Kaya 2015

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Electr Eng

DOI 10.1007/s00202-015-0352-3

ORIGINAL PAPER

PI-PD controllers for controlling stable processes with inverse


response and dead time
Ibrahim Kaya1

Received: 30 July 2014 / Accepted: 22 November 2015


© Springer-Verlag Berlin Heidelberg 2015

Abstract Many tuning methods for controlling stable 1 Introduction


processes with dead time can be found in the literature. Usu-
ally, proportional-integral-derivative (PID) controllers are Proportional-integral-derivative (PID) controllers are still
used for this purpose. In practice, stable processes with widely used in industrial systems, despite the significant
inverse response characteristics can also be encountered. developments in control theory and technology during recent
Addition of inverse response characteristic makes the con- years. The controller has only three parameters to be adjusted.
trol of a process more troublesome. Therefore, the use of PID Appropriate values for these parameters can be found using
controllers for tuning inverse response processes may lead to many theoretical approaches if a plant transfer function is
unsatisfactory closed loop responses. On the other hand, PI- given or by one of several tuning rules, which can be found
PD controllers have been shown to perform in very good in the literature, based on typical process models. Usually
closed loop responses where PID controllers cannot result the PID controller is taken to have the error as its input to the
in satisfactory closed loop responses. Therefore, this paper closed loop system which produces an undesirable ‘deriva-
adopts the structure previously suggested by the author for tive kick’ at its output for a step input to the feedback loop
controlling processes without inverse response. This adapta- even when the D term has a filter. Also, it is well known that
tion, in conjunction with a simple but powerful factorization, it is difficult to get good closed loop step responses for some
eliminates the affect of the inverse response term in the closed class of processes with PID controllers [1]. On the other hand,
loop transfer function, hence resulting in a simplified design PI-PD controllers have been shown to give improved perfor-
procedure. Tuning parameters of the PI-PD controller used in mance for a class of processes in different control schemes
the control structure are found using standard forms, which [2–6].
is a simple algebraic approach and proven to be resulting in Publications addressing control of different type of
very satisfactory closed loop responses. Simulation exam- processes using PID controllers can widely be found in the
ples are provided to illustrate the superiority of the proposed literature. For example, recent works on PID tuning for stable
method over some existing ones. processes can be found in references [7–12]. Recent studies
on PID tuning for unstable processes can be found in [8,13]
Keywords Inverse response · Smith predictor · PID · and references in there. Publication on tuning PID controllers
PI-PD for integrating processes can be found in [8,9,14]. All these
publications are basically suggested for controlling processes
without inverse response. Hence using them for controlling
processes with inverse response may lead to unsatisfactory
closed loop responses. In order to achieve better closed loop
performances, particular design methods and control struc-
B Ibrahim Kaya tures must be developed for inverse response processes.
ikaya@dicle.edu.tr
Examples of processes with inverse response are usu-
1 Department of Electrical and Electronics Engineering, Faculty ally encountered in chemical process, such as the level of
of Engineering, Dicle University, 21280 Diyarbakır, Turkey drum boiler in a distillation column, the exit temperature of

