] Ho WK, Xu W. PID tuning for unstable processes based on gain and phase
0.0073e−0.0667s
Gp = 59.29s2 +25.65s+1
. margin specifications. IEE Proc 1998;145:392.
[13] Luyben WL. External versus internal open loop processes. Ind Eng Chem Res
After designing inner loop controller (by autotuning, KC1 = 1998;37:2713–20.
1.2361) and outer loop controller by present tuning method (by [14] Huang HP, Chen CC. Control system synthesis for open loop unstable processes
IMC-PID Laurent [8] as KC = 0.5435, τi = 25.118, τD = with delay. IEE Proc 1997;144:334.
[15] Jung CS, Song HK, Hyun JC. A direct synthesis method of unstable first-order
2.1008, λ = 0.5), a setpoint filter is designed with transfer time delay processes. J Process Control 1999;9:265.
function Gf = 0.5s1+1 . This system is used to control liquid level [16] Visioli A. Optimum tuning of PID controllers for integrating and unstable
in the 2nd tank (setpoint tracking). The set point is kept at 12 cm processes. IEE Proc 2001;148:180.
[17] Yang XP, Wang QG, Hang CC. IMC-Based control system design for unstable
level. It is found from Fig. 7 that the ITAE value in this case becomes processes. Ind Eng Chem Res 2002;41(17):4288–94.
8.652 × 104 . In a separate trial (attempt) on the same process, [18] Saraf V, Zhao FT, Bequette BW. Relay autotuning of cascade-controlled open-
controller was designed by Shamsuzzoha and Lee [21] method and loop unstable reactors. Ind Eng Chem Res 2003;42(20):4488–94.
PID parameters were obtained as KC = 0.7678, τi = 5.85, τD = [19] Tan W, Marquez HJ, Chen TW. IMC design for unstable processes with time
delay. J Process Control 2003;13:203–13.
1.773, λ = 1.5. According to their suggestion, a filter with transfer [20] Panda RC. Synthesis of PID controller for unstable and integrating processes.
1.17s+1
function Gf = 10.37s 2 +5.85s+1 and a compensator with transfer Chem Eng Sci 2009;64:2807–16.

function 00..0333s+1 [21] Shamsuzzoha M, Lee Moonyong. Design of advanced PID controller for
4261s+1
were implemented on the real time setup and enhanced disturbance rejection of second-order processes with time delay.
test runs were conducted. It is evident from the Figure that present AIChE J 2008;54(6):1526–36.
method yields faster response with less overshoot compared to [22] Zhang Weidong. Optimal design of the refined Zigler–Nichols proportional—
integral-derivative controller for stable and unstable processes with time
the other method. The performance (ITAE) value in this case was delays. Ind Eng Chem Res 2006;45:1408–19.
recorded as 1.294 × 105 as shown in Table 4. This table shows that [23] Nie Zhuo-Yun, Wang Q-G, Wu M, He Y, Qin Q. Lead/Lag compensator design for
performance of present scheme is better and it needs only a simple unstable delay processes based on new gain and phase margin specifications.
Ind Eng Chem Res 2011;50:1330–7.
setpoint filter which is very easy to design and implement in real [24] Lee Y, Park S, Lee M, Brosilow C. PID controller tuning for desired closed-loop
time case. responses for SISO systems. AIChE J 1998;44(1):106–15.

Anda mungkin juga menyukai