You are on page 1of 12

www.ietdl.

org
Published in IET Control Theory and Applications
Received on 19th December 2010
Revised on 2nd May 2011
doi: 10.1049/iet-cta.2010.0746

ISSN 1751-8644

Note on fractional-order proportional integral


differential controller design
C. Yeroglu1 N. Tan2
1

Computer Engineering Department, Inonu University, Malatya 44280, Turkey


Electrical and Electronics Engineering Department, Inonu University, Malatya 44280, Turkey
E-mail: cyeroglu@inonu.edu.tr

Abstract: This study deals with the design of fractional-order proportional integral differential (PID) controllers. Two design
techniques are presented for tuning the parameters of the controller. The rst method uses the idea of the Ziegler Nichols and the
strom Hagglund methods. In order to achieve required performances, two non-linear equations are derived and solved to obtain
A
the fractional orders of the integral term and the derivative term of the fractional-order PID controller. Then, an optimisation
strategy is applied to obtain new values of the controller parameters, which give improved step response. The second method
is related with the robust fractional-order PID controllers. A design procedure is given using the Bode envelopes of the
control systems with parametric uncertainty. Five non-linear equations are derived using the worst-case values obtained from
the Bode envelopes. Robust fractional-order PID controller is designed from the solution of these equations. Simulation
examples are provided to show the benets of the methods presented.

Introduction

In recent studies, many researchers use fractional calculus in


the control system applications [1 5]. The signicance of
fractional-order representation comes from its nature. The
fractional-order differential equations can describe realworld systems more adequately [6]. Furthermore, the
implementations of fractional-order differential equations
[7, 8] have brought in new horizons in control engineering.
Consequently, many studies have been done for the
fractional-order control systems (FOCSs) [9 14].
Widespread usage of the PID controllers motivated many
researchers to look for better design methods or alternative
controllers [15, 16]. For example, the fractional-order
algorithm for the control of dynamic systems has been
introduced and the performance of CRONE (French
abbreviation for Commande Robuste dOrdre Non Entier),
over the PID controller, has been demonstrated by
Oustaloup [17, 18]. Podlubny has proposed a generalisation
of the PID controller as PIlDm controller. He also
demonstrated that the fractional-order PID (PIlD m)
controller has better response than classical PID controller
[1, 19]. Also, many valuable studies have been done for
fractional-order controllers and their implementations [13,
20 25]. Tuning of the PIlDm controller using the
frequency-domain approaches are studied in many papers.
For example, [26] proposes a method based on optimisation
strategies. Tuning of H1 controllers for fractional singleinput single-output (SISO) system was suggested in [23]. A
new design method for PIa controller is given in [27].
Some tuning rules for robustness to plant uncertainty for
PIl controller are given in [28]. However, in order to
1978
& The Institution of Engineering and Technology 2011

achieve better results, there are still needs for new methods
to obtain the parameters of PIlDm controllers.
On the other hand, it is known that the parameters of
physical systems cannot be expressed precisely and contain
uncertainty due to the tolerance values of elements, nonlinear effects or environmental conditions. Therefore the
parameter uncertainty is inevitable in the systems [5]. In
general, system uncertainties are analysed in two groups
such as parameter uncertainty and model uncertainty. In
control theory, robust control methods have been developed
for the analysis and design of uncertain systems. These
issues are addressed under robust control [29]. The
computation of the frequency responses of uncertain
transfer functions plays an important role in the application
of frequency-domain methods for the analysis and design of
robust control systems [30, 31]. Many studies on the
computation of the frequency responses of the integer order
control systems with parameter uncertainty structure can be
found in the literature [29, 32, 33]. However, in order to
apply classical controller design method to FOCSs with
parameter uncertainty structure, it is necessary to compute
the frequency responses of a given fractional-order interval
transfer function (FOITF). The procedures for the
computation of the Bode and Nyquist envelopes of FOITF
can be found in [6, 34]. The structures of PIlDm controller
have been widely used in recent papers. However, there is
no enough study for parameter uncertainty and robust
control design issues [13, 28]. Therefore the results
obtained in this study provide an important contribution to
this eld.
In this paper, two methods have been proposed for tuning
of the PIlDmcontroller. In the rst method, a tuning technique
IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 19781989
doi: 10.1049/iet-cta.2010.0746

www.ietdl.org
for a PIlDm controller, which is inspired from the classical
strom Hagglund tuning methods, is
Ziegler Nichols and A
introduced. The proposed method uses the classical
Ziegler Nichols tuning rules to obtain the values of kp and
strom Hagglund
ki . The value of kd is obtained using the A
method. In order to achieve specied phase margin, two
non-linear equations have been obtained using the idea of
strom Hagglund tuning method. The values of l and
the A
m are obtained from these two equations. In the second
method, a tuning strategy, which is based on the Bode
envelopes of the FOITF, is introduced for robust PIlDm
controller to control the rst-order and rst-order plus dead
time (FOPDT) systems with the parametric uncertainty
structure. In this method, the Bode envelopes of the system
are successfully combined with ve design criteria, which
Monje Vinagre et al. have used in their papers [13, 21,
28], to obtain new robust PIlDm controller that make the
given plant robust under parameter uncertainties. Thus, the
novelty of the results obtained in this paper is the
development of a new tuning method for PIlDm controller
and the presentation of a new method to design a robust
PIlDm controller. The method presented for the robust
PIlDm controller is an extension of the Monje Vinagre
et al. method, which is given in Section 4.1. The
improvement over those in [13, 21, 28] is that the parameter
uncertainty has been considered. Then an improved robust
method is obtained. Examples are provided to illustrate the
results.
The paper is organised as follows: Mathematical
background of fractional-order representation is given in
Section 2. In Section 3, a tuning method for PIlDm
controller is introduced. Design of robust PIlDm controllers
for rst-order and FOPDT systems with parametric
uncertainty structure is provided in Section 4. Section 5
includes concluding remarks.

