Professional Documents
Culture Documents
Yumuk2018 PDF
Yumuk2018 PDF
Abstract
In this study, we deal with systems that can be represented by single fractional order pole models and propose an integer
order proportional–integral/proportional–integral–derivative controller design methodology for this class. The basic
principle or backbone of the design methodology of the proposed controller relies on using the inverse of the fractional
model and then approximating this fractional controller transfer function by a low integer order model using Oustaloup
filter. The emerging integer order controller reveals itself either in pre-filtered proportional–integral or proportional–
integral–derivative form by emphasizing on the dominancy concept of pole-zero configuration. Parameters of the pro-
posed controllers depend on the parameters of the single fractional order pole model and the only free design para-
meter left is the overall controller gain. This free design parameter is determined via some approximating functions
relying on an optimization procedure. Simulation results show that the proposed controller exhibits either satisfactory
or better results with respect to some performance indices and time domain criteria when they are compared to classi-
cal integer order proportional–integral–derivative and fractional order proportional–integral–derivative controllers.
Moreover, the proposed controller is applied to real-time liquid level control system. The application results show that
the proposed controller outperforms the other controllers.
Keywords
Fractional order model, fractional order proportional–integral–derivative, pre-filtered integer order proportional–inte-
gral/proportional–integral controller, inverse control, liquid level control system
for fractional order dynamic control systems. The frac- keeping in mind that this is still an approximation. In
tional calculus is used for modeling and control of sev- this respect, proposing an integer low order controller
eral systems. Modeling and/or control of heat flow which preserves the dominant dynamic characteristics
platform,8,9 hexapod robot,10 liquid level system,11 of these high order approximations will be advanta-
servo-system,11–13 twin rotor helicopter,14 power elec- geous in design and implementation. An integer order
tronic buck converters15 and so on are investigated in PID controller and reduced integer order inverse con-
various studies. troller design for a fractional order system are pre-
In control area, fractional calculus provides three sented in Yumuk et al.,23,24 respectively.
different combinations in terms of the controller and In this study, we propose integer order pre-filtered
system types: fractional order control for integer order PI/PID controllers for a certain class of fractional order
system,9,16–19 fractional order control for fractional systems which is composed of a single fractional order
order system8,20–22 and integer order control for frac- pole. The single fractional order pole models (SFPM)
tional order system.23,24 In this respect, fractional cal- are generalized form of the first integer order models.
culus is an excellent tool to describe a control system In the design method, first the inverse of the fractional
compared to intr order calculus. model is taken into consideration for the proposed con-
There are various controller design strategies for trollers. The inverse fractional order controller is then
fractional order controllers in the literature. These represented by the second order Oustaloup filter
mostly rely on and end up with integer order control approximation. The integer order PI/PID controllers
structures. Some of related studies in this respect can be with an integer first order filter is finally obtained by
stated as fractional order composite model reference emphasizing the dominant pole-zero concept. It is a
adaptive control,25,26 fractional order sliding mode con- well-known fact that integer PID type controllers are
trol27,28 and fuzzy fractional order sliding mode con- still widely used in industrial implementations. Since
troller.29 The most widely used fractional controllers the evolving controller is in an integer PI/PID control-
are fractional order PID controllers. The tuning meth- ler form, it can easily be used in real-time applications.
ods for fractional order PID controller can be divided The parameters of the proposed controllers depend on
into three categories: analytical methods,8,20,21 numeri- the parameters of the single fractional order pole model
cal methods18,19,22 and rule-based methods.16,17 Finally, and the only free design parameter left is the overall
recent developments for design and implementation of controller gain. This free design parameter is deter-
fractional PID can be found in the survey.30 mined via some approximating functions based on an
The stability analysis in fractional calculus is an optimization procedure. Simulations are done to com-
important issue as it is in integer calculus. A stability pare the performances of proposed controllers with
analysis for integer order system is generalized to frac- classical integer order PID and fractional order PID
tional commensurate and incommensurate order sys- controllers. The results exhibit that the proposed con-
tem using state space representation.31,32 There are also troller provides either a comparable or satisfactory per-
some studies invoking the effect of dead time into frac- formance with respect to some performance indices and
tional order system stability. In that respect, numerical, time domain criteria. Moreover, the proposed control-
analytical and graphical stability tests are derived and ler, classical integer order PID and fractional order
given in Hwangi and Cheng,33 Merrikh-Bayat and PID controllers are all applied to a liquid level control
Karimi-Ghartemani,34 Shi and Wang,35 Yu and system in real time. The application results show that
Wang,36 Kheirizad et al.37 and Hamamci.38 the proposed controller outperforms the others.