123
Electr Eng

a tubular exothermic reactor or outlet concentration of an optimization approach which can be found even in some text
autocatalytic CSTR [15,16]. There are only a few works, books. Another approach to direct optimization which has
among them are references [15,17,18] which address tuning been little discussed for many years is the direct synthesis
of processes with inverse response. Luyben [18] suggested approach where the closed loop transfer function is syn-
using a PI (proportional-integral) controller in the classic thesized to a standard form. Interested readers can refer to
single-input single-output (SISO) system. The method sug- references [2–6] for details of standard forms approach and
gested by Luyben [18] has a limitation on the dead time of not its applications in different control structures.
being larger than 3.2. Chien et al. [19] have also suggested In order to obtain the standard form transfer function para-
a tuning method for controlling processes including a zero. meters integral performance criterion defined by
Still, the classical SISO structure with PID controller was
employed. Camacho et al. [20] gave application of variable ∞
structure control to chemical processes with inverse response. Jn = [t n e(t)]2 dt (1)
Balaguer et al. [21] studied identification of inverse response 0
processes using second order model based on transient step
responses. More recently, Jeng and Lin [22] proposed a is used. Here n = 0, n = 1 and n = 2 corresponds to ISE,
PID controller to control stable/integrating processes with ISTE and IST2 E criteria respectively. Minimizing a control
inverse response and time delay. Their proposed control system by ISE (n = 0) usually gives responses with large
system design is based on a Smith-type compensator for oscillations. Hence it is generally not preferred in control-
non-minimum phase dynamics, which aims to remove these ling systems. Increasing n, that is using ISTE (n = 1) or
elements from the feedback loop. IST2 E (n = 2), better closed loop responses can be achieved.
Addition of inverse response makes control of processes Typically, for n = 0, 1, 2 overshoots of 20–30 %, around
more troublesome and particular tuning rules for inverse 5–10 % and <5 % can be expected, with possibly small
response processes must be considered. One way to over- increases in settling time as n decreases [5,6].
come this difficulty is somehow to eliminate the term in the Assuming a plant transfer function with no zero and a
plant transfer function causing inverse response. For this, the controller with a zero then a closed loop transfer function,
structure previously used by the author for processes without T1 j , of the form
inverse response has been adopted. By adopting this structure
two advantages have been gained. First, by introducing a sim- c1 s + 1
T1 j = (2)
ple but powerful factorization, affect of the inverse response s j + d j−1 s j−1 + · · · + d1 s + 1
term in the closed loop transfer function has been eliminated.
This elimination results in a very simplified design procedure is obtained, where the subscript ‘1’ in T1 j indicates a zero
and hence analytical expressions to be used in determining in the numerator of the standard form and the subscript ‘j’
controller tuning parameters. Second, as PI-PD controller is indicates the order of the denominator. Also, for a unit step
used in the structure, improved closed loop performances can set point, the error is obtained as
be achieved for the control system. Tuning parameters of the
PI-PD controller are found using standard forms, which is s j−1 + d j−1 s j−2 + · · · + (d1 − c1 )
E1 j = (3)
a simple algebraic approach to controller design. Another s j + d j−1 s j−1 + · · · + d1 s + 1
advantage of the standard forms method are that it focuses
directly on the closed-loop step response and allows one Minimizing E 1 j for the different performance indices, the
to examine the effects of designing for different response optimum values of the d’s as functions of c1 are shown in
speeds. Fig. 1 for T13 (s) [5]. In the figure, c1 , d1 and d2 are the
The rest of paper is organized as follows. Section 2 coefficients in numerator and denominator of the standard
presents a brief review of standard forms, since it is used to transfer function T13 (s). As it is seen, for different integral
find tuning parameters of PI-PD controller. Controller design criteria very similar d1 values are achieved. However, d2 val-
procedure for inverse response processes is given in Sect. 3, ues show differences for different integral criteria. Also it is
followed by robustness analysis in Sect. 4. Simulation exam- observed that always larger values of d2 are obtained as the
ples are provided in Sect. 5. Finally, conclusions are supplied time weighting in the integral criterion is increased. On the
in Sect. 6. other hand, this is not the case for d1 values.
Figure 2 shows how J0 , J1 and J2 , the minimum values
2 Standard forms for the ISE, ISTE and IST2 E criteria respectively, vary as
c1 increases for T13 (s) [5]. The figure illustrates that as c1
Optimization has always been a powerful tool for identify- increases the step response of the closed loop improves. Also,
ing tuning parameters of a controller. One way is the direct the step responses for the J1 criterion for a few different c1

123
Electr Eng

T13 (s) 1.4


6
ISE
5.5 ISTE 1.2
IST2 E
5
d1 1
4.5

4 0.8

Output
2

d2
d ,d

3.5
1

d2 0.6
3
d1
2.5
d1 d2 0.4 c 1=0
2 c 1=2
0.2 c 1=4
1.5
c 1=6
1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0
c1 0 5 10 15 20 25
Time

Fig. 1 Optimum values of d1 and d2 for varying c1 values


Fig. 3 Step responses for T13 (s) and J1 criterion

T13(s)
12 d
ISE r + + y
ISTE Gc1 G
+ - -
10 IST2E
+
Gm G
-

Gc2
8
J0, J1, J2

+ +
6

Fig. 4 Control structure for controlling inverse response processes


4

2 It is again well known that, due to its structural limitations


[4], PID controllers may result in unsatisfactory closed loop
0 performances. Therefore, using PI-PD controller, shown to
0 1 2 3 4 5
c1 result in very satisfactory closed loop responses, in the Smith
predictor structure is more rational. As a result of this dis-
Fig. 2 J0 , J1 and J2 integral values for T13 (s) cussion, the control structure given in Fig. 4 is suggested for
tuning processes with inverse response and dead time. If the
controller G c2 (s) = 0, it is the well known Smith predictor
values are shown in Fig. 3 for T13 (s) [5]. It is seen that as c1 scheme [23]. The structure was first proposed [4] for tuning
increases the step responses get faster with very little change processes with large dead times. Here it is used for tuning
in overshoot, a typical characteristic for all the criteria. processes with inverse response and dead time is given.
In Fig. 4, G c1 and G c2 are, respectively, ideal PI and PD
controllers which are given by
3 PI-PD controller tuning for inverse response
processes  
1
G c1 (s) = K c 1 + (4)
Ti s
Because of the term causing inverse response in the plant
G c2 (s) = K f + T f s (5)
transfer function, control of inverse response processes are
more cumbersome. Hence, instead of using the classical sin-
gle input single output control structure, a control scheme In the structure, G is the process transfer function model to
that will help to get rid of the inverse response term is more be controlled which is given by
reasonable. The well known smith predictor configuration
[23] is a good candidate for this. However, in the classi- K (−T s + 1)e−θs
cal Smith predictor scheme a PID type controller is used. G(s) = (6)
s 2 + as + b