Fractional-order Caputo expression can be given as


r
a Dt f (t) =

r
a Dt f

(t) = lim h

[ta/h]


h0

j=0

 
r
f (t jh)
(1)
j
j

= 1,

c(r)
j



1 + r (r)
cj1
= 1
j

(2)

Riemann Liouville denition can be given as


r
a Dt f

1
dn
(t) =
G(n r) dt n

t

f ( t)
dt
rn+1
a (t t)

(3)

where n 2 1 , r , n and G(.) is a Gamma function.


IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 1978 1989
doi: 10.1049/iet-cta.2010.0746

f (n) (t)
dt
rn+1
a (t t)

(4)

1
G(m) =

eu um1 du

(5)

Numerical solutions for Grunwald Letnikov, Riemann


Liouville and Caputo expressions can be obtained using
denitions given in [1, 35 37].
Generally, dynamic behaviours of the systems can be
analysed using transfer function of the control system.
Thus, the Laplace transformations of the integrodifferential expressions for FOCSs are important.
Fortunately, there is not big difference between the Laplace
transformation of the fractional-order expression and that of
the integer order. The most general formula for the Laplace
transformations of the integro-differential expressions can
be given as [37]


 m 
n1
m1k

d f (t)
d
f
(t)
L
= sm L{f (t)}
sk
dt m
dt m1k
k=0

(6)

t=0

where n is an integer number and m satises, n 2 1 , m , n.


The above expression is simplied as follows if all the
derivatives of f (t) are zero
 m 
d f (t)
L
= sm L{f (t)}
dt m

(7)

Consider a SISO control system. Let y(t) be the output and


x(t) be the input of the system. The relation between input
and output of the system can be dened as

an

dan y(t)
dan1 y(t)
da0 y(t)
+ an1
+ + a0
a
a
dt n
dt n1
dt a0
bm
bm1
d x(t)
d
x(t)
db0 x(t)
= bm
+
b
+

+
b
m1
0
dt bm
dt bm1
dt b0

(8)

(1)

 
r
where (1)j
are the binomial coefcients c(r)
j , ( j 0,
j
1, . . .). Following expressions can be used to obtain the
coefcients [36]
c(r)
0

t

where n 2 1 , r , n. The Gamma function G(m) can be


dened for a positive real m as follows [35]

Mathematical background

Fractional calculus can be considered to be generalisation of


integration and differentiation of the integer order
expressions to the non-integer order one. The most
frequently used integro-differential denitions are
GrunwaldLetnikov, Riemann Liouville (RL) and Caputo
expressions. The GrunwaldLetnikov denition of the
fractional-order derivative is given by the following
equation [35]

1
G(r n)

Transfer function of the system can be obtained as follows by


taking the Laplace transform of the above equation [6]
G(s) =

Y (s) bm sbm + bm1 sbm1 + + b0 sb0


=
an san + an1 san1 + + a0 sa0
X (s)

(9)

where an . an21 . . . . . a0 0, bm . bm21 . . . . .


b0 0, ak (k 0, 1, 2, . . . , n) and bl (l 0, 1, 2, . . . , m)
are constants [6, pp. 285 290]. Substituting s jv in the
transfer function of the control system, frequency-domain
analysis of the FOCS can be studied. Since s m ( jv)m, the
expression for ( jv)m can be given as [4]

p
p
(jv)m = vm cos m + j sin m
2
2

(10)

1979

& The Institution of Engineering and Technology 2011

www.ietdl.org
3

Tuning method for PIlDm controller

One can write C( jvcp) using (11) and (14) as

This section presents the development of a tuning method for


the PIlDm controller, which is inspired from the Ziegler
strom Hagglund tuning methods for the
Nichols and A
systems which have integer order transfer functions. The
proposed method uses the Ziegler Nichols tuning rules to
obtain the values of kp and ki . The initial value of kd is
strom Hagglund method. In order to
obtained using the A
achieve specied phase margin, two non-linear equations
have been obtained using critical point information, namely
critical frequency vc and critical gain kc , in the similar
strom Hagglund tuning method. Fine
manner as the A
tuning has been done for kd to achieve the best numerical
solutions of these two equations. The values of l and m are
obtained from these equations using the fsolve
optimisation toolbox of MATLAB. Fine tuning of the
controller parameters may be required to achieve better step
response of the system. In that case, an optimisation model,
which has been developed using Simulink MATLAB, is
used. The controller parameters obtained by the proposed
method are chosen as initial values for optimisation. Then
the new values for the controller parameters are produced
using the optimisation model. Preliminary study of this
section has been presented in the conference [38].
3.1

Computation of PI lD m controller parameters

Consider the negative unity feedback control system shown in


Fig. 1. Transfer function of the plant is an integer order.
However, the controller of the system is a PIlDm controller
of the form
C(s) = kp +

ki
+ kd sm
sl

(11)

The proportional gain constant kp , the constant of integral


term ki , the constant of derivative term kd , the fractionalorder of differentiator l and the fractional order of
integrator m of the controller C(s) can be obtained using the
proposed method.
Let fpm be the required phase margin and vcp be the
frequency of the critical point on the Nyquist curve of G(s)
at which
arg(G(jvcp )) = 1808

(12)

and dene gain margin as


gm =

Considering (14) and (15), one can calculate the following


equations

p
l
l + kd vmcp cos
m
f1 (l, m) = kp + ki v
cp cos
2
2
kc ( cos fpm )
=0

(16)

p
l
l + kd vmcp sin m
f2 (l, m) = ki v
cp sin
2
2
kc ( sin fpm )
=0

(17)

