Anda di halaman 1dari 8

www.ietdl.

org
Published in IET Control Theory and Applications
Received on 18th February 2010
Revised on 11th June 2010
doi: 10.1049/iet-cta.2010.0082

ISSN 1751-8644

Simultaneous closed-loop automatic tuning


method for cascade controllers
M. Veronesi A. Visioli
Dipartimento di Ingegneria dellInformazione, University of Brescia Italy, Via Branze 38, I-25123 Brescia, Italy
E-mail: antonio.visioli@ing.unibs.it

Abstract: A new automatic tuning method for cascade control systems is presented. The technique consists in estimating the
parameters of the primary and secondary process simultaneously by evaluating a closed-loop set-point step response. Then,
based on the estimated model, the two proportional-integral-derivative controllers are tuned by applying an internal model
control strategy. A performance assessment strategy can also be implemented in this context. Both self-regulating and nonself-regulating processes are considered. Simulation results show the effectiveness of the methodology.

Introduction

The achievement of a satisfactory performance in addition to


a fast (low-cost) setting-up are essential requirements for a
control system to be adopted in the industrial context.
Proportional-integral-derivative (PID) controllers are widely
adopted in industry because an acceptable performance can
be obtained despite their relative simplicity, which is given
by their simple structure and by the presence of many
(auto-)tuning methods that have been devised in the last 60
years [1, 2].
When more demanding control specications are required
for a given application, PID controllers can be still adopted
as a basis of more complex control schemes where
couplings between simple control systems are exploited.
A typical example in this context is cascade control, which is
available as a standard tool in almost all industrial process
controller. It consists in employing an additional sensor in
order to separate the fast and the slow dynamics of the process
to obtain a fast load disturbance rejection [3]. Despite in the
last decade more sophisticated cascade control scheme have
been devised [46], the standard control scheme consists of
two nested loops with two PID controllers. The controller of
the inner loop is the so-called secondary controller, whereas
that of the outer loop is the primary controller. Thus, the
design of the overall control system involves the tuning of two
(PID) controllers. Usually, the secondary controller is tuned
rst by setting the primary controller in manual mode. Then,
the primary controller is tuned while the previously tuned
secondary controller is applied to the inner loop. It appears
how the design procedure is longer and more complicated
than that of a standard single-loop (PID) control.
In order to provide users with an effective aid in the design of
the cascade control system, a relay feedback autotuning
technique has been proposed in [79]. It basically consists in
applying the typical relay feedback-based autotuning [10] to
the two loops, sequentially as in the conventional approach.
IET Control Theory Appl., 2011, Vol. 5, Iss. 2, pp. 263270
doi: 10.1049/iet-cta.2010.0082

Thus, the procedure is still time consuming. To avoid the


sequential nature of the tuning procedure a simultaneous
online (closed-loop) automatic tuning technique has been
proposed in [11, 12]. However, in both cases an ad hoc
experiment (involving again the presence of a relay) has to be
performed. Another simultaneous tuning procedure has been
proposed in [13]. It is based on the evaluation of an open-loop
process step response and therefore also in this case an ad hoc
experiment has to be performed.
Further, tuning rules have also been proposed in the
literature (see e.g. [14 16]) for the primary and secondary
controller, but in these cases it has not been specied how
the procedure can be automated. A comprehensive review
of design methodologies for cascade control can be found
in [17].
In this paper we propose a new automatic tuning technique
in which the parameters of the primary and secondary process
are estimated simultaneously by evaluating a set-point step
response of the cascade control system where the PID
controllers have been previously roughly tuned. Thus, the
procedure can exploit routine operating data. Then, once the
process parameters have been estimated, the PID controllers
can be tuned according to any method devoted for this
purpose. However, here we propose to use an internal
model control (IMC) strategy [18], which has been shown
to give a satisfactory performance and, at the same time,
allows one to implement also a performance assessment
technique, as it will be shown in the paper. Note that the
overall methodology can be applied to both self-regulating
and non-self-regulating (integral) processes [19].
The paper is organised as follows. A cascade control scheme
is described in Section 2, mainly with the aim of introducing
the notation. The parameters estimation technique is
presented in Section 3. The PID tuning algorithm is
described in Section 4. The performance assessment method
is discussed in Section 5. Simulation results are given in
Section 6 and conclusions are given in Section 7.
263

