You are on page 1of 13

Journal of Process Control 21 (2011) 69–81

Contents lists available at ScienceDirect

Journal of Process Control


journal homepage: www.elsevier.com/locate/jprocont

Tuning rules for optimal PID and fractional-order PID controllers


Fabrizio Padula, Antonio Visioli ∗
Dipartimento di Ingegneria dell’Informazione, University of Brescia, Via Branze 38, I-25123 Brescia, Italy

a r t i c l e i n f o a b s t r a c t

Article history: In this paper we present a set of tuning rules for standard (integer-order) PID and fractional-order PID con-
Received 25 February 2010 trollers. Based on a first-order-plus-dead-time model of the process, the tuning rules have been devised
Received in revised form 4 June 2010 in order to minimise the integrated absolute error with a constraint on the maximum sensitivity. The
Accepted 5 October 2010
achieved performance indexes can also be used for the assessment of the controller performance. Both set-
point following and load disturbance rejection tasks are considered. By comparing the results obtained for
the two kinds of controllers, it is shown that the use of fractional-order integral action is not advantageous,
Keywords:
while the use of a fractional-order derivative action provides a performance improvement.
Fractional-order controllers
PID control © 2010 Elsevier Ltd. All rights reserved.
Tuning
Optimisation
Performance assessment

1. Introduction damping property. However, they are not general because they are
valid only for first-order-plus-dead-time (FOPDT) processes with
Fractional-order proportional-integral-derivative (FOPID) con- specific values of the time constant and of the dead time, and
trollers have received a considerable attention in the last years both the gain cross-over frequency is selected a priori. A more general
from an academic and industrial point of view (see, for example tuning rule, which is valid for every FOPDT process, based on the
[1–5]). In fact, in principle, they provide more flexibility in the con- fractional maximum sensitivity constrained integral gain optimisa-
troller design, with respect to the standard PID controllers, because tion method has been proposed in [16]. However, the technique is
they have five parameters to select (instead of three). However, this restricted to PI controllers. Finally, with the aim of minimising the
also implies that the tuning of the controller can be much more effects of low frequency load disturbances, another set of tuning
complex. In order to address this problem, different methods for the rules has been developed in [17,18]. Indeed, it has to be remarked
design of a FOPID controller have been proposed in the literature. that the load disturbance rejection task is often overlooked in the
They are based on the use of evolutionary algorithms, where differ- literature related to fractional order controllers, despite this task is
ent objective functions are considered [6–8], or on the solution of a often of main concern in the process industry.
nonlinear constrained optimisation problem where, in particular, In this paper we propose a new set of tuning rules for PID and
the iso-damping property (namely, the robustness to variations in FOPID controllers based on the minimisation of the integrated abso-
the gain of the process) is considered [9–11]. lute error (which is meaningful because this yields, in general, a low
It is however recognised that, for their widespread use in indus- overshoot and a low settling time at the same time [19]), subject
try, FOPID controllers should possess the same ease of use of to a constraint on the maximum sensitivity (as in the well-known
standard PID controllers and the improvement in the performance Kappa–Tau tuning rules for standard PID controllers [20]) in order
they are capable to provide should be clear. Actually, one of the to provide a required level of robustness. It will be shown that, in
reasons of the great success of standard PID controllers is the pres- this context, if the controller is restricted to a PI structure, then
ence of a lot of tuning rules [12] and of the automatic tuning feature the use of a fractional integral action is not useful in improving
[13] that simplify significantly their design. For this reason, tuning the performance and therefore a standard PI controller is the best
rules for FOPID controllers have been proposed in the literature. option. Conversely, the use of a fractional derivative action allows
In [14,15] they have been developed in order to achieve the iso- to improve the control performance. Both the set-point following
and the load disturbance rejection tasks will be considered explic-
itly. An analytical expression of the performance index is also given
∗ Corresponding author. Tel.: +39 030 3715460; fax: +39 030 380014. and this can be exploited in a performance assessment context.
E-mail addresses: fabrizio.padula@ing.unibs.it (F. Padula), The paper is organised as follows. In Section 2 the problem is
antonio.visioli@ing.unibs.it (A. Visioli). formulated. The tuning rules are described in Section 3. Simula-

0959-1524/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jprocont.2010.10.006
70 F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81

d
10

r e 9
C P y
8

7
Fig. 1. The considered control scheme. 6

IAEsp
5

8 4

3
7
2
6 1

5 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Kp

normalised dead time τ


4
Fig. 3. Determination of the performance index function IAEsp for a PI controller for
3 set-point following (Ms = 1.4). Plus sign: optimal value of IAEsp . Solid line: interpo-
lating function (9).
2

1 where, evidently, K is the gain, T is the time constant and L is the


dead time. The process dynamics can be conveniently characterised
by the normalised dead time defined as L/T or, equivalently, as
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
normalised dead time τ L
= , (2)
L+T
Fig. 2. Determination of the Kp tuning rule for a PI controller for set-point following
(Ms = 1.4). Plus sign: optimal value of Kp . Solid line: interpolating function (12). which represents a measure of difficulty in controlling the process.
Note that, using (2) we have, in general,  ∈ [0, 1]. In this paper,
values of the normalised dead time in the range 0.05 ≤  ≤ 0.8 have
tion results are presented in Section 4, where a comparison with
been considered. Actually, for values of  < 0.05 the dead time can
other tuning rules is performed. Conclusions are drawn in Section
be virtually neglected and the design of a controller is rather trivial,
5.
while for values of  ≥ 0.8 the process is significantly dominated by
the dead time and therefore a dead time compensator should be
2. Problem formulation
employed.
The standard PID controller is defined in series (interacting)
We consider the unity-feedback control scheme of Fig. 1 where
form as
the process is assumed to have a FOPDT dynamics, namely,
Ti s + 1 Td s + 1
K C(s) = Kp , (3)
P(s) = e−Ls , (1) Ti s Td s + 1
Ts + 1 N

Table 1
Kp tuning rule parameters for a PI controller.