The numerical solutions for l and m can be found using (16)


and (17).
3.2

Tuning method for PIlDm controller

All the parameters of the PIlDm controller, which are given in


(11), can be obtained by using the following procedure:
Specify the value of required phase margin fpm .
Obtain kp and ki from the Ziegler Nichols tuning rules.
Obtain (16) and (17).
strom
Specify the initial value for kd using the A
Hagglund method.
Simulation results show that especially variation in kd
affects the numerical solution of the equations seriously.
Therefore ne tuning can be required for kd to achieve the
best numerical solution for (16) and (17).
Find the numerical solutions for l and m from (16) and
(17), considering new value of kd .
If the step response of the system is not satisfactory
enough, an optimisation can be done by using the
optimisation model to obtain better values for the controller
parameters.
3.3

Application of the proposed tuning method

Example 1: Consider the negative unity feedback system in


Fig. 1. The transfer function of the system is given as
1
= kc
|G(jvcp )|

(13)

Then, in order to make the phase margin of the system equal


to fpm and |C( jvcp)G( jvcp)| 1, the following equation
must be satised
C(jvcp ) =

p
l
m
C(jvcp ) = kp + ki v
cos
l
v
cos
m
+
k
cp
d cp
2

2p

p
l
(15)
+ j ki v
l + kd vmcp sin m
cp sin
2
2

1
e jfpm = kc cos fpm + jkc sin fpm (14)
|G(jvcp )|

G1 (s) =

1
s(s + 3)(s + 4)

The phase crossover frequency and the gain margin


p of the
system can be obtained, respectively, as vcp 12 and
kc 84. Constants of proportional, integral and derivative
terms of the controller are obtained by using the Ziegler
Nichols rules as kp 50.40, ki 55.60 and kd 10.96. Let
PID controller obtained from Ziegler Nichols method be
C1ZN(s) which is dened as follows
C1ZN (s) = 50.40 +

Fig. 1 Negative unity feedback system


1980
& The Institution of Engineering and Technology 2011

(18)

55.60
+ 10.96 s
s

(19)

strom Hagglund method, the values of PID


Using the A
controller parameters have been calculated for specied
IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 19781989
doi: 10.1049/iet-cta.2010.0746

www.ietdl.org
phase margins which are shown in Table 1. Let C1AH(s)
strom
shows the PID controller obtained from the A
Hagglund method for fpm 308, which can be written
from Table 1 as
72.7461
C1AH (s) = 72.7461 +
+ 18.1865 s
s

(20)

The proposed method takes the values of kp and ki from


Ziegler Nichols method. The initial values for derivative
strom Hagglund
term kd have been found by using the A
method for the specied phase margins. Fine tuning has
been done for the term kd to achieve the best numerical
solution of (16) and (17) for each specied phase margin.
These two equations have been solved by using
optimisation toolbox fsolve of the MATLAB to obtain
numerical values of l and m by considering the new value
of kd for each specied phase margin. Table 2 shows all the
values of kp , ki , kd , l and m obtained by the proposed
method for each of the specied phase margin. Let C1(s)
shows the desired PIlDm controller designed by the
proposed method, which can be written from Table 2 for
fpm 308 as
C1 (s) = 50.4 +

55.6
+ 22 s0.8564
s0.7569

(21)

In order to obtain better step response, an optimisation model


has been developed using Simulink by considering least
square method for optimisation. This optimisation model
has been used to obtain new optimised values for the
parameters kp , ki , kd , l and m. Table 3 shows the
optimisation results for different initial values, which are
taken from Table 2. Let C1opt (s) shows the PIlDm controller
with optimised values, which can be written as
C1 opt (s) = 42.4580 +

87.2733
+ 53.1352s0.9623
s0.5030

(22)

Consequently, four types of controllers, namely C1ZN(s),


Table 1

Values of PID controller parameters calculated by


AstromHagglund method for fpm 308, 408, 508 and 608
Astrom-Hagglund

fpm

308
408
508
608

kp

ki

kd

72.7461
64.3477
53.9942
42.0000

72.7461
51.9716
34.0387
19.4923

18.1865
19.9177
21.4122
22.6244

Optimised values of PIlDm controller parameters where


initial values are taken from Table 2 for fpm 308, 408, 508 and 608

Table 3

Initial values
form Table 2
for fpm 308
for fpm 408
for fpm 508
for fpm 608

Proposed method with optimised values


kp

ki

kd

42.4580
41.2422
41.0156
41.7490

87.2733
89.5344
95.4288
88.8996

53.1352
54.6636
55.3521
53.8169

0.5030
0.5003
0.4975
0.4975

0.9623
0.9570
0.9639
0.9623

C1AH(s), C1(s) and C1opt (s), have been designed for the
given plant as follows:
Parameters of C1ZN(s) are calculated using the Ziegler
Nichols method, such as kp 50.40, ki 55.60 and
kd 10.96 for all specied phase margin.
strom
Parameters of C1AH(s) are calculated using the A
Hagglund method, for specied phase margins as shown in
Table 1.
Parameters of C1(s) are obtained using the proposed
method for the specied phase margins as shown in Table 2.
Optimisation model has been used to obtain better step
response for the controller C1(s). The controller C1opt (s) is
obtained with the new optimised values of kp , ki , kd , l and
m as given in Table 3.
It is known that the step response of a system gives
valuable information, such as maximum overshoot, rise
time, peak time and settling time. The step responses of the
system for the controllers C1ZN(s), C1AH(s), C1(s) and
C1opt (s) in (19) (22) are obtained using the nintblocks of
MATLAB, which is developed by Duarte Valerio [39] as
shown in Fig. 2. The performance specications for these
controllers are given in Table 4. One can conclude from
Fig. 2 and Table 4 that the performance specications of
the proposed method are much better than the Ziegler
strom Hagglund tuning methods.
Nichols and A
Bode plots and Nyquist plots of the system for the C1(s) for
phase margins fpm 308, 408, 508, 608 and Nyquist plot of
the system for C1opt (s) are given in Figs. 3 5, respectively.
Looking at the Figs. 3 and 4, it can be observed that the
system satises each of the specied phase margin for
C1(s). Fig. 5 shows that the values of the gain and phase
margins of the system for C1opt (s) are suitable.