& The Institution of Engineering and Technology 2011

www.ietdl.org
2

Cascade control

In a typical cascade control scheme (see Fig. 1) the process has


one input and two outputs. Indeed, in order to provide an
effective rejection of the load disturbance d, an additional
sensor is employed to separate as much as possible the fast
dynamics of the process from the slow one (i.e. the one with
the slowest poles and the non-minimum-phase part) [3]. The
process transfer function is denoted by P(s) = P2 (s)P1 (s)
where P2 (s) is the transfer function that models the fast
dynamics of the process and P1 (s) is the one of the slow part,
C2 (s) is the secondary controller and C1 (s) is the primary
controller. As it is typical in the industrial context, the
secondary process P2 (s) is assumed to have a rst-orderplus-dead-time (FOPDT) dynamics, whereas the primary
process P1 (s) is assumed to have an FOPDT or a secondorder-integrator-plus-dead-time
(SOIPDT)
dynamics,
according to the presence of a self-regulating or non-selfregulating process, respectively (see Remark 2 for a
discussion about the robustness issue). Thus, the two transfer
functions P2 (s) and P1 (s) can be written as

m eu2 s
P2 (s) = 2
t2 s + 1

Finally, r represents the set-point signal, u1 and u2 are the


output of the primary and secondary controller, respectively,
and e1 and e2 are the control errors of the primary and
secondary loop, respectively.
It appears that the effectiveness of a cascade control
scheme is due to the fact that disturbances affecting the
(fast) secondary loop are effectively compensated before
they affect the main process output y1 .
For the purpose of the analysis that will be done in the
following sections, it is worth stressing the following
relations which can be trivially deduced from the scheme of
Fig. 1
U1 (s) =

G2 (s) =

and

3
P1 (s) =

u1 s

m1 e
s(t1 s + 1)

(3)

It is then assumed that the controllers are of PID type in series


conguration (the use of other forms is straightforward by
suitably applying translation formulae to determine the
values of the parameters [17]), that is

Cj (s) = Kpj

Tij s + 1
Tij s




Tdj s + 1
,
Tf j s + 1

(4)

where (j = 1, 2) Kpj is the proportional gain, Tij is the integral


time constant, Tdj is the derivative time constant and Tf j is the
time constant of the lter that makes the controller proper. The
controller transfer functions can be rewritten as
Cj (s) =

Kpj C j (s)
,
sTij

j = 1, 2

(5)

C j (s) =

(1 + sTij )(1 + sTdj )


,
(1 + sTf j )

j = 1, 2

nG2 (s)

(9)

dG2 (s)

Parameters estimation

Self-regulating processes

Assume that a set-point step signal of amplitude As is applied


to the cascade control system with process (1) (2) at time
t = 0. Without loss of generality, we assume null initial
conditions. The process parameters can be estimated by
evaluating the transient response. Indeed, the gain m1 of the
primary process can be obtained by integrating the variable
e1 (t) = r(t) y1 (t). In fact
E1 (s)
1
=
R(s) 1 + C1 (s)G2 (s)P1 (s)
=

where, obviously

m2 Kp2 C 2 (s)esu2
sTi2 (t2 s + 1) + m2 Kp2 C 2 (s)esu2

The rst part of the automatic tuning methodology consists in


estimating the process parameters. As stated in the
Introduction, this is achieved by evaluating the set-point
step response of the cascade control system where the PID
controllers have been previously roughly tuned (thus, by
exploiting routine operating data).
3.1

j = 1, 2

(8)

C2 (s)P2 (s)
1 + C2 (s)P2 (s)

=:

(2)

or

C2 (s)
U (s)
1 + C2 (s)P2 (s) 1

where

(1)

m1 eu1 s
t1 s + 1

(7)

U2 (s) =

=
P1 (s) =

C1 (s)
R(s)
1 + C1 (s)G2 (s)P1 (s)

sTi1 (t1 s + 1)dG2 (s)


sTi1 (t1 s + 1)dG2 (s) + m1 m2 Kp1 Kp2 C 1 (s)C 2 (s)e(u1 +u2 )s
(10)

(6)
Thus, by integrating e1 (t) and by applying the nal value
theorem, we have
1
lim

t+1 0

Fig. 1 Considered cascade control scheme


264
& The Institution of Engineering and Technology 2011

e1 (t) dt = lim s
s0

As E1 (s)
T A
= i1 s
s s
m1 Kp1

(11)