123
Electr Eng

The key point in order to obtain the standard closed loop of K c and Ti may involve a trade-off between the values
transfer function and hence for deriving expressions to cal- chosen for α and c1 .
culate PI-PD controller tuning parameters is to factorize the
plant transfer function as G(s) = G m (s)Ḡ(s) where Remark 1 An experienced engineer can use the above given
guidelines to determine the four tuning parameters of the
K
G m (s) = (7) PI-PD controller. However, in order to make the tuning pro-
+ as + b
s2 cedure easier, one can choose K c and Ti values of an already
Ḡ(s) = (−T s + 1)e−θs (8) designed PI or PID controller as initial values and the remain-
ing other two controller parameters, namely T f and K f , can
In this case, the closed loop transfer function of the structure be calculated from (14) and (15). If, satisfactory closed loop
given in Fig. 4 is responses are not achieved, then either K c or Ti value can be
kept at its previous value and the other is changed slightly
G c1 (s)G m (s)Ḡ(s) until a satisfactory solution is obtained.
T (s) = (9)
1 + G m (s)[G c1 (s) + G c2 (s)]
Remark 2 As there are four tuning parameters of the PI-PD
According to (9), two controllers, namely G c1 and G c2 , can
controller, it requires more effort to obtain suitable tuning
be designed using the delay and zero free part of the model.
parameters. However, the standard forms which is used in
Using appropriate expressions in (9), the following closed
this paper to calculate four tuning parameters minimizes this
loop transfer function is obtained.
effort. The guidelines given after (15) and in Remarks 1 are
also useful for designing the PI-PD controller. One can expect
T13 (s)
that, due to its more tuning parameters, the PI-PD controller
K K c (Ti s+1)
= may result in less robust closed responses when compared
Ti s 3 +(a + K T f )Ti s 2 + (b + K K f + K K c )Ti s + K K c to PI/PID controllers. Nevertheless, as shown by examples
(10) provided in Sect. 4 and by papers given in the literature, see
[3–6], PI-PD controllers result in quite satisfactory closed
Using the normalization, loop performances even under assumed parameter changes
 1/3 in the process transfer function and band limited white
Ti s noise.
sn = s = (11)
K Kc α

which means the response of the system will be faster than the
normalized response by a factor of α, results in the standard 4 Robustness analysis and performance
closed loop transfer function
4.1 Robustness analysis
c1 sn + 1
T13 (sn ) = 3 (12) Robustness is an important issue for a control system to
sn + d2 sn2 + d1 sn + 1
result in satisfactory closed loop performances under un-
where, estimated parameter changes in the plant transfer function.
Hence, this section illustrates the robustness of the proposed
c1 = αTi (13) control structure and design method. Kharitonov theorem and
(a + K T f ) related approaches can be used for the robustness analysis of
d2 = (14) control systems with parametric uncertainty.
α
(b + k K f + k K c ) The Kharitonov theorem states that an interval polynomial
d1 = (15) family, which has an infinite number of members, is Hurwitz
α2
stable if and only if a finite small subset of four polynomi-
In principle α can be selected by the choice of K c and c1 by als known as the Kharitonov polynomials of the family are
the choice of Ti . Based on the value of c1 , the coefficients d2 Hurwitz stable. The most significant results following this
and d1 can be found from Fig. 4 and then the value of T f and theorem have been edge theorem of Bartlett et al. [24] and
K f can be computed from (14) and (15) respectively. the generalized Kharitonov theorem of Chapellat and Bhat-
Note that for a selected Ti , choosing larger K c values result tacharyya [25]. The proofs of these results are mainly based
in larger α and c1 values. This implies faster closed loop on the zero exclusion principle and the value set concept.
system response. In practice, K c will be constrained, possibly Extensions of these methods and a discussion of the exten-
to limit the initial value of the control effort, so that the choice sive literature on this subject can be found in [25,26].