Values of PIlDm controller parameters calculated by


proposed method for fpm 308, 408, 508 and 608

Table 2

Proposed method

fpm

308
408
508
608

kp

ki

kd

50.4
50.4
50.4
50.4

55.6
55.6
55.6
55.6

22.000
24.000
24.050
23.935

0.7569
0.9714
0.9762
0.9208

0.8564
0.8823
0.9766
1.0744

IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 1978 1989
doi: 10.1049/iet-cta.2010.0746

Fig. 2 Step responses of the system for C1ZN(s), C1AH(s), C1(s) and
C1opt (s)
1981

& The Institution of Engineering and Technology 2011

www.ietdl.org
Table 4

Step response specifications of C1ZN(s), C1AH(s) and C1opt(s)

Step response
specifications
max. overshoot, %
peak time, s
rise time, s
settling time (%5)
settling time (%2)

Ziegler Nichols
PID

AstromHagglund
PID

Proposed
fractional PID

Proposed fractional
PID with optimised values

73.5
2.35
1.73
6.02
7.30

59.0
2.10
1.65
3.20
4.25

52.5
1.83
1.47
3.75
3.98

31.5
1.47
1.31
2.60
2.83

Fig. 3 Bode plots of the system controlled with C1(s) for the phase
margins fpm 308, 408, 508 and 608
Fig. 5 Nyquist plot of the system for C1opt (s)

with the parameter uncertainty structure. Bode envelopes of


the rst-order and FOPDT systems with parameter
uncertainty structure are successfully combined with ve
design criteria to obtain the robust PIlDm controller.
As known, FOPDT systems provide simple
characterisation of a process and give valuable information
about dynamics of many applications in process control
industry. Since the plants are commonly modelled with
FOPDT transfer functions in the process industry, most of
the engineers are familiar with the parameter of FOPDT
model [40]. A FOPDT system can be represented
mathematically as follows
G(s) =

Fig. 4 Nyquist plots of the system controlled with C1(s) for the
phase margins fpm 308, 408, 508 and 608

Design of robust PIlDm controller

This section presents the development of a tuning method for


robust PIlDm controller for the rst-order and FOPDT
systems with parameter uncertainty structure. All parameters
of the PIlDm controller are calculated to satisfy robust
performances of the plant. Five unknown parameters of the
PIlDm controller are estimated solving ve non-linear
equations that satisfy ve design criteria. The proposed
method is an extension of Monje Vinagre et al. tuning
method [13, 21, 28] for the rst-order and FOPDT systems
1982
& The Institution of Engineering and Technology 2011

k
eLs
ts + a

(23)

where k is the steady-state gain, L represents the process delay


time, t . 0 is the time constant. Sign and magnitude of a
determines the open-loop stability and steady-state gain of
the process, respectively. In (23), parameters of the plant k,
t and L might be uncertain parameters of the system. Exact
values of these parameters may not be known. But these
parameters can be estimated at certain intervals. Therefore
modelling of this system as an interval time delay system is
a realistic approach. Parameters of FOPDT system with
parametric uncertainty structure can be dened as
k [ [k, k],

t [ [t, t],

L [ [L, L]

(24)

where k, t and L are lower limits, k, t and L are upper


limits of the parameters, respectively. The fractional-order
controller is designed to obtain the desired performances for
IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 19781989
doi: 10.1049/iet-cta.2010.0746

www.ietdl.org
the given interval system. Preliminary study of this section
has been presented in the conference [41].
4.1

for the following plant


Gp (s) =

Monje Vinagre et al. method

Owing to the tuning method, which Monje, Vinagre and their


colleagues have used in their papers [13, 21, 28], we preferred
to use Monje Vinagre et al. method as a title of this section.
Monje Vinagre et al. proposed a PIlDm controller tuning
algorithm for the system to satisfy ve design criteria, such
as magnitude at gain crossover frequency, phase margin,
robustness to plant uncertainties, high-frequency noise
attenuation and sensitivity functions. Five design criteria of
the Monje Vinagre et al. method are given as follows.
4.1.1 Phase margin and gain crossover frequency:
The gain and phase margins are two important frequencydomain specications and two important measures of
robustness. The phase margin is related to the damping of
the system. Thus, the following equations should be satised
|C(jvcg )G(jvcg )|dB = 0 dB

(25)

(Arg(C(jvcg )G(jvcg )) = p + fpm

(26)

where vcg is the gain crossover frequency and fpm is the


required phase margin.
4.1.2 Robustness to variation in the gain of the
plant: Satisfying the following constraint [42]


d(Arg(C(jv)G(jv)))
=0
dv
v=vcg

(27)

the phase is forced to be at at vcg and the phase plot is almost


constant within the interval around vcg . Consequently, the
phase plot around the specied frequency vcg is locally at,
which implies that the system will be more robust to gain
variation and overshoots of the step responses are almost
constant within the interval.
4.1.3 High-frequency noise rejection: To satisfy the
robustness due to high-frequency noise, the following
condition must be fullled




T (jv) = C(jv) G(jv)  A dB

1 + C(jv) G(jv)

(28)

k
ts+ 1

(30)

Parameters of the plant Gp(s) are taken as t 80 and


1 k 5. Design specications for the plant are given as
follows: gain crossover frequency vcg 0.675 rad/s, phase
margin fpm 608, robustness to variation in the gain of the
plant must be fullled, desired value of the noise
attenuation
is
A 220 dB
for
the
frequency
v vt 10 rad/s and desired value of sensitivity function
is B 220 dB for the frequency v v2 0.01 rad/s. Five
non-linear equations are obtained satisfying all of these
specications. These ve non-linear equations are solved
using fmincon optimisation toolbox of MATLAB, to
obtain all the parameters of the PIlDm controller as follows
CMV (s) = 4.8694 +