Control task Ms = 1.4 Ms = 2.0

a b c a b c

Set-point 0.3220 −1.049 −0.2292 0.5261 −1.074 −0.2157


Load disturbance 0.2958 −1.014 −0.2021 0.5327 −1.029 −0.2428

Table 2
Ti tuning rule parameters for a PI controller.

Control task Ms = 1.4 Ms = 2.0

a b c a b c

Set-point 0.1726 1.156 0.9907 0.4626 0.9252 0.9393


Load disturbance 1.624 0.2269 −0.5556 1.44 0.4825 −0.1019

Table 3
Kp tuning rule parameters for an integer-order PID controller.

Control task Ms = 1.4 Ms = 2.0

a b c a b c

Set-point 0.429 −1.032 −0.4654 0.6566 −1.076 −0.6705


Load disturbance 0.1724 −1.259 −0.05052 0.2002 −1.414 0.06139
F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81 71

Table 4
Ti tuning rule parameters for an integer-order PID controller.

Control task Ms = 1.4 Ms = 2.0

a b c a b c

Set-point 0.06949 −0.5366 0.7119 0.1604 −0.1288 0.7814


Load disturbance 0.5968 0.6388 0.07886 0.446 0.9541 0.1804

Table 5
Td tuning rule parameters for an integer-order PID controller.

Control task Ms = 1.4 Ms = 2.0

a b c a b c

Set-point 0.4568 1.153 0.001962 0.49 1.069 0.001149


Load disturbance 0.5856 0.5004 −0.1109 0.6777 0.4968 −0.1499

where Kp is the proportional gain, Ti is the integral time constant, Td that an additional first-order filter has been employed in (4) in
is the derivative time constant. The fractional-order PID controller order to make the controller proper. Indeed, in order to implement
is defined as a generalisation of the standard PID controller (3), the fractional-order controller, the Oustaloup continuous integer-
namely, order approximation [21] has been employed. It consists in using
the following approximation based on a recursive distribution of
Ti s + 1 Td s + 1
C(s) = Kp Td
, (4) zeros and poles:
Ti s s+1
N

where  and  are the noninteger orders of the integral and deriva- 
N s
1 + ωz,n
tive terms respectively. It can be easily noted that, by selecting s ∼
=k s ,  > 0, (5)
1+ ωp,n
 =  = 1, a standard PID controller is obtained. It can also be noted n=1

SP Ms=1.4 SP Ms=2.0
45 60

40
50
35

30 40
IAE increment [%]
IAE increment [%]

25
30
20

15 20

10
10
5

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
normalised dead time normalised dead time

LD Ms=1.4 LD Ms=2.0
140
80
120
70

60 100
IAE increment [%]
IAE increment [%]

50 80

40
60
30
40
20
20
10

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
normalised dead time normalised dead time

Fig. 4. Increment of the IAE value by using optimal PI controllers (dashed line) or optimal integer-order PID controllers (solid line) with respect to optimal fractional-order
PID controllers. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left: load disturbance step response with Ms = 1.4.
Bottom-right: load disturbance step response with Ms = 2.0.
72 F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81

SP 1.4 SP 2.0
1.2 1.2
process variable

1 1

process variable
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
time time
6 8
5
control variable

control variable
4
3 4
2
2
1
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
time time
LD 1.4 LD 2.0
0.3
0.2
0.25
process variable

process variable
0.2 0.15
0.15 0.1
0.1
0.05 0.05
0 0
−0.05
0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 8
time time

0 0
control variable

−0.2
control variable

−0.4 −0.5
−0.6
−0.8
−1
−1
−1.2
−1.5
0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 8
time time

Fig. 5. Simulation results for process P1 (s) with a PI controller. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left:
load disturbance step response with Ms = 1.4. Bottom-right: load disturbance step response with Ms = 2.0. Solid line: proposed tuning rules for PI controllers. Dashed line:
tuning rules proposed in [16]. Dash-dot line: tuning rules proposed in [18].

which is valid in a frequency range [ωl , ωh ] and where the gain A usually specified control requirement is to minimise the inte-
k is adjusted so that the right side of (5) has unity gain at the grated absolute error
gain crossover frequency of s . In this paper the value N = 8 has  ∞  ∞
been chosen, while ωl and ωh have been selected as 0.001ωc and IAE = |e(t)|dt = |r(t) − y(t)|dt, (7)
1000ωc respectively, where ωc is the gain crossover frequency. It 0 0
is well-known that with standard PID controllers, the choice of the
because this yields, in general, a low overshoot and a low settling
derivative filter can be critical [22]. Actually, this aspect has been
time at the same time [19] in the set-point or load disturbance
often overlooked in fractional-order PID controllers. Here we select
step response. However, aiming at just obtaining the theoretical
N = 10 if T = 1, so that the pole is at frequency ω = 10/Td , namely, it
minimum integrated absolute error that can be achieved for the
does not influence the controller dynamics significantly and it does
single-loop system might not be sensible in practical cases because
filter the high-frequency noise at the same time (note that this is
the robustness issue and the control effort have also to be taken
a typical choice in standard PID controllers [23]). In all the other
into account. For this reason, the devised tuning rules aim at min-
cases, the value of N is conveniently modified in order to rigidly shift
imising the integrated absolute error by constraining at the same
the Bode magnitude plot along the abscissa axis without changing
time the maximum sensitivity (as in the well-known Kappa–Tau
its shape, (i.e., in order to scale the system step response propor-
tuning rules for standard PID controllers [20]), which is defined as
tionally to the value of T when the normalised dead time value is
kept constant): 1
Ms = max , (8)
ω ∈ [0,+∞) 1 + C(s)P(s)

and which represents also the inverse of the maximum distance


N = 10T (−1) . (6) of the Nyquist plot from the critical point (−1,0). Obviously, the
F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81 73