Fractional order controllers are mostly designed in The article is organized as follows. In section
frequency domain since the frequency response of a ‘‘Fractional calculus,’’ a brief introduction to fractional
fractional expression can exactly be obtained. calculus is provided. In section ‘‘Integer order pre-
Moreover, the translation or transformation of fre- filtered PI/PID controller design,’’ the single fractional
quency domain characteristics to time domain is not a order pole model is introduced and an integer order
unique and easy task for fractional order representa- pre-filtered PI/PID controller design method is pro-
tion. Time domain design can only be achieved in an posed for two different regions specified according to
approximate manner after some integer order approxi- the fractional order of this model. Simulation studies
mations for fractional order expressions. In addition, are presented in section ‘‘Simulation results’’ to show
when a fractional order controller is somehow obtained the effectiveness of the proposed method and compare
via any domain criteria, the implementation of this its results with those of classical integer order PID and
controller is another challenge for the designer. The fractional order PID controllers. Section ‘‘Experimental
reason is that we still do not have a direct physical rep- results’’ presents the application of the proposed
resentation of fractional order operator. The best way method on double-tank liquid level control system. The
of implementing fractional order controllers can be tank system is modeled by both the first order dead
achieved through high integer order transfer functions time and the single fractional order pole model. Various
Yumuk et al. 3
comparative studies are done and the results are pre- Thus, a continuous fractional order transfer function
sented over the emerged controllers. Finally, conclusion G(s) can be represented as follows
and discussions are given in section ‘‘Conclusion.’’
Y(s) bm sam + bm1 sam1 + + b0 sa0
G(s) = =
U(s) am sbm + am1 sbm1 + + a0 sb0
Fractional calculus ð7Þ
There are many definitions of fractional derivative or where ak (k = 0, 1, 2, . . . , m) and bk (k = 0, 1, 2, . . . , n)
integral in the literature. The commonly used three of are real numbers, bk \ bk1 \ \ b0 ak \ ak1
these definitions are proposed by Grunwald–Letnikov, \ \ a0 and ak (k = 0, 1, 2, . . . , m) and bk(k=0,
Riemann–Liouville and Caputo definitions.39 1, 2, . . . , n) are constant coefficients. The continuous
30
Grunwald–Letnikov definition for numerical evalua- fractional order transfer function in equation (7) is
tions is described as follows called a commensurate transfer function when all ak
and bk orders are integer multiples of the least common
½trh
X divisor (a 2 R+ ). The commensurable transfer func-
1 n
a
r Dt g(t)= lim a ( 1)j g(t jh), tion can be written as follows
h!0 h
j=0 j ð1Þ
P
m
n 1\a\n bk ska
Y(s) k=0
G(s) = = n ð8Þ
where r and t are limits of the operator, h is time incre- U(s) P
ak ska
ment, a is a real number, ((t r)=h) denotes integer k=0
n It is said that the transfer function given in equation
part of the upper limit of summation and bino-
j
(8) is Bounded Input-Bounded Output (BIBO) stable
mial coefficient is given by if and only if 0 \ a \ 2 and j arg (sak )j . a(p=2)
(k = 1, P. . . , n) where sk are the roots of the polynomial
n G(n + 1)
= ð2Þ P(s) = nk = 0 ak ska which is the denominator of the
j G(j + 1)G(n j + 1)
commensurate fractional order transfer function in
where gamma function G(n) is equation (8).