43.0382
+ 0.0637s0.9899
s0.3776

(31)

Step responses, Bode plot, magnitude plot of T(s) and S(s) of


the system CMV(s) Gp(s) are given in Figs. 6 9, respectively.
The step responses in Fig. 6 demonstrate that the stability of
the system CMV(s)Gp(s) is satised and overshoots of the step
responses are almost constant for the values of 1 k 5. In
Fig. 7, phase plot shows that the system is robust to gain
variation within the interval around the gain crossover
frequency v vcg . Figs. 8 and 9 show, respectively, that
the system CMV(s)Gp(s) satises the desired noise
attenuation and the desired value of sensitivity function at
the specied frequencies.
4.2

Design of robust PIlDm controller

This section proposes a design procedure for robust PIlDm


controller to control rst-order and FOPDT system with
parametric uncertainty structure. The FOPDT system with
parametric uncertainty can be represented as
G(s) =

[k, k]
e[L,L]s
[t, t]s + 1

(32)

The proposed method uses the Bode envelopes of the FOPDT


plant given in Fig. 10, to satisfy robust performance of the
system. Consider the Bode envelopes given in Fig. 10.
Minimum and maximum plots of the gain are obtained by
the following transfer functions, respectively

dB

where A is the desired value of the noise attenuation for the


frequency v vt rad/s.
4.1.4 Good output disturbance rejection: To ensure a
good output disturbance rejection, the following constraint
must be satised




1
S(jv) =
 B dB

1 + C(jv) G(jv)

(29)

GR1 (s) =

k
eLs
ts + 1

and

GR2 (s) =

k
eLs
ts + 1

(33)

Time delay L does not have any effect on the gain plot of the
plant. Similarly, minimum and maximum plots of phase are
obtained by the following transfer functions, respectively.
GR3 (s) =

k
eLs
ts + 1

and

GR4 (s) =

k
eLs
ts + 1

(34)

dB

where B is the desired value of sensitivity function for the


frequency v vs rad/s.
Application of Monje Vinagre et al. method: An
application of the Monje Vinagre et al. method is given
IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 1978 1989
doi: 10.1049/iet-cta.2010.0746

Steady-state gain k does not have any effect on phase plot of


the plant. In order to design robust PIlDm controller, (25)
should be satised with the transfer function GR2(s), namely
vcg must be taken at the point a in Fig. 10. Equation (26)
should be satised with the transfer function GR4(s), namely
1983

& The Institution of Engineering and Technology 2011

www.ietdl.org

Fig. 6 Step response of CMV(s)Gp(s) for 1 k 5

Fig. 9 Magnitude of S(s) for CMV(s)Gp(s)

(25) to (29) as follows

Fig. 7 Bode plots of CMV(s)Gp(s)

|C(jvcg )GR2 (jvcg )|dB = 0 dB

(35)

(Arg(C(jvcg )GR4 (jvcg )) = p + fm

(36)

Fig. 8 Magnitude of T(s) for CMV(s)Gp(s)

the point b in Fig. 10 should be the minimum phase margin


of the system. Similarly, (27) should be satised with the
transfer function GR4(s) for minimum phase margin. GR1(s)
and GR2(s) minimise (28) and (29), respectively, to satisfy
the specied conditions. According to the explanations
given above, the following equations can be obtained from
1984
& The Institution of Engineering and Technology 2011

Fig. 10 Bode envelopes of a FOPDT plant


IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 19781989
doi: 10.1049/iet-cta.2010.0746

www.ietdl.org


d(Arg(C(jv)GR4 (jv)))
=0
dv
v=vcg

(37)





T (jv) = C(jv) GR1 (jv)  A dB

1 + C(jv) G (jv)

(38)

R1

dB





1
S(jv) =
 B dB

1 + C(jv) GR2 (jv)dB

(39)

In order to satisfy the given specications for the system and


full ve design criteria, the following equations can be
derived with ve unknown parameters (kp , ki , kd , l and m),
using (33) and (34) in (35) (39)






k



(r)2 + (s)2  = 0 dB



2
((t vcg ) + 1)


a tan

s
r

(40)

(41)

1
(su r s ru)
t

L = 0 (42)
2
2
1 + (s/r)
(r)
1 + (t vcg )2







2
2
k (rt) + (st)


 20 dB (43)


(1 + k rt)2 + (t vt + k st)2 

dB






2


(t v s ) + 1


 20 dB



(1 + k rs)2 + (t vs + k ss)2 

4.3 Application of the proposed method to design


robust PIlDm controller
Example 2: Consider the following rst-order interval plant
G2 (s) =

dB

a tan (t vcg ) vcg L = p + fpm

Use ve design criteria, which are given in Section 2,


with the transfer functions given in (33) and (34) and
obtain (35) (39).
Derive the (40) (44) using the transfer functions given in
(35) (39) with the desired design specication.
Solve (40) (44) for ve unknown parameters (kp , ki , kd , l
and m) using fmincon optimisation toolbox of MATLAB
and obtain robust controller C(s).