Hence, the gain of the primary process P1 (s) can be


IET Control Theory Appl., 2011, Vol. 5, Iss. 2, pp. 263 270
doi: 10.1049/iet-cta.2010.0082

www.ietdl.org
it is possible to write

computed as

1
T A
1i1 s
m1 =
Kp1 0 e1 (t) dt

(12)

lim

t+1

v(t) dt
0

= lim s
Similarly, the gain u2 of the secondary process P2 (s) can be
obtained by integrating the variable e2 (t) = u1 (t) y2 (t). In
fact
U1 (s)
E2 (s) =
1 + C2 (s)P2 (s)
sTi2 (t2 s + 1)
=
U1 (s)
sTi2 (t2 s + 1) + m2 Kp2 C 2 (s)eu2 s

(13)

Then, by taking into account that


U1 (s)
C1 (s)
=
R(s)
1 + C1 (s)P1 (s)
Kp1 C 1 (s)(t1 s + 1)
=
sTi1 (t1 s + 1) + m1 Kp1 C 1 (s)eu1 s

(14)

and by integrating e2 (t) and applying the nal value theorem


we have that
1
lim

t+1 0

e2 (t) dt = lim s
s0

As /m1 E2 (s)
Ti2 As
=
s
m1 m2 Kp2
s

(15)

s0

As (t1 s + 1) G2 (s)eu1 s
s
s

= As lim

(t1 s + 1)dG2 (s) m2 Kp2 C 2 (s)e(u1 +u2 )s


sdG2 (s)

s0

sTi2 (t1 s + 1)(t2 s + 1) + ((t1 s + 1) esu1 )


m2 Kp2 C 2 (s)esu2
= As lim
s0
s(sTi2 (t2 s + 1) + m2 Kp2 C 2 (s)eu2 s )


Ti2
(t1 s + 1) eu1 s
+ lim
= As
m2 Kp2 s0
s



Ti2
(t1 s + 1) 1 1 eu1 s
+
+ lim
= As
m2 Kp2 s0
s
s


Ti2
= As
+ t1 + u 1
(19)
m2 Kp2
Dene as T0j , j = 1, 2 the sum of the dead time and of the
time constant of the process Pj (s)
T0j := tj + uj
From (19) we have
T01 =

Thus, the gain of the secondary process can be computed as


Ti2 As
1
m2 =
m1 Kp2 0 e2 (t) dt

(16)

1
v(t) dt
0

Ti2
m2 Kp2

(21)

For the secondary loop, the following variable can be


introduced
(22)

so that we have
(17)

V (s) C1 (s)(m1 G2 (s)P1 (s))


=
R(s)
1 + C1 (s)G2 (s)P1 (s)


m2 C2 (s)
C2 (s)P2 (s)

U (s)
1 + C2 (s)P2 (s) 1 + C2 (s)P2 (s) 1

W (s) =

so that we have

C2 (s)
C1 (s)
(m2 P2 (s))
R(s)
1 + C2 (s)P2 (s)
1 + C1 (s)P1 (s)
(23)

Then, by integrating w(t) and by applying the nal value


theorem, we obtain

Kp1 C 1 (s) sTi1 (t1 s + 1)dG2 (s)(m1 G2 (s)P1 (s))


=
sTi1 sTi1 (t1 s + 1)dG (s) + m1 m2 Kp1 Kp2 C 1 (s)

1

C 2 (s)e(u1 +u2 )s

lim

m1 Kp1 C 1 (s)dG2 (s)((t1 s + 1) G2 (s)eu1 s )


=
sTi1 (t1 s + 1)dG2 (s) + m1 m2 Kp1 Kp2 C 1 (s)C 2 (s)e(u1 +u2 )s
(18)
By integrating v(t) and by applying the nal value theorem,
IET Control Theory Appl., 2011, Vol. 5, Iss. 2, pp. 263270
doi: 10.1049/iet-cta.2010.0082

1
As

w(t) = m2 u2 (t) y2 (t)

Then, the following variable is introduced


v(t) = m1 u1 (t) y1 (t)