123
Electr Eng

Consider an integer order interval polynomial a following interval characteristic polynomial can be obtained

K (s) = p0 + p1 s + p2 s 2 + p3 s 3 + p4 s 4 + p5 s 5 + · · · (s) = 1.5s 3 + [1, 2.457]s 2 + [0.61, 1.59]s


(16) + [0.234, 0.249] = 0 (20)

where pi ∈ [ pimin , pimax ] i = 0, 1, 2, . . . , pimin and pimax The four Kharitonov polynomials are found to be
are specified lower and upper bound of the ith perturbation,
respectively. Kharitonov showed that the stability of the inter- K 1 (s) = 0.234 + 0.61s + 2.457s 2 + 1.50s 3
val polynomial family of (15) could be found by applying the
Routh criterion to the following four polynomials K 2 (s) = 0.234 + 1.59s + 2.457s 2 + 1.50s 3
K 3 (s) = 0.249 + 0.61s + 1.00s 2 + 1.50s 3
K 1 (s) = p0min + p1min s + p2max s 2 + p3max s 3 + p4min s 4 + p5min s 5 + · · · K 4 (s) = 0.249 + 1.59s + 1.00s 2 + 1.50s 3 (21)
K 2 (s) = p0max + p1max s + p2min s 2 + p3min s 3 + p4max s 4 + p5max s 5 + · · ·
K 3 (s) = p0max + p1min s + p2min s 2 + p3max s 3 + p4max s 4 + p5min s 5 + · · · Now consider case 2, where G c1 (s) = 0.50(1 + 1/1.5s)
K 4 (s) = p0min + p1max s + p2max s 2 + p3min s 3 + p4min s 4 + p5max s 5 + · · ·
and G c2 (s) = −0.316 − 0.492s from Table 2. G m (s) is
(17) again given by (7). Once more, note that the controllers cor-
respond to the IST2 E criterion. Following a similar procedure
as for case 1, the following four Kharitonov polynomials are
Lets consider Example 1 of the paper. Two cases will be con-
obtained:
sidered in order to show the trade-offs available in the design
and to compare how the robustness of the closed loop sys-
tem is affected by the choice of PI controller gain K c . From K 1 (s) = 0.45 + 1.30s + 2.936s 2 + 1.50s 3
(9), the closed loop characteristic equation of the proposed K 2 (s) = 0.234 + 2.254s + 2.936s 2 + 1.50s 3
control structure is
K 3 (s) = 0.55 + 1.30s + 1.588s 2 + 1.50s 3

(s) = 1 + G m (s)[G c1 (s) + G c2 (s)] = 0 (18) K 4 (s) = 0.55 + 2.254s + 1.588s 2 + 1.50s 3 (22)

First consider the case 1, where G m (s) is given by (6), and The roots of Kharitonov polynomials for both cases are
G c1 (s) = 0.26(1 + 1/1.5s) and G c2 (s) = −0.528 − 0.847s summarized in Table 1 to show that they all have negative
from Table 2. Note that the controllers correspond to the real parts and hence satisfy Hurwitz condition. Kharitonov
IST2 E criterion. Substituting these into (18), the following rectangles of the closed loop system are given in Figs. 5 and
closed loop characteristic equation is obtained: 6 for case 1 and case 2, respectively. It can be seen that the
Kharitonov rectangles do not include the origin. Therefore,
from zero exclusion principle one can say that the inter-
(s) = 1.5s 3 + (1.5a − 1.27K )s 2 + (1.5b − 0.402K )s
val characteristic equations for both cases are stable. Thus,
+ 0.26K = 0 (19) designed controllers for two cases are robust. From Figs. 5
and 6 it can be seen that the value set for the first case is closer
Nominal values of plant transfer function for Example 1 to origin then the second case. It can be concluded that the
are K = 1, a = 2 and b = 1. It is assumed that K ∈ controller design for second case is more robust than the first
[0.9, 1.1], a ∈ [1.6, 2.4] and b ∈ [0.7, 1.3]. In this case, the case.

Table 1 The roots of


K 1 (s) K 2 (s) K 3 (s) K 4 (s)
Kharitonov polynomials
Case 1
−1.4299 −0.2029 −0.5087 −0.1702
−0.1041 + j0.3135 −0.7175 + j0.5039 −0.079 + j0.5658 −0.2482 + j0.9560
−0.1041 − j0.3135 −0.7175 − j0.5039 −0.079 − j0.5658 −0.2482 + j0.9560
Case 2
−1.5162 −0.2973 −0.6171 −0.2861
−0.2205 + j3863 −0.8300 + j5659 −0.2208 + j7385 −0.3863 + j1.0642
−0.2205 + j3863 −0.8300 − j5659 −0.2208 − j7385 −0.3863 − j1.0642

123
Electr Eng

0.8
design. The performance of controller structure and design
0.6 method is compared with design methods of Luyben [18],
Chien et al. [19] and Jeng and Lin [22] as they have also pro-
0.4
posed PI/PID controller design methods for tuning processes
0.2 with inverse response characteristics. Simulation examples
0
are provided to compare the performance of the proposed
control structure and design method with the design meth-
Im

-0.2 ods cited above.