SP 1.4 SP 2.0
1.5
1.2

process variable
process variable

1
0.8 1
0.6
0.4 0.5
0.2
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time time
15 30
25

control variable
control variable

20
10
15
10
5 5
0
0 −5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time time
LD 1.4 LD 2.0
0.4 0.4

process variable
process variable

0.3 0.3
0.2 0.2

0.1 0.1

0 0

−0.1 −0.1
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
time time
0 0
−0.2
control variable
control variable

−0.4
−0.5
−0.6
−0.8
−1 −1
−1.2
−1.4 −1.5
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
time time

Fig. 6. Simulation results for process P1 (s) with a PID controller. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left:
load disturbance step response with Ms = 1.4. Bottom-right: load disturbance step response with Ms = 2.0. Solid line: proposed tuning rules for fractional-order PID controllers.
Dashed line: proposed tuning rules for integer-order PID controllers. Dash-dot line: AMIGO tuning rules. Dotted line: tuning rules proposed in [15].

higher the value of Ms is, the less robust is the system to modelling to derive suitable tuning rules [13,20,26]. In this context, different
uncertainties. interpolating functions have been considered [27], by taking into
account the aim of providing tuning rules where the resulting per-
formance is scaled by the time constant T (see Section 4.4). As a
3. Optimal tuning
result, the following (general) structure for the controller parame-
ters has been devised (obviously, for an integer-order controller it
In order to find the tuning rules for the minimisation of the
is  =  = 1):
integrated absolute error by constraining the maximum sensitiv-
ity value, an approach similar to that employed in [24] has been 1
Kp = (a b + c), (9)
used. In particular, the set-point following and the load disturbance K
rejection tasks have been considered separately and different pro-   b 
 L
cesses with different values of the normalised dead time have been Ti = T a +c , (10)
T
considered. For each of them, the values of the parameters of both
integer and fractional-order PI and PID controllers have been found
  b 
L
by means of a genetic algorithm [25] (which are known to provide a Td = T  a +c , (11)
T
global optimum of a problem in a stochastic frame) in order to min-
imise the integrated absolute error in a step response, by imposing while piecewise constant functions have been considered for 
two typical values of the maximum sensitivity, namely, Ms = 1.4 and and . The different cases are then considered separately in the
Ms = 2.0 [20]. Note that Ms = 1.4 and Ms = 2 are two different signif- following subsections.
icant cases in the range of suitable values, namely, one where the
robustness issue is of primary concern (Ms = 1.4) and one where the 3.1. PI controller
aggressiveness is more important (Ms = 2.0) (see Section 4). Finally,
for each considered controller, the optimal coefficients found for If a PI controller is considered, (namely, Td = 0), as a result
the different values of L/T or , have been interpolated in order of the optimisation procedure for fractional-order PI controllers,
74 F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81

SP 1.4 SP 2.0
1.2 1.2
process variable

process variable
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 5 10 15 0 1 2 3 4 5 6 7 8 9 10
time time
2
1.5
control variable

control variable
1.5
1
1
0.5
0.5

0 0
0 5 10 15 0 1 2 3 4 5 6 7 8 9 10
time time
LD 1.4 LD 2.0
0.8 0.8

process variable
process variable

0.6 0.6

0.4 0.4

0.2 0.2

0 0

−0.2 −0.2
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15
time time

0 0
control variable

control variable

−0.2 −0.2
−0.4 −0.4
−0.6 −0.6
−0.8 −0.8
−1 −1
−1.2 −1.2
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15
time time

Fig. 7. Simulation results for process P2 (s) with a PI controller. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left:
load disturbance step response with Ms = 1.4. Bottom-right: load disturbance step response with Ms = 2.0. Solid line: proposed tuning rules for PI controllers. Dashed line:
tuning rules proposed in [16]. Dash-dot line: tuning rules proposed in [18].

Table 6
Kp tuning rule parameters for a fractional-order PID controller.

Control task Ms = 1.4 Ms = 2.0

a b c a b c

Set-point 0.6503 −0.9166 −0.6741 0.9294 −0.933 −0.9205


Load disturbance 0.2776 −1.095 −0.1426 0.1804 −1.449 0.2319

the value  = 1 is obtained for all the cases addressed. This employed in the different cases are shown in Tables 1 and 2.
means that, for the problem considered, there is no point in As an example of how the tuning rules have been obtained,
using a fractional-order integrator. The tuning rules obtained the optimal values of Kp for the different values of  and
by interpolating the results are therefore (9) and (10) with the corresponding interpolating function has been plotted in
 = 1 where the values of the parameters a, b and c to be Fig. 2.

Table 7
Ti tuning rule parameters for a fractional-order PID controller.