In real-time applications, the implementation of
Б
tn1 et dt n 2 R fractional order differentiation is a challenging prob-
G(n) = 0 ð3Þ lem. For this purpose, there are various integer order
(n 1)! n2N
approximations in the literature. These approximations
Riemann–Liouville definition, which is appropriate are generally divided into two categories: continuous-
to find analytical solution of simple functions like et , tb , time implementations (Oustaloup filter approximation,
and so on40 is described as follows continued fraction approximation, etc.) and discrete-
time implementations (finite impulse response (FIR) fil-
ðt ter approximation, approximation using step or
a 1 dn g(t)
r Dt g(t) = dt, impulse response invariants).2,39 One of the most
G(n a) dtn (t t)an + 1 ð4Þ
r widely used approximation is Oustaloup filter approxi-
n 1\a\n mation,2,11,41–43 and it is defined as follows
YN
s + vk0
where r and t are limits of the operator, and a is a real sa ’ O(sa ) = K0 ð9Þ
number. Caputo fractional derivative which provides k=1
s + vk
the use of initial conditions y(0), y0 (0) instead of frac-
where a 2 (0, 1) is the order of derivative, N is the order
tional initial conditions y0:5 (0), y1:2 (0) and so on, unlike
of filter, the poles (vk ), zeros (vk0 ) and the gain (K0 ) are
Riemann–Liouville definition, is given as follows
as follows
ðt n 2k1a
a 1 d g(t) vh 2N
r Dt g(t) = dt, vk0 = vl
G(n a) dtn (t t)an + 1 ð5Þ vl
r
2k12N+ a ð10Þ
n 1\a\n vh
vk = vl
vl
where r and t are limits of the operator, and a is a real K0 = vah
number. Using these fractional derivative definitions,
the Laplace transformation of the fractional derivative where vl and vh are the lower and upper frequency bound
which have zero initial conditions, can be defined as values. The Oustaloup filter approximation completely
follows represents the fractional order differentiation when N is
equal to infinity. In the literature, N is usually taken as 11
L fr Dat g(t)g = sa G(s) ð6Þ in order to have satisfactory approximation. When the
4 Proc IMechE Part I: J Systems and Control Engineering 00(0)
Tsa + 1
C(s) = Kc ð12Þ Here, subscripts p and z indicate the poles and zeros of
s the controller in equation (13).
The above controller possesses an integrator to elim- The controller C(s) given in equation (13) has real
inate the steady-state error to step inputs and a control- poles and the pole p2 is not dominant. In order to
ler gain represented by Kc . All the work in the study is express C(s) in pre-filtered PID form, the pole p2 is
done on the basic classical unity feedback control dia- neglected and this controller is named as IIS PID
gram given in Figure 1. In the figure, R(s), E(s), U(s) (Inverse Integer order Single fractional order pole model
and Y(s) represent reference, error, control and output PID). The final form of IIS PID controller is as follows
signals, respectively.
In our study, we prefer the Oustaloup approximation Kc Ki
CIIS PID (s) = Kp + + Kd s
to fractional order differentiation. If high order approx- ts + 1 s
ð17Þ
imation is used, the integer order version of controller Kc (s z1 )(s z2 )
(12) will possess great number of zeros and poles. Even =
s(s p1 )
if the number of poles and zeros of system model are
the same as the order of the controller gets higher, the where the parameters of the controller and filter time
control effort may dramatically increase. PID control- constant are given by
lers are the most preferred controller types due to their
simple structure, easy implementation, well-known z1 + z2 z1 z2 1 1
Kp = , Ki = , Kd = , t =
effects and characteristics of its each components. p1 p1 p1 p1
Therefore, the aim of this study is to find an integer ð18Þ
order PID controller where its parameters are based on
the model parameters of single fractional order pole The parameter (Kc ) in equation (17) is a design para-
system model (11). For this reason, the second order meter. An optimization-based procedure is given in sec-
Oustaloup approximation of the fractional order differ- tion ‘‘A formula generation procedure for the free
entiation is used. Two different cases emerge designated design parameter Kc ’’ for the selection of this
by a 2 (0, 1 and a 2 (1, 2 in this design methodology. parameter.