(20)

t+1 0

w(t) dt = lim s
s0

As 1
(t s + 1) eu2 s 1
m2 2
s m2
s(t2 s + 1) m1

As
(t s + 1) eu2 s As
= (u2 + t2 )
lim 2
m1 s0 s(t2 s + 1)
m1
(24)

Thus, the sum of the secondary process time constant and


265

& The Institution of Engineering and Technology 2011

www.ietdl.org
dead time can be computed as
T02

m
= 1
As

1
w(t) dt

(25)

Then, the apparent dead times u2 and u1 of the two processes


can be evaluated by considering, respectively, the time
interval from the application of the step signal to the setpoint and the time instant when the secondary and primary
output attains the 2% of the steady-state value
corresponding to the new set-point value As , namely, when
the conditions y2 . 0.02As /m1 and y1 . As occur. Actually,
from a practical point of view, in order to cope with the
measurement noise, a simple sensible solution is to dene a
noise band NB [20] (whose amplitude should be equal to
the amplitude of the measurement noise) and to rewrite the
condition as yj . NB, j = 1, 2.
It is clear at this point that the time constants of the two
processes can be trivially obtained as

tj = T0j uj ,
3.2

j = 1, 2

(26)

lags and of the apparent dead time. Thus, we can consider the
model reduction technique known as the half rule, which
states that the largest neglected (denominator) time constant
is distributed evenly to the effective dead time and the
smallest retained time constant [22]. This means that the
initial (high-order) model and the reduced FOPDT model
have the same value of the sum of the lags and of the
apparent dead time. In other words, in case the process has
a high-order dynamics, the proposed method estimates the
parameters of the reduced FOPDT model [21].

PID tuning

Once the process parameters have been estimated, the PID


controllers can be tuned simultaneously according to a
given tuning rule. Here we propose to use an IMC approach
[18] because, in addition to provide good results both in the
load disturbance rejection and set-point following task (see
Section 6), it allows to devise simply a performance
assessment strategy (see Section 5) and it is capable to cope
with the possible model uncertainties (see Remark 2).

Non-self-regulating processes

The same rationale of the technique described for self-regulating


processes can be applied also to non-self-regulating processes
(1) and (3). In this case, v(t) has to be redened as
1
u1 (t) dt y1 (t)

v(t) := m1

4.1

Self-regulating processes

According to the IMC strategy, the secondary controller


parameters are selected as

(27)

Then, it is therefore possible to determine

m1 = As
T01

Ti1

Kp1
1

1
=
As

m2 =

1 t

e (n) dn dt
0 1

Kp2 =

u2 /2
,
m2 (l2 + u2 )

Tf 2 =

l2 u2
2(l2 + u2 )

Ti2 =

u2
,
2

Td2 = t2 ,
(32)

(28)
where

T
v(t) dt i2
m2 Kp2

(29)

l2 = max{0.25u2 , 0.2t2 }

T A
1 i2t s
m1 Kp2 0 0 e2 (n) dn dt

(30)

is the desired closed-loop time constant. In fact, with these


parameters the inner loop transfer function results (by
employing a rst-order Pade approximation)

(33)

and

m
T02 = 1
As

1 t
w(n) dn dt
0

G2 (s) =

(31)

1
eu2 s
l2 s + 1

(34)

Finally, by estimating the dead times u2 and u1 as in the previous


subsection, expression (26) can be employed to determine the
time constants of the process.
Remark 1: The values of process parameters m1 , m2 , T01 and
T02 are determined by considering the integral of signals and
therefore the method is inherently robust to the measurement
noise [21]. Further, they are obtained independently on the
values of the PID parameters. This is an advantage with
respect to the use of other methods for the identication of
the process transfer function (e.g. the least squares
approach), whose result depends on the control variable and
process variable signals.
Remark 2: The proposed method addresses also implicitly the
robustness issue. Actually, if the processes have multiple lags,
the value of T01 and T02 represents in any case the sum of the
266
& The Institution of Engineering and Technology 2011

Denote as G(s) the process seen by the primary controller,


given by the series of G2 (s) and P1 (s). Its transfer function
can be written as
G(s) := G2 (s)P1 (s) =

m1
e(u2 +u1 )s
(l2 s + 1)(t1 s + 1)

(35)

By applying the IMC-based tuning rule proposed in [22], which


provides a very good compromise between performance and
robustness, the primary controller can be tuned as follows
Kp1 =