(44)

dB

where

p
l
r = kp + ki v
l + kd vmcg cos m
cg cos
2
2

p
l
l + kd vmcg sin m
s = ki v
cg sin
2
2

p
l1
cos l + kd mvmcg1 cos m
ru = ki lvcg
2
2

p
l1
m1
su = ki lvcg sin l + kd mvcg sin m
2
2

p
l
l + kd vmt cos m
rt = kp + ki v
t cos
2
2

p
l
l + kd vmt sin m
st = ki v
t sin
2
2

p
l
m
rs = kp + ki vs cos l + kd vs cos m
2
2

p
l
l + kd vms sin m
ss = ki v
s sin
2
2
Equations (40) (44) with ve unknown parameters (kp , ki ,
kd , l and m) can be solved to obtain the parameters of
PIlDm for the robust stability of the given plant.
The following procedure can be applied to design robust
PIlDm controller for FOPDT system:
Specify the plant with parametric uncertainty in (23).
Find the transfer functions, which will give the gain and
phase extremums of the Bode envelopes.
Obtain Bode envelopes of system using (33) and (34).
IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 1978 1989
doi: 10.1049/iet-cta.2010.0746

[k, k]
[t, t]s + 1

(45)

where k [ [k, k] = [2, 4] and t [ [t, t] = [60, 80]. In order


to calculate the Bode envelopes, four different transfer
functions are obtained using gain and phase extremums of
the plant given in (45) as follows
G21 (s) =

2
80 s + 1

and

G22 (s) =

4
60 s + 1

(46)

G23 (s) =

4
60 s + 1

and

G24 (s) =

2
80 s + 1

(47)

Design specications for the system are given as follows: gain


crossover frequency vcg 2.1 rad/s, phase margin
fpm 808, robustness to variation in the gain of the plant
must be fullled, desired noise attenuation A 220 dB for
the frequency v vt 10 rad/s and desired value of
sensitivity function B 220 dB for the frequency
v vs 0.01 rad/s. In order to satisfy the given
specications for the system and full ve design criteria,
(46) and (47) are used in (40) (44). FMINCON
optimisation toolbox of MATLAB is used for solution of
ve non-linear equations. Equation (40) is taken as a main
function and the other equations are taken as non-linear
constraints for optimisation. Values of the ve unknown
parameters (kp , ki , kd , l and m) are calculated. Then, PIlDm
controller to control G2(s) is obtained as
C2 (s) = 6.7441 +

27.7461
+ 0.0063s0.7874
s0.1461

(48)

Step responses of C2(s)G2(s) for 25 different values of the


parameters of G2(s) are obtained by using the nintblocks
of MATLAB [39], as shown in Fig. 11. The step responses
of the system C2(s)G2(s) show that the system is more
robust to gain changes and overshoot of the step responses
is almost constant within the interval of uncertain parameters.
Bode plots, magnitude plots of T2(s) and S2(s) of the
system C2(s)G2(s) for 100 different values of the uncertain
parameters by taking 10-point for each interval of the
uncertain parameters of the G2(s) are obtained, respectively,
as in Figs. 12 14 using the toolbox developed in
MATLAB [43]. Fig. 12 shows that the phase envelopes of
the system are almost at and almost constant within an
interval around vcg 2.1 rad/s with the specied
constraints. Magnitude plots of the high-frequency noise
attenuation function and sensitivity function show,
respectively, that the system satises desired noise
attenuation and the desired value of sensitivity in Figs. 13
and 14. One can conclude from the results of Figs. 11 14
1985

& The Institution of Engineering and Technology 2011

www.ietdl.org
that the controller satises the robust performance of the
system.

Example 3: Now, consider the plant in Example 2 with time


delay as follows

G3 (s) =

Fig. 11 Step responses of the C2(s)G2(s)

[k, k]
e[L,L]s
[t, t]s + 1

(49)

where [k, k] = [2, 4], [t, t] = [60, 80] and [L, L] =


[0.5, 1]. Gain crossover frequency is vcg 0.16 rad/s. The
gain and phase envelopes can be obtained from the
following equations, respectively
G31 (s) =

2
e1s
80 s + 1

G33 (s) =

and

4
e0.5s
60 s + 1

G32 (s) =

and

4
e1s (50)
60 s + 1

G34 (s) =

2
e1s
80 s + 1
(51)

Controller C3(s) can be obtained using the same procedure in


Example 2 as follows
C3 (s) = 2.8664 +

Fig. 12 Bode plots of C2(s)G2(s)

0.0956
+ 1.2878 s0.9721
s0.8483

(52)

The overshoots of the step responses of the system


C3(s)G3(s), which are obtained for 125 different transfer
functions by taking 5-point from each uncertain parameters
of the plant C3(s), are almost the same as given in Fig. 15.
Bode plots, magnitude of T3(s) and magnitude of S3(s) of
the system C3(s)G3(s) are obtained in Figs. 16 18,
respectively, for 125 different values of the uncertain
parameters of the G3(s). The phase envelopes of the system
are almost at and almost constant within an interval
around vcg 0.16 rad/s with the specied constraints.
Magnitude plots of the high-frequency noise attenuation
function and sensitivity function show that the system
satises the desired noise attenuation and the desired value
of sensitivity.

Fig. 13 Magnitudes of T2(s) for C2(s)G2(s)

Fig. 14 Magnitudes of S2(s) for C2(s)G2(s)


1986
& The Institution of Engineering and Technology 2011

Fig. 15 Step responses of the system controlled with C3(s)


IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 19781989
doi: 10.1049/iet-cta.2010.0746

www.ietdl.org
phase extremums of the plant given in (53) as follows
G41 (s) =

2.7
e55s ,
443 s + 1

G42 (s) =

3.5
e55s (54)
423 s + 1

G43 (s) =

3.5
e45s ,
423 s + 1

G44 (s) =

2.7
e55s (55)
443 s + 1

Design specications for the system are given as follows: gain


crossover frequency vcg 0.008 rad/s, phase margin
fpm 608, robustness to variation in the gain of the plant
must be fullled, desired noise attenuation A 220 dB for
the frequency v vt 1 rad/s and desired value of
sensitivity function B 220 dB for the frequency
v vs 0.001 rad/s. The controller C4(s) to control G4(s)
is calculated as in Example 2 as follows
C4 (s) = 0.5775 +

Fig. 16 Bode plots of C3(s)G3(s)

0.0047
+ 4.3865s0.5253
s0.9733

(56)

Step responses of the system controlled with C4(s) for 125


different values of the parameters of the G4(s) are obtained
as in Fig. 19. Overshoots of the step responses are almost
constant within the interval of uncertain parameters and the