Yumuk et al. 5
Pre-filtered PI/PID controller design for a 2 (1, 2 controller, the dominant pole can again be found easily
Similarly, using the second order Oustaloup approxi- but the dominant zeros cannot be identified quite easily
mation in fractional order differentiation, the inverse as in the previous case. The dominancy of these zeros
controller in equation (12) is approximated in the fol- will depend on T, a, vl and vh . In order to make a
lowing parametric form dominance analysis based on two fractional system
model parameters; namely, T and a, we fix the other
s3 rv1 m
a1
2 + s2 m
a1
2 + rm
a4
4 n
3a2
+ svl r + m 4 n + vma1 two parameters that emerge from implementation of
C(s) = Kc h
a1
a
a1
s m 2 s2 + svl m 4 n + vm 2 fractional differentiation in the system model, namely,
= Kc (sz 1 )(sz2 )(sz3 )
s(sp1 )(sp2 )
vl and vh . We set vl and vh values to 103 and 103 rad/
s, respectively, as it is commonly used in the
ð19Þ
literature.42,46
pffiffiffiffi
where m = (vh =vl ), n = 1 + m, r = Tvah , v = vh vl When these frequencies bound values are placed, k2 ,
and p1 , p2 are the poles and z1 , z2 , z3 are the zeros of k1 and k0 in equation (23) are obtained as follows
C(s). The poles (p1 , p2 ) of controller in equation (19)
are found in parametric form using equation (14) where 3(1 + a) 103a
k2 = 1001:10 2 +
T
a pffiffiffiffi a1 3(1 + a)
ap = 1, bp = vl m4 1 + m , cp = vh vl m 2 1001:10 2
k1 = + 103a ð24Þ
ð20Þ T
1
In order to find the zeros (z1 , z2 , z3 ) of the controller k0 =
T
in equation (19) in parametric form, the third order
equation in the numerator should be solved. For this As it is mentioned above, all zeros of equation (19)
purpose, we refer to Holmes44 and Mitchell45 and find might be on left half s plane real axis. In this case, two
the following parametric closed form formulas dominant zeros can be considered in the design of the
proposed PID controller with a pre-filter (IIS PID).
z1 = 2( p)0:5 cos 13 arccos qp ( p)0:5 0 2p 3 k2 When two of the zeros are in complex conjugate form
3 the dominancy will either occur on the real zero or on
z2 = 2( p)0:5 cos 13 arccos qp ( p)0:5 1 2p 3 k2
3 the complex conjugate ones. In this case, if the complex
z3 = 2( p)0:5 cos 13 arccos qp ( p)0:5 4 2p 3 k32 conjugate zeros are dominant, still PID type of control-
ler will be obtained. Otherwise, if the real zero is domi-
ð21Þ nant, then the controller will be in pre-filtered PI form
where which we named as IIS PI (Inverse Integer order Single
fractional order pole model PI). The final form of
3k1 3k22 2k32 9k1 k2 + 27k0 IIS PID controller is as follows
p= , q= ð22Þ
9 54
Kc Ki
and CIIS PID (s) = Kp + + Kd s
ts + 1 s
ffiffiffi ð25Þ
m
a1
2 + Tvah m
a4 p
4 ð1 + mÞ Kc (s z10 )(s z20 )
k2 = =
Tva1
a1
m 2 s(s p1 )
h
3a2 p ffiffiffi
vl Tvah + m 4 ð1 + mÞ
k1 = a1
ð23Þ where the parameters of the controller and filter time
Tva1 m 2
h constant are given by
vh vl ma1
k0 = a1
Tva1 m 2
h z10 + z20 z10 z20 1 1
Kp = , Ki = , Kd = , t =
In the pre-filtered PID controller design for the case p1 p1 p1 p1
of a 2 (0, 1, the lower and upper frequency bound val- ð26Þ
ues (vl and vh , respectively) take place in equation (13)
directly. Moreover, the poles and zeros of equation and z10 and z20 are the dominant zeros of the controller
(13) are on the left half s plane real axis and the domi- in equation (19). The IIS PI will be in the form
nant ones can obviously be identified from equations
Kc Ki Kc (s z10 )
(14)–(16). While expressing IIS PID as in equation (17) CIIS PI = Kp + = ð27Þ
using equation (13), non-dominant pole is neglected ts + 1 s s(s p1 )
and IIS PID parameters depend on vl and vh .