-0.4
Example 1 Here, the example studied by Luyben [18] is con-
-0.6 sidered. The transfer function model used by Luyben [18] is
-0.8
(−T s + 1)e−θs
G(s) =
-1 (s + 1)2
-2.5 -2 -1.5 -1 -0.5 0 0.5
Re
Two values of the dead time (θ = 0.2 and 1.6) and also two
Fig. 5 Kharitonov rectangles for G m (s) = 1/(s + 1)2 , G c1 (s) = values of T parameter (T = 0.2 and 1.6), and hence four
0.26(1 + 1/1.5s) and G c2 (s) = −0.528 − 0.847s sub-examples, were considered by Luyben [18]. However,
1.2
here, only one value of T parameter, namely T = 1.6, and
one dead time value, namely θ = 1.6, are considered. This
1 case corresponds to the most sluggish and unsatisfactory
closed loop responses, when the design methods of Luyben
0.8 [18] and Chien et al. [19] are considered. Controller para-
meters used in the simulations by Luyben [18] and Chien
0.6 et al. [19] are, respectively, K c = 0.26, Ti = 1.50 and
K c = 0.15, Ti = 1.00, Td = 1.00.
Im

0.4
Table 2 summarizes tuning parameters of controllers
0.2 obtained for the proposed control structure for two cases.
In the first case, K c and Ti are limited to tuning parameters
0 of the PI controller suggested by Luyben [18], in order to
make a fair comparison and show the use of remark 1. In the
-0.2 second case, K c is limited to 0.5 and Ti is kept at its previous
-2.5 -2 -1.5 -1 -0.5 0 0.5 1
Re value to show the trade-offs available in the design. Figure 7
Fig. 6 Kharitonov rectangles for G m (s) = 1/(s + 1)2 , G c1 (s) =
gives the closed loop system responses to a unit step set-point
0.50(1 + 1/1.5s) and G c2 (s) = −0.316 − 0.492s change for both cases using the different design criteria. It is
seen that faster responses for case 2 are achieved, which fits
to context given in Sect. 2. Also, it is observed that the ISE
4.2 Performance criterion gives the most oscillatory responses for both cases.
Another observation is that the ISTE and IST2 E criteria give
Integral of squared error ∞ (ISE), see (1), and Integral of quite similar responses.
absolute error IAE = 0 |e(t)|dt are two widely used cri-
teria for evaluating closed loop performances. Hence, they
Table 2 Proposed controller parameters for Example 1
are used to compare the closed loop performances of the
proposed control structure and design method with the ones Criterion c1 d2 d1 Kc Ti Kf Tf
existing in the literature and used for comparison. Case 1
The total variation (TV) of the input u(t) is a good measure ISE 0.84 1.170 2.224 0.26 1.50 −0.572 −1.353
of the smoothness
∞ of a signal [27]. The TV is computed by
ISTE 0.84 1.730 2.250 0.26 1.50 −0.564 −1.044
TV = i=1 |u i+1 − u i | and should be as small as possible.
IST2 E 0.84 2.087 2.365 0.26 1.50 −0.528 −0.847
Case 2
5 Simulation examples ISE 1.04 1.234 2.332 0.50 1.50 −0.379 −1.145
ISTE 1.04 1.816 2.351 0.50 1.50 −0.370 −0.741
This section provides several examples to illustrate the use IST2 E 1.04 2.175 2.462 0.50 1.50 −0.316 −0.492
of the proposed controller structure and PI-PD controller

123
Electr Eng

1.4 1.6

Case 1
1.2
Case 2
1.4
1

0.8 1.2

Control Signal Magnitude


0.6
1
Output, y

0.4

0.8
0.2

0
0.6

-0.2 2
Proposed IST E Kc=0.5
2 0.4 2
-0.4 IST E Proposed IST E Kc=0.26
ISTE Chien
ISE Luyben
-0.6 0.2
0 5 10 15 20 25 0 5 10 15 20 25 30
Time Time

Fig. 7 Responses for the proposed method to a unit step set point Fig. 9 Comparison of control signals for Example 1
change for Example 1

1.4
Also, the proposed method for K c = 0.5 results in lesser ISE
1.2 and IAE values but slightly larger TV value when compared
with the proposed method for K c = 0.26 for both nominal
1
system and perturbed system.
0.8
Example 2 An inverse response process with large time con-
0.6
Output, y

stant is considered in this example:


0.4

5(−s + 1)e−2s
0.2 G(s) =
(100s + 1)(10s + 1)
2
0 Proposed IST E Kc=0.5
2
Proposed IST E Kc=0.26 Since the time constants are different, the method suggested
-0.2
Chien by Luyben [18] cannot be used in this case. The above given
transfer function can be re-arranged as G(s) = 0.005(−s +
Luyben
-0.4
0 10 20 30 40 50 60
Time
1)e−2s /(s 2 + 0.11s + 0.001) to put in the form of (7). In
Fig. 8 Comparison of responses to a unit step set point change and order to find tuning parameters of the PI-PD controller, PI
disturbance for Example 1 controller gain K c is limited to 20 and the integral time con-
stant Ti is selected as 2. This results in α = 0.368 and hence
c1 = 0.737. The standard form T13 (s) to minimize J2 for
Figure 8 illustrates responses for the proposed controller c1 = 0.737 has d2 = 2.318 and d1 = 2.040, which gives
structure and design method to both a unit set-point change T f = 131.255 and K f = 43.347. The method of Chien et
and a disturbance with magnitude of −0.5. To avoid crowd- al. [19] has PID parameters of K c = 1.742, Ti = 100, Td =
ing in Fig. 8, results for the proposed method are only given 10.J2 criterion values of the proposed design method and the
for the IST2 E criterion. For comparison, results for the design design method of Chien et al. [19] are, respectively, given
methods of Luyben [18] and Chien et al. [19] are also pro- by 428.323 and 40206.513. Responses to a unit step change
vided. Clearly, the proposed method gives better closed loop are given in Fig. 10. Responses to disturbance with different
responses for both the set point tracking and disturbance magnitudes are illustrated in Fig.11. It is seen that disturbance
rejection. Control signals for all design methods are given in rejection of both methods are satisfactory for disturbances
Fig. 9. Note the initial high control effort of design method with large magnitudes. The proposed approach gives supe-
suggested by Chien et al. [19]. rior performance than design method of Chien et al. [19] for
Performance specifications for all three design methods all cases. Comparing to the first example the superiority of
are given in Table 3 for nominal system and perturbed sys- the proposed method is now more obvious.
tem. For the perturbed system, it was assumed that there are
+30 % change in K , T, a, b and θ . It seen that the proposed Smith predictor structures are quite sensitive to modeling
method gives less ISE, IAE and TV values when compared errors. Hence, responses under ±30 % changes in K , T and
with design methods of Luyben [18] and Chien et al. [19]. θ are given in Fig. 12, as the proposed controller scheme

123
Electr Eng

Table 3 Performance specifications


Scheme Nominal system (set-point + disturbance) Perturbed system (set-point + disturbance)

IAE ISE TV IAE ISE TV

Example 1
Proposed K c = 0.5 3.5234 1.7623 0.0030 5.4258 2.1612 0.0051
Proposed K c = 0.26 4.5847 2.3807 0.0028 6.5992 2.8308 0.0040
Chien et al. 10.7844 7.2561 0.0054 13.5084 8.9977 0.0066
Luyben 12.0308 7.6780 0.0033 16.5291 10.1118 0.0046
Example 2
Proposed 15.2196 3.6274 0.0665 16.9603 3.7268 0.0614
Chien et al. 39.8006 11.8487 0.0450 44.6973 13.4542 0.0438
Example 3
Proposed 11.1272 5.5328 0.0037 286.6185 1607.300 0.016 × 104
Jeng and Lin 11.9789 8.1269 0.0056 287.1330 1613.700 48.092 × 104

1.2 1.4
+30% Change

1 1.2

1
0.8

0.8
0.6
Output, y

Output, y

0.6

0.4 -30% Change


0.4

0.2
0.2

0
0
Proposed Proposed
Chien Chien
-0.2 -0.2
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time Time

Fig. 10 Step input responses for Example 2 Fig. 12 Step input responses under assumed parameter changes in
K , T and θ
0.05

-0.05 loop responses are achieved although a smith predictor based


-0.1 controller configuration is used and large changes in the para-
-0.15
meters are assumed.
From Fig. 12, it is seen that the proposed method results in
Output

-0.2
larger overshoots in the case of an assumed +30 % changes
-0.25 in K , T and θ when compared with the method of Chien
-0.3 et al. [19]. This situation can be expected, because, for the
-0.35
Proposed, d=-0.2
Proposed, d=-0.6
nominal case the method of Chien et al. [19] is very slug-
Proposed, d=-1.0 gish, while the proposed method gives quite faster response.
-0.4 Chien, d=-0.2
Chien, d=-0.6 Hence, although there are similar deterioration in responses
Chien, d=-1.0
-0.45
0 50 100 150 200 250 300 350 400 450 500 for the proposed design method and the design method of
Time
Chien et al. [19], it looks as if the proposed design method
Fig. 11 Disturbance responses for Example 2 is more sensitive to parameter changes. On the other hand,
when the case of assumed −30 % changes in K , T and θ is
is based on smith predictor structure. In the simulations, considered, the proposed method gives a faster closed loop
the controller tuning parameters are kept at their original response with no overshoot when compared with the method
values. As is seen, from Fig. 12 quite satisfactory closed of Chien et al. [19]. As a result, for large positive changes