Fig. 17 Magnitudes of T3(s) for C3(s)G3(s)

Fig. 19 Step responses of the system controlled with C4(s)


Final value of the reference input step 0.47 as in [13]

Fig. 18 Magnitudes of S3(s) for C3(s)G3(s)

Example 4: Consider the following real liquid-level system


given in [13]

G4 (s) =

k
3.13
eLs =
e50s
ts+ 1
433.33 s + 1

(53)

Interval representation of this plant is a realistic approach. The


parameters of this real plant can be considered as
[k, k] = [2.7, 3.5], [t, t] = [423, 443]
and
[L, L] =
[45, 55], which are interval representations of the
parameters. In order to calculate the Bode envelopes, four
different transfer functions are obtained using gain and
IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 1978 1989
doi: 10.1049/iet-cta.2010.0746

Fig. 20 Bode plots of C4(s)G4(s)


1987

& The Institution of Engineering and Technology 2011

www.ietdl.org

Fig. 21 Magnitudes of T4(s) for C4(s)G4(s)

these non-linear equations. Gain and phase extremums of


Bode envelopes of the plant is used to satisfy robust
performances of the system. As seen from the results of the
numerical example, the PIlDm controller satises robust
performance of the system.
It is necessary to point out that there are many other tuning
methods in the literature for PID controllers, which may give
strom
better results than the Ziegler Nichols and A
Hagglund methods for some cases. Some tuning methods
for PIlDm controller are also proposed in recent years. The
comparison studies of the proposed methods for tuning of
PIlDm controllers certainly will be very important. On the
other hand, the robust PIlDm controller design for different
types of plants with parameter uncertainty structure might
be a subject of future work.

Fig. 22 Magnitudes of S4(s) for C4(s)G4(s)

controller satises the robust stability of the FOPDT system,


which is more robust to gain changes. Bode plots, magnitude
of T4(s) and magnitude of S4(s) of the system C4(s)G4(s) for
125 different values of the uncertain parameters of the
G4(s) are shown, respectively, in Figs. 20 22. The Bode
plots of the system shows that the phase of the system are
almost at and almost constant within an interval around
vcg 0.008 rad/s with the specied constraints. Magnitude
plots of the high-frequency noise attenuation function and
sensitivity function show that the system satises desired
noise attenuation for v vt 1 rad/s and the desired value
of sensitivity for v v2 0.001 rad/s. One can conclude
from Figs. 19 22 that the controller satises the robust
performance of the system.

Conclusions

In this paper, two methods for tuning of PIlDm controller have


been proposed. The rst method is based on the idea of using
strom Hagglund method
the Ziegler Nichols and A
together. Values of the kp and ki parameters of PIlDm
controller have been computed from the Ziegler Nichols
method and the remaining parameters kd , l and m have
strom Hagglund method using
been found from the A
critical point information. Values of the controller
parameters are optimised to achieve better step response.
The simulation results demonstrated that the PIlDm
controller has better response than the classical PID
controllers. In the second method, we proposed a new
robust tuning method for a PIlDm controller to control rstorder systems with parameter uncertainty structure. The
proposed method benets from design specications given
in Monje Vinagre et al. tuning method. Five design
specications of the Monje Vinagre et al. method are used
to derive ve non-linear equations. Values of the ve
unknown parameters of PIlDm controller are obtained from
1988
& The Institution of Engineering and Technology 2011

References

1 Podlubny, I.: Fractional-order systems and PIlD m controllers, IEEE


Trans. Autom. Control, 1999, 44, (1), pp. 208 214
2 Sabatier, J., Poullain, S., Latteux, P., Thomas, J.L., Oustaloup, A.:
Robust speed control of a low damped electromechanical system
based on CRONE control: application to a four mass experimental test
bench, Nonlinear Dyn., 2004, 38, pp. 383 400
3 Chen, Y.Q., Ahn, H.S., Podlubny, I.: Robust stability check of
fractional order linear time invariant systems with interval
uncertainties, Signal Process., 2006, 86, pp. 26112618
zguven, O
.F., O
zyetkin, M.M.: Robust stability analysis of
4 Tan, N., O
fractional order interval polynomials, ISA Trans., 2009, 48,
pp. 166172
5 Nataraj, P.S.V.: Computation of spectral sets for uncertain linear
fractional-order systems using interval constraints propagation. Third
IFAC Workshop on Fractional Differentiation and its Application,
Ankara, 2008
6 Xue, D., Chen, Y.Q., Atherton, D.P.: Linear feedback control analysis
and design with MATLAB (SIAM Press, 2007)
7 Nakagava, N., Sorimachi, K.: Basic characteristics of a fractance
device, IEICE Trans. Fundam., 1992, E75-A, 12, pp. 1814 1818
8 Hwang, C., Cheng, Y.C.: A numerical algorithm for stability testing of
fractional delay systems, Automatica, 2006, 42, pp. 825 831
9 Petras, I.: The fractional order controllers: methods for their synthesis
and application, J. Electr. Eng., 1999, 50, pp. 284 288
10 Manabe, S.: Early development of fractional order control. Proc.
ASME 2003 Design Engineering Technical Conf., Chicago, Illinois,
2003
11 Valerio, D., Da Costa, J.S.: Time domain implementation of fractional
order controllers, IEE Proc. Control Theory Appl., 2005, 5,
pp. 539552
12 Machado, J.: Discrete-time fractional-order controllers, Fract. Calc.
Appl. Anal., 2001, 4, (1), pp. 4766
13 Monje, C.A., Vinagre, B.M., Feliu, V., et al.: Tuning and auto-tuning of
fractional order controllers for industry applications, Control Eng.
Pract., 2008, 16, pp. 798 812
14 Chen, Y.Q., Moore, K.L.: Discretization schemes for fractional order
differentiators and integrators, IEEE Trans. Circuits Syst. I: Fundam.
Theory Appl., 2002, 49, (3), pp. 363 367
15 Ogata, K.: Modern control engineering (Prentice-Hall, New Jersey,
2002)
strom, K., Hagglund, T.: PID controllers: theory, design and tuning
16 A
(Instrument Society of America, North Carolina, 1995)
17 Oustaloup, A.: LaDerivation Non Entiere (HERMES, Paris, 1995)
18 Oustaloup, A., Levron, F., Mathieu, B., Nanot, F.M.: Frequency-band
complex noninteger differentiator: characterization and synthesis,
IEEE Trans. Circuits Syst. I: Fundam. Theory Appl., 2000, 47, (1),
pp. 2539
19 Podlubny, I.: Fractional differential equations (Academic Press, San
Diego, California, 1999)
20 Bettou, K., Charef, A., Mesquine, F.: A new design method for
fractional PIlD m controller, IJ-STA, 2008, 2, pp. 414 429
21 Vinagre, B.M., Monje, C.A., Calderon, A.J., Suarez, J.I.: Fractional
PID controllers for industry application, J. Vibr. Control, 2007, 13,
(9 10), pp. 14191429
22 Vinagre, B.M., Podlubny, I., Dorcak, L., Feliu, V.: On fractional PID
controllers: a frequency domain approach. IFAC Workshop on
Digital Control. Past, Present and Future of PID Control, Terrasa,
Spain, 2000, pp. 5358
IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 19781989
doi: 10.1049/iet-cta.2010.0746