T01 u1
,
m1 (l1 + u)

Tf 1 =

Td1
10

Ti1 = T01 u1 ,

Td1 = l2 ,
(36)

IET Control Theory Appl., 2011, Vol. 5, Iss. 2, pp. 263 270
doi: 10.1049/iet-cta.2010.0082

www.ietdl.org
where

u = u1 + u2 ,
4.2

l1 = u

(37)

Non-self-regulating processes

In case of a non-self-regulating process, the same tuning rule


(32) can be employed for the secondary controller, whereas
for the primary controller it is convenient to use a PD
controller whose parameters are set, accordingly to the IMC
methodology, as
Kp1 =

1
,
2m1 l1

Td1 = T01 u1 + l2

(38)

where

l1 = u1 + u2

(39)

Note that, if the value of the secondary time constant t1 is


negligible, then a negative value of Td1 can be determined
from (38). In this case it is sufcient to set Td1 = 0, namely,
to use a simple proportional controller (see Section 6.4).

J = 0.6. In any case, another value of J can be selected by


the user depending on how tight are its control
specications. In case the performance is not satisfactory,
the controllers can be retuned according to the technique
described in Section 4. It is worth stressing that the
obtained value of the performance index J can be also
greater than one because the target values of the integrated
absolute error have been chosen greater than the theoretical
minimum integrated absolute error that can be achieved.
Remark 3: It is worth noting that, especially in case the
controller employed yields a signicant overshoot, the setpoint following performance can be improved by using a
two degree-of-freedom control [12, 15], namely by ltering
the set-point signal.
Remark 4: As it is well known in the IMC context, parameters
l1 and l2 allows to handle the trade-off between
aggressiveness and robustness (and control activity) and
therefore they can be easily modied by the user in order to
meet a particular control specications. Actually, the
proposed performance assessment strategy is meaningful
from this point of view [see (41)].

Performance assessment

6.1
The devised tuning strategy, being analytical, can be
exploited effectively also in a performance assessment
context [23]. Indeed, the assessment of the performance of
a control loop is generally performed by rst calculating a
performance index based on the available data and then by
evaluating the current control performance against a
selected benchmark, which represents the desired
performance. Usually, minimising the integrated absolute
error
1
IAE =

1
|e(t)| dt =

|r(t) y(t)| dt

(40)

e1 (t) dt =
0

As Ti1
= As (l1 + u) = 2As u
m1 Kp1

Example 1

As a rst example we consider the following self-regulating


process [12]
P1 (s) =

1
,
(1 + 10s)(1 + 4s)(1 + s)2

P2 (s) =

1
2s + 1

(43)

With the purpose of showing the effectiveness of the method,


as an initial tuning we select a very sluggish tuning, namely
Kp1 = 0.5,
Kp2 = 1,

is meaningful because this yields, in general, a low overshoot


and a low settling time at the same time [24]. With the tuning
rule presented in Section 4, the response to a step set-point
change is, in principle, not oscillatory and therefore the
integrated absolute error can be substituted by the integrated
error which can be determined as follows. In the presence
of a set-point step change, by considering expressions (12)
and (36) (37) we have
1

Simulation results

Ti1 = 20,
Ti2 = 2,

Td1 = 0,
Td2 = 0,

Tf 1 = 0
Tf 2 = 0

(44)

(note that they are two PI controllers). The set-point unit step
response obtained with this cascade controller is shown in
Fig. 2, where a very slow step response can be observed.
The resulting integrated absolute error is IAEsp = 39.98. By
applying the identication strategy explained in Section 3,

(41)

Thus, after having applied the set-point step change with


given controllers, we can estimate the process parameters
and compare the obtained integrated absolute error with the
desired one given by expression (41). Indeed, a
performance index can be dened as [21]
J = 1
0

2As u
|e1 (t)| dt

(42)

In principle, the performance obtained by the control system


is considered to be satisfactory if J = 1. From a practical
point of view, however, based on extensive simulations, the
controller can be considered to be well-tuned if J . J with
IET Control Theory Appl., 2011, Vol. 5, Iss. 2, pp. 263270
doi: 10.1049/iet-cta.2010.0082