Here, z10 is the dominant zero and the parameters of
In the pre-filtered PI/PID controller design for the
the controller and filter time constant are given by
case of a 2 (1, 2, the poles and zeros of equation (19)
also depend on vl and vh . The poles and zeros of equa- 1 z1 0 1
tion (19) are on the left half s plane. While all poles take Kp = , Ki = , t = ð28Þ
p1 p1 p1
place on the real axis, the zeros might be in complex
conjugate form. Thus, in design of the proposed where p10 = p1 and z10 = z1 .
6 Proc IMechE Part I: J Systems and Control Engineering 00(0)
J = ISE + TV ð29Þ
ð‘
ISE = (r(i) y(i))2 ð30Þ
i=0
X
‘
TV = ju(i + 1) u(i)j ð31Þ
Figure 2. PID and PI regions for different T and a values. i=1
Figure 3. Optimal K values (a) a 2 ½0, 1, T 2 ½0, 100 and (b) a 2 ½1, 2, T 2 ½0, 100.
Yumuk et al. 7
(a) (b)
(c) (d)
Figure 4. Optimized Kc K values and fitting approximating function curves (a) a 2 ½0, 0:15, (b) a 2 ½0:15, 0:75, (c) a 2 ½0:75, 1 and
(d) a 2 ½1, 2, T 2 ½0, 100.
a k4 k3 k2 k1 k0
This function is generated by using fit command in In this case, it is observed that only one approximat-
MATLAB based on optimum values found via the opti- ing function is sufficient and it is obtained as follows
mization procedure.
0:00004072T3 0:007273T2 + 0:6526T + 5:73
Kc = f(T) =
Case A: a 2 ½0, 1 K
ð33Þ
In this case, extensive analysis on the resulting data Optimized Kc K values and the four fitting function
forced us to generate more than one approximating curves are shown in Figure 4.
function. As a result, various Kc approximating func- The procedure of the integer order pre-filtered PI/
tions are generated for different a intervals given in PID controller design method proposed can be given as
Table 1. In that way, we have obtained lesser root mean follows:
square error (RMSE) compared to single approximat-
ing function.
1. Inverse of the fractional order model is taken.
2. The steady-state error is eliminated via insertion of
Case B: a 2 ½1, 2 an integrator in the loop.
8 Proc IMechE Part I: J Systems and Control Engineering 00(0)
(a) (b)
Figure 5. (a) Step responses and (b) control signals for system (36) using C PID, IIS PID and FO PID controllers.
3. Oustaloup second order filter approximation is uti- pre-filtered PI/PID controller design.’’ The related
lized in driving integer order controller from frac- parameters of the classical integer order PID (C PID)
tional order type. and the classical fractional order PID (FO PID) are
4. A pre-filtered PID or PI controller form is found via the BB-BC optimization algorithm over the
obtained using domination rules. performance measure given in equation (29). Finally,
5. The design parameter is determined via approxi- the following controller transfer functions are obtained
mating functions given in Table 1 and equation as follows
(33).