www.ietdl.org
23 Petras, I., Hypiusova, M.: Design of fractional-order controllers
via H1 norm minimization, Sel. Top. Model. Control, 2002, 3,
pp. 5054
24 Xue, D., Chen, Y.Q.: A comparative introduction of four fractional
order controllers. Proc. Fourth World Congress, Intelligent Control
and Auto, 2002, vol. 4, pp. 32283235
25 Hamamci, S.E.: An algorithm for stabilization of fractional-order time
delay systems using fractional-order PID controllers, IEEE Trans.
Autom. Control, 2007, 52, pp. 1964 1969
26 Barbosa, R.S., Machado, J.A.T., Ferreira, I.M.: Tuning of PID
controllers based Bodes ideal transfer function, Nonlinear Dyn.,
2004, 38, pp. 305321
27 Maione, G., Lino, P.: New tuning rules for fractional PIa controllers,
Nonlinear Dyn., 2007, 49, (12), pp. 251257
28 Monje, C.A., Calderon, A.J., Vinagre, B.M., Chen, Y.Q., Feliu, V.: On
fractional PIl controllers: some tuning rules for robustness to plant
uncertainties, Nonlinear Dyn., 2004, 38, pp. 369 381
29 Bhattacharyya, S.P., Chapellat, H., Keel, L.H.: Robust control: the
parametric approach (Prentice-Hall, 1995)
30 Kharitonov, V.L.: Asymptotic stability of an equilibrium position of a
family of systems of linear differential equations, Diff. Eqns., 1979,
14, pp. 14831485
31 Bartlett, A.C., Hollot, C.V., Lin, H.: Root location of an entire polytope
of polynomials: it sufces to check the edges, Math. Controls, Signals
Syst., 1988, 1, pp. 6171
32 Hollot, C.V., Tempo, R.: On the Nyquist envelope of an interval plant
family, IEEE Trans. Autom. Control, 1994, 39, pp. 391396
33 Tan, N., Atherton, D.P.: Frequency response of uncertain systems:
a 2q-convex parpolygonal approach, IEE Proc. Control Theory Appl.,
2000, 147, (5), pp. 547 555

IET Control Theory Appl., 2011, Vol. 5, Iss. 17, pp. 1978 1989
doi: 10.1049/iet-cta.2010.0746

zyetkin, M.M.: Frequency response


34 Yeroglu, C., Tan, N., O
computation of fractional order interval transfer functions,
Int. J. Control Autom. Syst., 2010, 8, (5), pp. 1009 1017
35 Caponetto, R., Dongola, G., Fortuna, L., Petras, I.: Fractional order
systems, modeling and control applications, world scientic series on
nonlinear science, series A (World Scientic Publishing Co. Pvt.
Ltd., 2010, vol. 72)
36 Dorcak, L.: Numerical models for simulation the fractional-order
control systems, UEF-04-94 (The Academy of Sciences, Institute of
Experimental Physics, Kosice, Slovakia, 1994)
37 Oldham, K.B., Spanier, J.: The fractional calculus: theory and
applications of differentiation and integration to arbitrary order
(Dover Books on Mathematics, 2006)
38 Yeroglu, C., Onat, C., Tan, N.: A new tuning method for PIlDm
controller. ELECO 2009 Sixth Int. Conf. Electrical and Electronics
Engineering, Bursa, Turkey, 2009
39 Valerio, D.: Ninteger v. 2.3 fractional control toolbox for MatLab.
http://web.ist.utl.pt/~duarte.valerio, 2005
40 Roy, A., Iqbal, K.: PID controller tuning for the rst-order-plus-deadtime process model via Hermite-Biehler theorem, ISA Trans., 2004,
44, pp. 363 378
41 Yeroglu, C., Ozyetkin, M.M., Tan, N.: Design of robust PIlDm
controller for FOPDT systems. Fourth IFAC Workshop on Fractional
Differentiation and its Applications, Badajoz, Spain, 2010
42 Chen, Y.Q., Hu, C.H., Moore, K.L.: Relay feedback tuning of robust
PID controllers with iso-damping property. Proc. 42nd IEEE Conf.
on Decision and Control, Maui, Hawaii, USA, 2003
43 Yeroglu, C., Tan, N.: Development of a toolbox for frequency response
analysis of fractional order control systems. 19th European Conf. on
Circuit Theory and Design, Antalya, August 2009

1989

& The Institution of Engineering and Technology 2011

You might also like