Control task Ms = 1.4 Ms = 2.0

a b c a b c

Set-point 0.04701 −0.2611 0.9276 −0.001427 −1.003 1.031


Load disturbance 0.6241 0.5573 0.0442 0.6426 0.8069 0.05627
F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81 75

SP 1.4 SP 2.0
1.4 1.5
1.2
process variable

process variable
1 1
0.8
0.6
0.4 0.5
0.2
0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
time time
50 50
control variable

control variable
0 0

−50 −50
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
time time
LD 1.4 LD 2.0
0.6 0.6
process variable

process variable
0.4 0.4
0.2 0.2
0 0
−0.2 −0.2
−0.4
0 5 10 15 20 25 −0.4
0 5 10 15 20 25 30
time
time

0 0
control variable

control variable

−0.5 −0.5

−1 −1

−1.5 −1.5
0 5 10 15 20 25 0 5 10 15 20 25 30
time time

Fig. 8. Simulation results for process P2 (s) with a PID controller. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left:
load disturbance step response with Ms = 1.4. Bottom-right: load disturbance step response with Ms = 2.0. Solid line: proposed tuning rules for fractional-order PID controllers.
Dashed line: proposed tuning rules for integer-order PID controllers. Dash-dot line: AMIGO tuning rules. Dotted line: tuning rules proposed in [15].

Table 8
Td tuning rule parameters for a fractional-order PID controller.

Control task Ms = 1.4 Ms = 2.0

a b c a b c

Set-point 0.3563 1.2 0.0003108 0.4203 1.229 0.01822


Load disturbance 0.4793 0.7469 −0.02393 0.597 0.5568 −0.09536

For the purpose of the performance assessment task, it is also by first calculating a performance index based on the available data
useful to determine the optimal value of the integrated absolute and then by evaluating the current control performance against a
error as a function of the process parameters. In fact, the assess- selected benchmark, which represents the desired performance.
ment of the performance of a control loop is generally performed The same approach used for the controller parameters have been

Table 9
 and  tuning rule parameters for a fractional-order PID controller.

Control task Ms = 1.4 Ms = 2.0

   

1.0 if  < 0.1


1.1 if  < 0.1
Set-point 1 1 1.1 if 0.1 ≤  < 0.4
1.2 if 0.1 ≤ 
1.2 if 0.4 ≤ 
1.0 if  < 0.1 1.0 if  < 0.2
Load disturbance 1 1.1 if 0.1 ≤  < 0.4 1 1.1 if 0.2 ≤  < 0.6
1.2 if 0.4 ≤  1.2 if 0.6 ≤ 
76 F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81

SP 1.4 SP 2.0
1.2
1.2
process variable

process variable
1
0.8
0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30
time time
1.2 1.2
control variable

control variable
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30
time time
LD 1.4 LD 2.0
1 1
0.8
process variable

0.8

process variable
0.6 0.6
0.4 0.4
0.2
0.2
0
0
−0.2
−0.2
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30
time time
0 0
control variable

control variable

−0.2 −0.2
−0.4 −0.4
−0.6 −0.6
−0.8 −0.8
−1 −1
−1.2 −1.2
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30
time time

Fig. 9. Simulation results for process P3 (s) with a PI controller. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left:
load disturbance step response with Ms = 1.4. Bottom-right: load disturbance step response with Ms = 2.0. Solid line: proposed tuning rules for PI controllers. Dashed line:
tuning rules proposed in [16]. Dash-dot line: tuning rules proposed in [18].

employed also for the determination of the integrated absolute shown in Tables 3–5. The optimal values of the integrated absolute
error function, namely, the optimal IAE values for different values error as a function of the process parameters can be expressed as
of  have been interpolated with a suitable function. The desired
performance can be therefore expressed as

T (0.0020 exp(9.837) + 0.8695 exp(2.270) + 1.158 − 0.9188) if Ms = 1.4
IAEsp = , (12)
T (0.0021 exp(9.046) + 0.5488 exp(2.244) + 1.292 − 0.5836) if Ms = 2.0

for the set-point following task and



KT (0.8220 exp(2.538) + 0.0261 exp(6.453) + 0.0877 − 0.9474) if Ms = 1.4
IAEld = , (13)
KT (0.0234 exp(6.485) + 0.3228 exp(2.474) + 0.5557 − 0.4255) if Ms = 2.0

for the load disturbance rejection task. As an illustrative example,


the interpolation for the set-point following case with Ms = 1.4 is
Table 10
plotted in Fig. 3. Comparison between tuning rules for integer and fractional-order PI controller with
 = 0.0909.
3.2. Integer-order PID controller Tuning rules Kp Ti  Ms IAEsp IAEId

SP 1.4 3.75 1.00 1 1.40 0.27 0.27


In case the derivative action is employed, the results obtained SP 2.0 6.70 0.99 1 1.92 0.22 0.15
for integer and fractional-order PID controllers are different. For LD 1.4 3.16 0.41 1 1.44 0.40 0.14
integer-order PID controllers, the tuning rules obtained for Kp and Ti LD 2.0 6.04 0.37 1 2.02 0.30 0.062
and Td are given by expressions (9)–(11) where, obviously,  =  = 1 [16] 3.26 0.41 0.70 1.71 0.69 0.54
[18] 3.39 0.74 1.12 1.35 0.46 0.31
and the values of the parameters a, b and c for the different cases are
F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81 77

SP 1.4 SP 2.0
1.4
1
1.2
process variable

process variable
0.8 1
0.6 0.8
0.6
0.4
0.4
0.2 0.2
0 0
0 5 10 15 20 25 30 0 2 4 6 8 10 12 14 16 18 20
time time
3 5
control variable