Fig. 2 Set-point step response with the initial tuning of the


controllers (example 1)
267

& The Institution of Engineering and Technology 2011

www.ietdl.org
we obtain u1 = 4.84, T01 = 16.01, u2 = 0.10, T02 = 2.00.
The performance index is J = 0.25, which suggests a new
tuning of the parameters. As a consequence of the
estimation procedure, we have l2 = 0.38 and l1 = 4.94
and the resulting PID parameters are [see (32) and (36)]
Kp1 = 1.13,
Kp2 = 0.104,

Ti1 = 11.17,
Ti2 = 0.050,

Td1 = 0.380,

Tf 1 = 0.038

Td2 = 1.90,

Tf 2 = 0.040
(45)

The set-point unit step response with the retuned controllers is


shown in Fig. 3, where the result obtained with the controllers
determined by applying the autotuning technique proposed in
[12] is also shown for the sake of comparison. Note the
change of the time scale with respect to Fig. 2. With the
controller parameters (45), the integrated absolute error is
IAEsp = 11.82, which yields a performance index
J = 0.84. With the method proposed in [12] the resulting
integrated absolute error is IAEsp = 12.048.
The performance obtained with the parameters (45) in the
load disturbance rejection task can be evaluated in Fig. 4.
In this case we have that the integrated absolute errors are
IAEd = 0.43 for the method proposed in this paper and
IAEd = 0.72 for the method proposed in [12].
Although it has to be noted that the comparison involves
techniques that have to be applied in different contexts and it is
therefore somewhat unfair, results conrm the effectiveness of
the proposed methodology. Actually, the tuning of the primary
controller might appear to be too aggressive for the set-point
following performance. It is worth stressing that, in any case, a
somewhat aggressive tuning allows one to achieve a high load
disturbance rejection performance and, as mentioned in
Remark 3, the overshoot in the set-point step response can be
easily reduced by low-pass ltering the set-point signal.
6.2

Example 2

As a second example we consider the following selfregulating process [7, 11]


P1 (s) =

es
,
(s + 1)2

P2 (s) =

e0.1s
1 + 0.1s

(46)

Fig. 4 Load disturbance step response of example 1


Solid line: PID controllers tuned with methodology proposed in this paper
Dashed line: PID controllers tuned with the methodology proposed in [12]

having applied the automatic tuning technique presented


therein, namely
Kp1 = 0.39,

Ti1 = 1.44,

Td1 = 0,

Tf 1 = 0

Kp2 = 0.53,

Ti2 = 0.17,

Td2 = 0,

Tf 2 = 0

(47)

The obtained set-point unit step response is shown in Fig. 5


(dotted line). We have IAEsp = 4.14. By applying the
identication method, we have u1 = 1.48, T01 = 3, u2 =
0.12, T02 = 0.2. As a consequence, the performance index
results J = 0.77. Even if the result can be considered to be
satisfactory, by applying the tuning procedure of Section 4
we obtain
Kp1 = 0.480,

Ti1 = 1.552,

Td1 = 0.027,

Tf 1 = 0.003

Kp2 = 0.400,

Ti2 = 0.055,

Td2 = 0.089,

Tf 2 = 0.011
(48)

The obtained set-point unit step response is shown again in


Fig. 5 as a solid line. In this case it is IAEsp = 3.57 and
J = 0.90.

As an initial tuning we consider that obtained in [11] after

Fig. 5 Set-point step response of example 2


Fig. 3 Set-point step response of example 1
Solid line: PID controllers tuned with methodology proposed in this paper
Dashed line: PID controllers tuned with the methodology proposed in [12]
268

& The Institution of Engineering and Technology 2011

Solid line: PID controllers tuned with methodology proposed in this paper
Dashed line: PID controllers tuned with the methodology proposed in [7]
Dotted line: PID controllers tuned with the methodology proposed in [11]
IET Control Theory Appl., 2011, Vol. 5, Iss. 2, pp. 263 270
doi: 10.1049/iet-cta.2010.0082

www.ietdl.org
The load disturbance rejection performance can be
evaluated in Fig. 6, where the method proposed in this
paper is compared again with the those described in [11, 7].
We have IAEd = 0.51 for the method proposed in [11],
IAEd = 0.45 for the method proposed in [7] and
IAEd = 0.16 for the method proposed in this paper.
Also this example conrms the effectiveness of the
proposed automatic tuning strategy.
6.3