2:207(s + 0:129415)(s + 18:1288)
CIIS PID (s) =
s(s + 0:177828)
Simulation results
1 48:5912
CC PID (s) = 52:9608 + + 6:8845s
In this section, the proposed controller is compared 2:8664s + 1 s
with pre-filtered integer order classical PID and classi- 32:9836
CFO PID (s) = 2:6797 + 0:7732 + 0:0001s1:1673
cal fractional order PID on two different system mod- s
els, which is given as follows ð37Þ
The unit step system responses and control inputs
1 Ki
CC PID (s) = Kp + + Kd s ð34Þ related to all controllers are given in Figure 5(a) and
ts + 1 s
(b). The measured performance values and resultant
Ki
CFO PID (s) = Kp + l + Kd sm ð35Þ three time domain criteria values are listed in Table 2.
s Mp , ts and tr in Table 2 indicate overshoot, settling time
The classical pre-filtered PID in equation (34) and and rise time, respectively.
fractional order PID in equation (35) possess four and It can be clearly seen that even though there is a
five design parameters, respectively. The fractional inte- small improvement in the listed time domain criteria
grator in CFO PID (s) is implemented as (O (s1l )=s) using for FO PID compared to IIS PID and C PID, the con-
Oustaloup approximation to eliminate steady-state error. trol signal is very aggressive and the magnitude of the
control signal is large compared to the cases of the
IIS PID and the C PID as it is illustrated in Figure
Simulation for a 2 (0, 1 5(b). Moreover, the IIS PID controller has the mini-
In this simulation, the following system in Luo et al.20 mum cost.
is considered as follows The robustness analysis is carried out for both
6%20 variation in Kc and T, and the results are given
1
G(s) = ð36Þ in Table 3. The robustness of the proposed controller is
0:4s0:5 +1 in a comparable level with respect to that of other
For fractional order differentiation, vl and vh are controllers.
taken as 103 and 103 rad/s, respectively. The simula- The Bode diagrams of the open loop transfer
tions are all run via FOMCON toolbox,42,46 in functions of the control systems attained by the con-
MATLAB. trollers IIS PID, C PID and FO PID are illustrated
The proposed PID controller is attained executing in Figure 6. The frequency domain criteria values
all steps of the procedure given in section ‘‘Integer order are depicted from the related Bode diagrams and
Yumuk et al. 9
Controller Mp ts tr ISE TV J
ISE: integral square error; TV: total variation of control input; PID: proportional–integral–derivative; IIS_PID: inverse integer order single fractional
order pole model PID; C_PID: classical integer order PID; FO_PID: classical fractional order PID.
ISE: integral square error; TV: total variation of control input; IIS_PID: inverse integer order single fractional order pole model proportional–integral–
derivative; C_PID: classical integer order proportional–integral–derivative; FO_PID: classical fractional order proportional–integral–derivative.
PM: phase margin; IIS_PID: inverse integer order single fractional order
pole model proportional–integral–derivative; C_PID: classical integer
order proportional–integral–derivative; FO_PID: classical fractional order
proportional–integral–derivative.
(a) (b)
Figure 7. (a) Step responses and (b) control signals for system (39) using IIS PIDr , P PID and FO PID controllers.
ISE: integral square error; TV: total variation of control input; IIS_PID: inverse integer order single fractional order pole model proportional–integral–
derivative; C_PID: classical integer order proportional–integral–derivative; FO_PID: classical fractional order proportional–integral–derivative.
Yumuk et al. 11
Figure 8. The Bode diagrams of open loop transfer functions Figure 9. Liquid level control system: (a) front view and (b)
of the control systems attained by the controllers IIS PID, C PID back view.
and FO PID. 1. Proportional valve; 2. level sensor; 3. tanks; 4. transition valves
between tanks; 5. drain valves; 6. reservoir tank; 7. start/stop buttons; 8.
tank level selector and transducer; 9. pressure/voltage (0–2 PSI/1–5 V)
transducer; 10. electropneumatic regulator (current/pressure (4–20 mA/
0.7–14 PSI) transducer); 11. electrical level data (to PLC); 12.
Table 7. The frequency domain criteria values. proportional valve position data (from PLC); 13. compressor; 14. input/
output unit of PLC; 15. pump; and 16. circuit breaker.