2.5 4

control variable
2 3
1.5 2
1 1
0.5 0
0 −1
0 5 10 15 20 25 30 0 2 4 6 8 10 12 14 16 18 20
time time
LD 1.4 LD 2.0
1 1
0.8
process variable

process variable
0.8
0.6
0.6
0.4
0.4 0.2
0.2 0
−0.2
0
−0.4
0 5 10 15 20 25 0 2 4 6 8 10 12 14 16 18 20
time time

0 0
control variable

control variable

−0.2
−0.4 −0.5

−0.6
−1
−0.8
−1
−1.5
0 5 10 15 20 25 0 2 4 6 8 10 12 14 16 18 20
time time

Fig. 10. Simulation results for process P3 (s) with a PID controller. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left:
load disturbance step response with Ms = 1.4. Bottom-right: load disturbance step response with Ms = 2.0. Solid line: proposed tuning rules for fractional-order PID controllers.
Dashed line: proposed tuning rules for integer-order PID controllers. Dash-dot line: AMIGO tuning rules.


T (0.0004 exp(12.279) + 0.6107 exp(2.0849) + 1.390 − 0.6546) if Ms = 1.4
IAEsp = , (14)
T (0.0020 exp(9.202) + 0.4256 exp(2.3008) + 0.6196 − 0.4405) if Ms = 2.0
for the set-point following task and

KT (0.5933 exp(2.587) + 0.0070 exp(8.0967) + 0.0850 − 0.6888) if Ms = 1.4
IAEld = , (15)
KT (0.0059 exp(8.116) + 0.1047 exp(2.8826) + 1.169 − 0.1882) if Ms = 2.0
for the load disturbance rejection task.

Table 11
Comparison between tuning rules for integer and fractional-order PID with  = 0.0909.

Tuning rules Kp Ti  Td  Ms IAEsp IAEld

SP 1.4 I 4.63 0.95 1 0.034 1 1.41 0.22 0.21


SP 2.0 I 8.00 1.00 1 0.043 1 2.02 0.15 0.12
LD 1.4 I 3.48 0.22 1 0.074 1 1.44 0.32 0.081
LD 2.0 I 6.00 0.23 1 0.066 1 1.95 0.23 0.042
AMIGO 1.4 4.77 0.43 1 0.038 1 1.39 0.28 0.093
AMIGO 2.0 8.67 0.20 1 0.043 1 1.97 0.28 0.034
SP 1.4 F 5.18 1.01 1 0.023 1.1 1.42 0.199 0.20
SP 2.0 F 7.78 1.02 1 0.043 1 1.97 0.15 0.13
LD 1.4 F 3.69 0.22 1 0.062 1 1.42 0.32 0.078
LD 2.0 F 6.06 0.16 1 0.070 1 2.14 0.25 0.033
[15] 8.67 0.20 1.12 0.043 0.1534 1.10 1.48 1.30
78 F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81

SP 1.4 SP 2.0
1.5
1

process variable
process variable

0.8
1
0.6
0.4 0.5
0.2
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 35 40
time time
6 10
5
control variable

control variable
8
4
6
3
4
2
1 2

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 35 40
time time
LD 1.4 LD 2.0
0.8
0.8
process variable

0.6

process variable
0.6 0.4
0.4 0.2

0.2 0
−0.2
0
−0.4
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
time time
0
0
control variable

control variable

−0.2
−0.5
−0.4
−0.6
−1
−0.8
−1
−1.5
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. 11. Simulation results for process P4 (s) with a PID controller. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left:
load disturbance step response with Ms = 1.4. Bottom-right: load disturbance step response with Ms = 2.0. Solid line: proposed tuning rules for fractional-order PID controllers.
Dashed line: proposed tuning rules for integer-order PID controllers. Dash-dot line: AMIGO tuning rules.

3.3. Fractional-order PID controller 3.4. Comparison

If a fractional-order PID controller (4) is considered, the optimal It is important at this point to evaluate quantitatively the
value  = 1 is still found (see Section 3.1), while an optimal value improvement of performance that can be achieved using the
 =/ 1 is found in general. The structure of the tuning rules for the derivative action and using a fractional-order controller with
PID parameters Kp , Ti and Td is the same of (9), (10) and (11), where respect to an integer-order one. The increment, in percentage, that
the values of a, b and c are shown in Tables 6–8. The tuning rules results using a PI controller tuned according to the formulae shown
for  and  are shown in Table 9. For the purpose of performance in Section 3.1 or an integer-order PID controller tuned according to
assessment, the optimal values of the integrated absolute error as the rules presented in Section 3.2 with respect to a fractional-order
a function of the process parameters can be expressed as PID controller, tuned according to the method proposed in Section


T (0.0011 exp(10.414) + 0.6304 exp(2.2226) + 0.7910 − 0.6547) if Ms = 1.4
IAEsp = , (16)
T (0.0017 exp(9.258) + 0.4573 exp(2.2474) + 0.4777 − 0.4711) if Ms = 2.0

3.3, is shown in Fig. 4 for both the set-point following and load
for the set-point following task and disturbance rejection task.