Example 3

As a third example the following non-self-regulating process


is considered
P1 (s) =

e3s
,
s(1 + 10s)(1 + 2s)

P2 (s) =

e0.5s
1+s

(49)
Fig. 7 Set-point step response of example 3

A starting condition in which the controller has been badly


tuned has been considered; the initial settings are as follows
Kp1 = 0.1,
Kp2 = 1,

Ti1 = 50,
Ti2 = 2,

Td1 = 2,
Td2 = 0,

Tf 1 = 0.02
Tf 2 = 0

Solid line: PID controllers tuned with methodology proposed in this paper
Dashed line: PID controllers with the initial tuning

(50)

After a set-point unit step change the following parameters are


estimated (see Fig. 7)

m1 = 1, T01 = 15.37,
m2 = 1, T02 = 1.60,

u1 = 5.81
u2 = 0.50

(51)

The value of the performance index with the initial manual


tuning is J = 0.19 (note that IAEsp = 65.82 and, in case of a
unit step load disturbance, IAEd = 97.66, see Fig. 8). The
new values suggested for the controller parameters are as follows
Kp1 = 0.079,

Td1 = 9.79,

Tf 1 = 0.098

Kp2 = 0.467,

Ti2 = 0.252,

Td2 = 0.22,

Tf 2 = 0.504
(52)

The set-point and load disturbance unit step responses are


shown, respectively, in Figs. 7 and 8. The value of the
performance index is improved to J = 0.78 (IAEsp = 13.34
and IAEd = 7.45).

Fig. 8 Load disturbance step response of example 3


Solid line: PID controllers tuned with methodology proposed in this paper
Dashed line: PID controllers with the initial tuning

6.4

Example 4

As a nal example, another non-self-regulating process is


considered [19]
P1 (s) =

2e2s
,
s

P2 (s) =

4es
1+s

(53)

As a starting condition, the following initial settings have


been considered
Kp1 = 0.02,
Kp2 = 0.2,

Ti1 = 50,
Ti2 = 1,

Td1 = 0,
Td2 = 0,

Tf 1 = 0
Tf 2 = 0

(54)

After a set-point unit step change the following parameters are


estimated (see Fig. 9)

Fig. 6 Load disturbance step response of example 2


Solid line: PID controllers tuned with methodology proposed in this paper
Dashed line: PID controllers tuned with the methodology proposed in [7]
Dotted line: PID controllers tuned with the methodology proposed in [11]
IET Control Theory Appl., 2011, Vol. 5, Iss. 2, pp. 263270
doi: 10.1049/iet-cta.2010.0082

m1 = 2.00,
m2 = 4.03,

T01 = 1.99
T02 = 1.98

(55)

The two dead times have been evaluated as u1 = 2.49 and


u2 = 1.62. The integrals of absolute errors have the
269

& The Institution of Engineering and Technology 2011

www.ietdl.org
simultaneously tune the two PID controllers after having
estimated the process parameters by evaluating a set-point
step response. A performance assessment index has also
been devised to evaluate if the PID controllers need to be
retuned. Simulation results have conrmed the effectiveness
of the method and that it is capable to provide a high load
disturbance rejection performance.

Acknowledgments

This work has been supported partially by MIUR scientic


research funds.

9
Fig. 9 Set-point step response of example 4
Solid line: PID controllers tuned with methodology proposed in this paper
Dashed line: PID controllers with the initial tuning

Fig. 10 Load disturbance step response of example 4


Solid line: PID controllers tuned with methodology proposed in this paper
Dashed line: PID controllers with the initial tuning

following values: IAEsp = 36.21 after the setpoint step


change, corresponding to a performance index J = 0.23,
and IAEd = 90.44 after the load disturbance occurrence.
Hence, the proposed autotuning method gives the following
values of the parameters
Kp1 = 0.06,
Ti2 = 0.81,

Td1 = 0.0,
Td2 = 0.51,

Kp2 = 0.1
Tf 2 = 0.16

(56)

which allow are to obtain IAEsp = 8.89 after the setpoint step
change, corresponding to a performance index J = 0.82, and
IAEd = 36.28 after the load disturbance occurrence. Results
are shown in Figs. 9 and 10. The improvement is evident. It
is worth noting that the results achieved are very similar to
those obtained by using the method proposed in [19], which,
in any case, employs a much more complex control scheme.