123
Electr Eng

1.2 1.6

1.4
1
1.2

0.8
1

0.8
0.6
Output, y

Output, y
0.6
0.4
0.4

0.2 0.2

0
0

Proposed -0.2
Proposed
Chien
-0.2 Jeng
0 10 20 30 40 50 60 70 -0.4
0 10 20 30 40 50 60 70 80 90 100
Time
Time

Fig. 13 Responses under measurement noise having maximum power Fig. 15 Step and disturbance responses for Example 3
of 0.02 for Example 2

25
Proposed (−2s + 1)
Chien G(s) =
(s + 1)5
20

Let us use the model G(s) = 0.822(−1.117s + 1)e−1.362s /


Control Signal Magnitude

15
(4.434s 2 + 3.79s + 1) which was used by Jeng and Lin
[22]. The model is rearranged as G(s) = 0.185(−1.117s +
1)e−1.362s /(s 2 + 0.855s + 0.226) to calculate the four PI-
10

PD controller parameters. PID controller parameters for the


5
method of Jeng and Lin [22] is given by K c = 0.795, Ti =
4.038, Td = 1.287. For the proposed PI-PD controller, K c
0 is limited to 0.4 which is approximately half of the con-
troller gain value of the method suggested by Jeng and Lin
-5
0 5 10 15 20 25 30
[22]. The integral time constant Ti is selected as 8.0. These
Time values result in α = 0.210 and hence c1 = 1.680. The
standard form T13 (s) to minimize J2 for c1 = 1.680 has
Fig. 14 Comparison of control signals for Example 2
d2 = 2.462 and d1 = 2.856, which gives T f = −1.828
and K f = −0.942.J2 criterion values are 7751.6 and 11087
in all plant transfer function parameters, one can prefer the for the proposed design method and for the design method
method of Chien et al. [19]. But, for negative changes in all of Jeng and Lin [22], respectively. Responses to a unit step
plant transfer function parameters, the proposed method is change and a disturbance with magnitude of −0.5 entering
yet more preferable as in nominal case. Since, in practice the closed loop system at time 50 s are shown in Fig. 15.
measurements are not noise free, the closed loop system for
both design methods are re-simulated assuming a unit step In the sense of overshoot, the proposed method gives a
set point change and a noise with maximum power of 0.02. better response than the method of Jeng and Lin [22]. Both
Results are given in Fig. 13. It is seen that quite reasonable methods result in a similar settling time. The method of Jeng
closed loop performances are achieved. Controller signals and Lin [22] gives slightly faster disturbance rejection. As
are illustrated in Fig. 14. It is seen that the proposed PI-PD stated for Example 2, in practice measurements are not noise
requires slightly larger control effort. free. Hence, the closed loop system responses, assuming a
The values of ISE, IAE and TV are provided in Table 3. unit step set point change and disturbance with magnitude of
For the perturbed system −30 % change in K , T and θ were −0.5 under a noise with maximum power of 0.02 are given in
assumed. It is observed that the proposed control scheme and Fig. 16. It can be concluded that both methods show a similar
design method gives lower performance specifications. sensitivity to noises. Figure 17 gives controllers signals for
the proposed method and the method suggested by Jeng and
Example 3 Consider a high-order process with inverse Lin [22]. The figure shows that the proposed method requires
response studied by Jeng and Lin [22]: lees initial control effort.

123
Electr Eng

2
closed loop responses, where PID controllers are unsatisfac-
tory, was used. Tuning parameters of the PI-PD controller
1.5
were determined from standard forms. Simulation examples
results have been provided to illustrate the value of the pro-
posed structure. It is worth to mention that an alternative
1 way to improve the closed loop performance is to use frac-
Output, y

tional order PID or PI-PD controllers. Fractional order PID


type controllers are nowadays attracting interest of many
0.5
researchers. However, the difficulty with the designing of
fractional order PID type controllers is the trouble in deter-
0 mining its tuning parameters. So, it can be concluded that the
use of fractional order PID controllers in controlling inverse
Proposed response processes is an open issue and may be a possible
Jeng
-0.5
0 10 20 30 40 50 60 70 80 90 100 investigation area.
Time