KT (0.6512 exp(2.394) + 0.0031 exp(8.663) + 0.1353 − 0.7456) if Ms = 1.4
IAEld = , (17)
KT (0.0025 exp(8.995) + 0.1188 exp(2.836) + 1.193 − 0.2005) if Ms = 2.0

It can be seen that, in general, the use of the derivative action


for the load disturbance rejection task. allows to improve the performance significantly and that the use
F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81 79

SP 1.4 SP 2.0
1.5
1

process variable
process variable

0.8 1
0.6
0.4 0.5
0.2
0 0
−0.2
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15
time time
5 10

control variable
control variable

4 8

3 6

2 4

1 2

0 0
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15
time time
LD 1.4 LD 2.0
0.8 0.8
process variable

process variable
0.6 0.6

0.4 0.4
0.2
0.2
0
0
−0.2
−0.2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
time time
0.2 0.5
control variable

0
control variable

−0.2 0
−0.4
−0.5
−0.6
−0.8 −1
−1
−1.5
0 5 10 15 20 25 30 0 5 10 15 20 25 30
time time

Fig. 12. Simulation results for process P5 (s) with a PID controller. Top-left: set-point step response with Ms = 1.4. Top-right: set-point step response with Ms = 2.0. Bottom-left:
load disturbance step response with Ms = 1.4. Bottom-right: load disturbance step response with Ms = 2.0. Solid line: proposed tuning rules for fractional-order PID controllers.
Dashed line: proposed tuning rules for integer-order PID controllers. Dash-dot line: AMIGO tuning rules.

of the fractional-order derivative action is convenient for all the Table 12


Comparison between tuning rules for integer and fractional-order PI controller with
processes and for all the considered control tasks. Obviously, the
 = 0.4.
performance improvement is paid by the higher controller order
and this trade-off should be evaluated in a given application. Tuning rules Kp Ti  Ms IAEsp IAEld

SP 1.4 0.61 1.10 1 1.42 1.80 1.79


4. Simulation results SP 2.0 1.19 1.26 1 2.04 1.36 1.05
LD 1.4 0.55 0.93 1 1.42 1.75 1.70
LD 2.0 1.12 1.08 1 2.05 1.41 0.97
4.1. FOPDT processes [16] 0.74 0.71 1 1.89 1.74 1.27
[18] 0.47 1.19 1.12 1.26 3.18 3.08
In order to verify the effectiveness of the tuning rules proposed,
the following processes with k = 1 and T = 1 (and with different Table 13
values of the dead time L) are first considered (the value of the Comparison between tuning rules for integer and fractional-order PID with  = 0.04.
normalised dead time is also shown): Tuning rules Kp Ti  Td  Ms IAEsp IAEld
1 −0.1s SP 1.4 I 0.64 0.80 1 0.29 1 1.42 1.34 1.26
P1 (s) = e ,  = 0.0909, (18)
s+1 SP 2.0 I 1.09 0.95 1 0.32 1 1.92 0.95 0.87
LD 1.4 I 0.50 0.54 1 0.37 1 1.43 1.42 1.21
1 −0.67s
P2 (s) = e ,  = 0.4, (19) LD 2.0 I 0.79 0.48 1 0.40 1 1.98 1.23 0.74
s+1 AMIGO 1.4 0.86 0.95 1 0.23 1 1.39 1.37 1.20
1 −2.5s AMIGO 2.0 1.53 0.67 1 0.25 1 2.05 1.57 0.85
P3 (s) = e ,  = 0.7143. (20) SP 1.4 F 0.83 0.98 1 0.22 1.2 1.42 1.26 1.20
s+1 SP 2.0 F 1.26 1.03 1 0.27 1.2 2.15 0.92 0.83
The tuning rules presented in the previous sections have been LD 1.4 F 0.61 0.54 1 0.33 1.2 1.44 1.38 1.12
LD 2.0 F 0.91 0.52 1 0.38 1.1 1.95 1.15 0.70
applied and the results for the different cases are summarised
[15] 0.91 0.52 1.35 0.38 1.09 1.69 4.31 3.91
in Tables 10–15, where the obtained values of the controller
80 F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81

Table 14 order controllers [13] have been considered (with both Ms = 1.4
Comparison between tuning rules for integer and fractional-order PI with  = 0.7143.
and Ms = 2.0). In particular, the rule that gives the best results have
Tuning rules Kp Ti  Ms IAEsp IAEld been considered between the two proposed sets in [15] (they aims
SP 1.4 0.23 1.49 1 1.40 6.50 6.50 at achieving different phase margins). Then, in order to provide a
SP 2.0 0.54 2.02 1 2.11 4.37 4.06 fair comparison, the ideal-form PID controller considered in [13]
LD 1.4 0.21 1.44 1 1.38 6.75 6.74 have been augmented with a low-pass first-order filter whose cut-
LD 2.0 0.51 2.14 1 1.94 4.30 4.19 off frequency has been selected as ten times the frequency of the
[16] 0.42 1.77 1.1 1.89 5.94 5.75
fastest zero of the controller in case of real zeros and as 10 times
[18] 0.16 1.74 1.12 1.26 10.56 10.49
the resonance frequency in case of complex conjugate zeros.
The set-point and load disturbance step responses are shown in
Figs. 5–10 for the different cases.
parameters as well as the obtained value of Ms and of the inte-
grated absolute error in both the set-point and load disturbance
step response are shown. Note that the tuning rule employed is 4.2. High-order process
described as SP or LD (which means that the set-point following or
the load disturbance rejection task is addressed, respectively) fol- The following high-order process is considered:
lowed by the target maximum sensitivity and the letter ‘I’ or ‘F’ if 1
P4 (s) = . (21)
the controller is of integer order or of fractional order respectively (s + 1)8
(note that for the PI case there is no need of distinguishing between
‘I’ and ‘F’). A comparison with other methods proposed in the liter- After having approximated it with a FOPDT transfer function
ature has also been performed. In particular, for fractional-order PI with K = 1, T = 3.06, and L = 4.95, the application of the tuning rules
controllers the methods proposed in [16] and [18] have been con- for integer-order and fractional-order PID controllers gives the
sidered. For PID controllers, the tuning rules proposed in [15] for results shown in Table 16, while the set-point and load disturbance
fractional-order controllers and the AMIGO tuning rules for integer- step responses are plotted in Fig. 11.