Conclusions

In this paper we have proposed an automatic tuning procedure


for a cascade control system. The technique allows one to

270
& The Institution of Engineering and Technology 2011

References

strom, K.J., Hagglund, T.: Advanced PID control (ISA Press,


1 A
Research Triangle Park, NC, 2006)
2 ODwyer, A.: Handbook of PI and PID controller tuning rules
(Imperial College Press, 2006)
3 Krishnaswami, P.R., Rangaiah, G.P., Jha, R.K., Deshpande, P.D.:
When to use cascade control, Ind. Eng. Chem. Res., 1990, 29,
pp. 2163 2166
4 Kaya, I.: Improving performance using cascade control and a Smith
predictor, ISA Trans., 2001, 40, pp. 223234
5 Liu, T., Gu, D., Zhang, W.: Decoupling two-degree-of-freedom control
strategy for cascade control systems, J. Process Control, 2007, 15,
pp. 159167
6 Kaya, I., Tan, N., Atherton, D.P.: Improved cascade control structure
for enhanced performance, J. Process Control, 2007, 17, pp. 316
7 Hang, C.C., Loh, A.P., Vasnani, V.U.: Relay feedback auto-tuning of
cascade controllers, IEEE Trans. Control Syst. Technol., 1994, 2,
pp. 4245
8 Zhuang, M., Atherton, D.P.: Optimum cascade PID controller design
for SISO systems. Proc. Int. Conf. on Control 1994 (UKACC), 1994,
pp. 606611
9 Vivek, S., Chidambaram, M.: Cascade controller tuning by relay auto
tune method, J. Indian Inst. Sci., 2004, 84, pp. 8997
10 Yu, C.C.: Autotuning of PID controllers (Springer-Verlag, London,
UK, 1999)
11 Tan, K.K., Lee, T.H., Ferdous, R.: Simultaneous online automatic
tuning of cascade control for open loop stable processes, ISA Trans.,
2000, 39, pp. 233242
12 Leva, A., Donida, F.: Autotuning in cascaded systems based on a single
relay experiment, J. Process Control, 2009, 19, pp. 896 905
13 Visioli, A., Piazzi, A.: An automatic tuning method for cascade control
systems. Proc. IEEE Int. Conf. on Control Applications, Munich,
Germany, 2006, pp. 29682973
14 Lee, Y., Park, S., Lee, M.: PID controller tuning to obtain desired
closed loop responses for cascade control systems, Ind. Eng. Chem.
Res., 1998, 37, pp. 18591865
15 Lee, Y., Oh, S., Park, S.: Enhanced control with a general cascade
control structure, Ind. Eng. Chem. Res., 2002, 41, pp. 26792688
16 Arrieta, O., Vilanova, R., Balaguer, P.: Procedure for cascade control
system design, Int. J. Comput., Commun. Control, 2008, 3,
pp. 235248
17 Visioli, A.: Practical PID control (Springer, London, UK, 2006)
18 Morari, M., Zariou, E.: Robust process control (Prentice-Hall,
Englewood Cliffs, NJ, 1989)
19 Kaya, I., Atherton, D.P.: Improved cascade control structure for
controlling unstable and integrating processes. Proc. IEEE Int. Conf.
on Decision and Control European Control Conf., Sevilla, Spain,
2005, pp. 71337138
strom, K.J., Hagglund, T., Hang, C.C., Ho, W.K.: Automatic tuning
20 A
and adaptation for PID controllers a survey, Control Eng. Pract.,
1993, 1, pp. 699 714
21 Veronesi, M., Visioli, A.: Performance assessment and retuning of PID
controllers, Ind. Eng. Chem. Res., 2009, 48, pp. 26162623
22 Skogestad, S.: Simple analytic rules for model reduction and PID
controller tuning, J. Process Control, 2003, 13, pp. 291 309
23 Jelali, M.: An overview of control performance assessment technology
and industrial applications, Control Eng. Pract., 2006, 14, pp. 441 466
24 Shinskey, F.G.: Feedback controllers for the process industries
(McGraw-Hill, New York, NY, 1994)

IET Control Theory Appl., 2011, Vol. 5, Iss. 2, pp. 263 270
doi: 10.1049/iet-cta.2010.0082

Anda mungkin juga menyukai