Fig. 16 Responses under measurement noise having maximum power


of 0.02 for Example 3 References

2 1. Kaya I, Tan N, Atherton DP (2006) A refinement procedure for


PID controllers. Electr Eng 88:215–221
1.8 2. Atherton DP, Boz AF (1998) Using standard forms for controller
design. In: Proceedings of UKACC Control’98, pp. 1066-1071
1.6 3. Kaya I (2003) A PI-PD controller design for control of unstable
and integrating processes. ISA Trans 42:111–121
Control Signal Magnitude

1.4 4. Kaya I (2003) A new smith predictor and controller for control of
processes with long dead time. ISA Trans 42:101–110
1.2
5. Kaya I, Tan N, Atherton DP (2007) Improved cascade control struc-
ture for enhanced performance. J Process Control 17:3–16
1
6. Kaya I, Atherton DP (2008) Use of Smith predictor in the outer
loop for cascaded control of unstable and integrating processes.
Ind Eng Chem Res 47:1981–1987
0.8
7. Shamsuzzoha M, Skogestad S (2010) The setpoint overshoot
method: a simple and fast closed-loop approach for PID tuning.
0.6
J Process Control 20:1220–1234
Proposed
Jeng
8. Matausek MR, Sekara TB (2011) PID controller frequency-domain
0.4
0 5 10 15 20 25 30 35 40 45 50 tuning for stable, integrating and unstable processes, including
Time dead-time. J Process Control 2:17–27
9. Shamsuzzoha M (2013) Closed-loop PI/PID controller tuning for
Fig. 17 Comparison of control signals for Example 3 stable and integrating process with time delay. Ind Eng Chem Res
52:12973–12992
10. Li K (2013) PID tuning for optimal closed-loop performance with
Performance specifications are supplied in Table 3. For specified gain and phase margins. IEEE Trans Control Syst Technol
the perturbed system, +30 % change in K and T and simul- 21:1024–1030
11. Lee J, Cho W, Edgar TF (2014) Simple analytic PID controller
taneously a band limited noise with maximum power of 0.02
tuning rules revisited. Ind Eng Chem Res 53:5038–5047
were supposed. 12. Pavkovic D, Polak S, Zorc D (2014) PID controller auto-tuning
based on process step response and damping optimum criterion.
ISA Trans 53:85–96
13. Cho W, Lee J, Edgar TF (2014) Simple analytic proportional-
6 Conclusions integral-derivative (PID) controller tuning rules for unstable
processes. Ind Eng Chem Res 53:5048–5054
Processes with inverse response characteristics can be obser- 14. Ali A, Majhi S (2011) Integral criteria for optimal tuning of PI/PID
ved in process control systems. There are only a few studies controllers for integrating processes. Asaian J Control 13:328–337
15. Ogunnaike BA, Ray WH (1994) Process dynamics, modelling and
in the literature considering the control of such processes. control. Oxford University Press, Oxford
Those studies have some shortcomings in controlling inverse 16. Stephanopoulos G (1984) Chemical process control: an introduc-
response characteristic processes. Hence, a controller struc- tion to theory and practice. Prentice-Hall, Englewood Cliffs
ture which was originally suggested by the author for 17. Waller KVT, Nygardas CG (1975) On inverse response in process
control. Ind Eng Chem Fundam 14:221–223
controlling processes with large dead times has been adopted 18. Luyben WL (2000) Tuning Proportional-Integral controllers for
to control processes with inverse response. In the proposed processes with both inverse response and dead time. Ind Eng Chem
structure, the PI-PD controller which was shown to have good Res 39:973–976

123
Electr Eng

19. Chien IL, Chung YC, Chen BS, Chuang CY (2003) Simple PID 23. Smith OJ (1959) A controller to overcome dead time. ISA J 6:28–
controller tuning method for processes with inverse response plus 33
dead time or large overshoot response plus dead time. Ind Eng 24. Barlett AC, Hollot CV, Lin H (1988) Root locations of an entire
Chem Res 42:4461–4477 polytope of polynomials: it suffices to check the edge. Math Control
20. Camacho O, Rojas R, Garcia W (1999) Variable structure control Signal Syst 1:61–71
applied to chemical processes with inverse response. ISA Trans 25. Bhattacharyya SP, Chapellat H, Keel LH (1995) Robust control:
38:55–72 the parametric approach. Prentice Hall
21. Balaguer P, Alfaro V, Arrieta O (2011) Second order inverse 26. Tan N (2002) Computation of the frequency response of multilinear
response process identification from transient step response. ISA affine systems. IEEE Trans Autom Control 47:1691–1696
Trans 50:231–238 27. Skogestad S (2003) Simple analytic rules for model reduction and
22. Jeng JC, Lin SW (2012) Robust proportional-integral-derivative PID controller tuning. J Process Control 13:291–309
controller design for stable/integrating processes with inverse
response and time delay. Ind Eng Chem Res 51:2652–2665

123

You might also like