Table 15
Comparison between tuning rules for integer and fractional-order PID with  = 0.7143.

Tuning rules Kp Ti  Td  Ms IAEsp IAEld

SP 1.4 I 0.14 0.75 1 1.32 1 1.44 5.32 5.32


SP 2.0 I 0.27 0.92 1 1.31 1 1.96 3.60 3.40
LD 1.4 I 0.21 1.15 1 0.82 1 1.41 5.41 5.40
LD 2.0 I 0.38 1.25 1 0.92 1 1.99 3.58 3.31
AMIGO 1.4 0.35 1.70 1 0.62 1 1.38 4.86 4.85
AMIGO 2.0 0.60 1.38 1 0.69 1 2.08 4.66 4.21
SP 1.4 F 0.21 0.96 1 1.07 1.2 1.38 4.81 4.80
SP 2.0 F 0.35 1.03 1 1.31 1.2 2.08 3.40 3.27
LD 1.4 F 0.26 1.08 1 0.93 1.2 1.44 4.44 4.42
LD 2.0 F 0.53 1.40 1 0.90 1.2 2.34 3.40 2.96
[15] Not applicable

Table 16
Comparison between tuning rules for integer and fractional-order PID with the high-order process P4 (s).

Tuning rules Kp Ti  Td  Ms IAEsp IAEld

SP 1.4 I 0.24 2.34 1 2.44 1 1.45 9.77 9.76


SP 2.0 I 0.43 2.85 1 2.51 1 1.94 7.63 6.87
LD 1.4 I 0.27 2.72 1 1.94 1 1.42 10.26 10.25
LD 2.0 I 0.46 2.71 1 2.17 1 2.02 8.25 7.13
AMIGO 1.4 0.45 4.17 1 1.40 1 1.39 9.50 9.44
AMIGO 2.0 0.78 3.27 1 1.58 1 2.10 11.25 9.34
SP 1.4 F 0.34 2.96 1 2.43 1.2 1.39 9.25 9.20
SP 2.0 F 0.54 3.15 1 2.97 1.2 2.08 6.88 6.48
LD 1.4 F 0.33 2.63 1 2.53 1.2 1.43 8.95 8.81
LD 2.0 F 0.59 3.07 1 2.62 1.2 2.09 7.28 6.40
[15] Not applicable

Table 17
Comparison between tuning rules for integer and fractional-order PID with the non-minimum-phase process P5 (s).

Tuning rules Kp Ti  Td  Ms IAEsp IAEld

SP 1.4 I 0.27 1.24 1 1.16 1 1.45 4.66 4.90


SP 2.0 I 0.48 1.51 1 1.20 1 1.94 3.20 3.42
LD 1.4 I 0.28 1.37 1 0.97 1 1.42 4.87 5.10
LD 2.0 I 0.48 1.34 1 1.09 1 2.03 3.43 3.19
AMIGO 1.4 0.48 2.12 1 0.69 1 1.39 4.58 4.76
AMIGO 2.0 0.83 1.64 1 0.78 1 2.10 4.66 3.99
SP 1.4 F 0.37 1.57 1 1.01 1.2 1.39 4.46 4.66
SP 2.0 F 0.59 1.67 1 1.24 1.2 2.08 3.15 3.32
LD 1.4 F 0.35 1.33 1 1.10 1.2 1.42 4.32 4.45
LD 2.0 F 0.61 1.51 1 1.10 1.1 2.16 3.09 2.91
[15] Not applicable
F. Padula, A. Visioli / Journal of Process Control 21 (2011) 69–81 81

4.3. Non-minimum-phase process A large number of simulation results confirm the effectiveness
of the devised tuning rules also with respect to other tuning rules
As a final illustrative example, the following non-minimum- proposed previously in the literature.
phase process is considered:
1−s References
P5 (s) = 3
. (22)
(s + 1) [1] I. Podlubny, Fractional-order systems and PI D controllers, IEEE Transactions
on Automatic Control 44 (1999) 208–214.
After having approximated it with a FOPDT transfer function [2] D. Valerio, Fractional robust system control, Ph.D. Thesis, University of Lisbon
with K = 1, T = 1.62, and L = 2.39, the application of the tuning rules (P), 2005.
for integer-order and fractional-order PID controllers gives the [3] C.A. Monje, Design methods of fractional order controllers for industrial appli-
cations, Ph.D. Thesis, University of Extremadura (E), 2006.
results shown in Table 17, while the set-point and load disturbance [4] B.M. Vinagre, C.A. Monje, A.J. Calderon, J.I. Suarez, Fractional PID controllers for
step responses are plotted in Fig. 12. industry application. A brief introduction, Journal of Vibration and Control 13
(2007) 1419–1429.
[5] C.A. Monje, B.M. Vinagre, V. Feliu, Y.Q. Chen, Tuning and auto-tuning of frac-
4.4. Discussion
tional order controllers for industry applications, Control Engineering Practice
16 (2008) 798–812.
From the results presented, it can be seen that the proposed [6] D. Maiti, A. Acharya, M. Chakraborty, A. Konar, R. Janarthanan, Tuning PID
and PI Dı controllers using the integral time absolute error criterion, in: Pro-
tuning rules have the following features.
ceedings IEEE International Conference on Information and Automation for
Sustainability, Colombo (CL), 2008.
• Despite the approximation introduced by the interpolation and [7] A. Biswas, S. Das, A. Abraham, S. Dasgupta, Design of fractional-order PI D
(in case of fractional-order controller) by the Oustaloup contin- controllers with an improved differential evolution, Engineering Applications
of Artificial Intelligence 22 (2009) 343–350.
uous integer-order approximation, the achieved performance is [8] M. Zamani, M. Karimi-Ghartemani, N. Sadati, M. Parniani, Design of fractional
very close to the optimal one. It is also true that the achieved order PID controller for an AVR using particle swarm optimization, Control
maximum sensitivity value is close to the target one. Engineering Practice 17 (2009) 1380–1387.
[9] C.A. Monje, B.M. Vinagre, Y.Q. Chen, V. Feliu, P. Lanusse, J. Sabatier, Proposals for
• The rules proposed for integer-order PID controllers provide a
fractional PI D tuning, Preprints IFAC Workshop on Fractional Differentiation
performance similar to that obtained by the AMIGO tuning rules. and its Applications, Bordeaux (F), 2004, pp. 156–161.
However, the rules proposed in this paper can be applied to PID [10] C.A. Monje, B.M. Vinagre, A.J. Calderon, V. Feliu, Y.Q. Chen, On fractional PI
controllers: some tuning rules for robustness to plant uncertainties, Nonlinear
controller both in series and ideal (non interacting) form. On the Dynamics 38 (2004) 369–381.
contrary, the AMIGO tuning rules might result in complex conju- [11] R.S. Barbosa, J.A. Tenreiro Machado, I.M. Ferreira, Tuning of PID controllers
gate controller zeros and cannot be therefore applied, in general, based on Bode’s ideal transfer function, Nonlinear Dynamics 38 (2004)
305–321.
to PID controllers in series form.
[12] A. O’Dwyer, Handbook of PI and PID Controller Tuning Rules, Imperial College
• Because of the structure of the provided tuning rules (for both Press, 2006.
integer-order and fractional-order PID controllers), the perfor- [13] K.J. Åström, T. Hägglund, Advanced PID Control, ISA Press, Research Triangle
Park (NC), 2006.
mance provided is scaled appropriately with respect to the time
[14] D. Valerio, J. Sa da Costa, Tuning-rules for fractional PID controllers, Preprints
constant T. In other words, with the same normalised dead time, IFAC Workshop on Fractional Differentiation and its Applications, Porto (P),
the value of the integrated absolute error that is obtained for a 2006.
given value of T is equal to T times the value of the integrated [15] D. Valerio, J. Sa da Costa, Tuning of fractional PID controllers with
Ziegler–Nichols-type rules, Signal Processing 86 (2006) 2771–2784.
absolute error that is obtained for T = 1. [16] Y.Q. Chen, T. Bhaskaran, D. Xue, Practical tuning rule development for fractional
• The considered AMIGO tuning rules are the same for both the set- order proportional and integral controllers, ASME Journal of Computational and
point following and the load disturbance rejection task, while two Nonlinear Dynamics 3 (2008) 0214031–0214037.
[17] J.J. Gude, E. Kahoraho, New tuning rules for PI and fractional PI controllers, in:
different tuning rules are used in this paper for the two different Proceedings European Control Conference, Budapest (HU), 2009.
tasks. However, results are still satisfactory if the tuning rules [18] J.J. Gude, E. Kahoraho, Simple tuning rules for fractional PI controllers, in: Pro-
devised for a given task are employed for the other one. ceedings IEEE International Conference on Emerging Technologies and Factory
Automation (ETFA), Mallorca (E), 2009.
[19] F.G. Shinskey, Feedback Controllers for the Process Industries, McGraw-Hill,
5. Conclusions New York, 1994.
[20] K.J. Åström, T. Hägglund, PID Controllers: Theory Design and Tuning, ISA Press,
Research Triangle Park, 1995.
In this paper we have proposed tuning rules for integer-order [21] A. Oustaloup, La Commande CRONE: Commande Robuste d’Ordre Non Entier,
and fractional-order PID controllers. In particular, the tuning rules Hermes, Paris (F), 1991.
allow to minimise the integrated absolute error (for either the [22] A. Visioli, Practical PID Control, Springer, London (UK), 2006.
[23] K.H. Ang, G. Chong, Y. Li, PID control systems analysis, design, and technology,
set-point following or the load disturbance rejection task) subject
IEEE Transactions on Control Systems Technology 13 (2005) 559–576.
to a constraint on the maximum sensitivity. The improvement of [24] A. Visioli, Optimal tuning of PID controllers for integral and unstable processes,
the performance that can be achieved by employing a fractional- IEE Proceedings – Control Theory and Applications 148 (2001) 180–184.
[25] M. Mitchell, An Introduction to Genetic Algorithms, MIT Press, Cambridge (MA),
order PID controller has been specified quantitatively. Indeed, if a
1998.
PI controller is considered, then the use of a fractional-order inte- [26] J.-C. Shen, New tuning method for PID controller, ISA Transactions 42 (2002)
gral action does not provide any improvement in the performance, 473–484.
while for PID controllers, the use a fractional-order derivative [27] The Mathworks, Curve Fitting Toolbox User’s Guide, Natick, MA, 2010.

action is convenient. Finally, analytical expressions of the perfor-


mance index have been provided so that they can be employed
effectively for the purpose of performance assessment.

You might also like