Controllers PM vc (rad/s) L (s) Ms
Figure 10. Overall block diagram of the double-tank liquid level control system.
Kf
Gf (s) = eLs ð41Þ
Tf s + 1 Figure 11. Open loop step responses of the real-time system
and its various models.
where Kf , Tf and L are the gain, time constant and
dead time of FOPDTM, respectively. Another reason
of selecting the model in equation (41) is that it has square error in the result of optimization. According to
three free parameters as in the single fractional pole the total squared error, the best fitted model is the sin-
model given in equation (11). Thus, a fair modeling gle fractional order pole model. The collected data
comparison is obtained. Both model parameters are from system and the step responses of the models are
obtained using the total squared error given below depicted in Figure 11.
In this section, we provide two different types of
t =X
Tsample
comparison for the proposed controller. First, the pro-
etotal = (ymodel (t) ymeasurement (t))2 ð42Þ
t=0
posed controller is compared with C PID and FO PID
designed using SFPM as it is done in section
where Tsample , ymodel (t) and ymeasurement (t) indicate the ‘‘Simulation results.’’ In the second part, the proposed
number of sample points, the output of the model and controller is compared with pre-filtered PID in Wang
collected data from the system, respectively. BB-BC et al.49 designed for FOPDTM.
Algorithm is utilized for optimization of equation (42). The performance measure in equation (29) is opti-
Table 8 shows the obtained model parameters and total mized using BB-BC optimization algorithm so as to
Yumuk et al. 13
(a) (b)
Figure 12. The liquid level control system real time: (a) step response and (b) control signal.
ISE: integral square error; TV: total variation of control input; IIS_PID: inverse integer order single fractional order pole model proportional–integral–
derivative; C_PID: classical integer order proportional–integral–derivative; FO_PID: classical fractional order proportional–integral–derivative.
find four parameters of C PID and five parameters of Comparison of the proposed controller with pre-
FO PID. The proposed PID controller is attained filtered PID designed for FOPDTM
executing all steps of the procedure given in section
As it is seen in Table 8, FOPDTM for double-tank
‘‘Integer order pre-filtered PI/PID controller design.’’
liquid level system is given as follows
The controllers are found as follows
0:88692
(s + 0:009007)(s + 1:9011) Gf (s) = e1:57s ð46Þ
GIIS PID = 20:9019 ð43Þ 119:5197s + 1
s(s + 0:5311)
According to total squared error values given in Table
1 10:3782
GC PID = 66:1761 + + 0:4383 8, it can easily be concluded that FOPDTM does not
0:1s + 1 s so adequately represent the real system compared to
ð44Þ SFPM.
In this section, the proposed controller (CIIS PID (s))
54:8377
GFO PID = 48:3351 + + 0:1816s0:2 : ð45Þ is compared with pre-filtered PID proposed by Wang
s
et al.49 designed for FOPDTM. The transfer function
In order to be able to use the same implementation of pre-filtered PID (CWang PID (s)) is obtained as follows
PLC blocks in all controller design cases, we had to
make a slight simplification in equation (45) control- 1 0:779s
CWang PID (s) = 46:075 1 + +
ler transfer function. Step responses and the control 120:3s 0:0779s + 1
signals obtained using controllers (43)–(45) are given (119:515s + 1)(0:862523s + 1)
= 0:383
in Figure 12. The performance of the proposed con- s(0:0779s + 1)
troller is better than C PID and FO PID with respect ð47Þ
to the performance criteria J and also gives satisfac-
tory time domain response. Moreover, the amplitude Step responses and the control signals obtained using
of the control signal of the proposed controller is controllers (43) and (47) are given in Figure 12. These
lower. The performance comparison of the controllers two controllers have the same overshoot values, but the
is given in Table 9. proposed controller has better rise time and settling
14 Proc IMechE Part I: J Systems and Control Engineering 00(0)
20. Luo Y, Chen YQ, Wang CY, et al. Tuning fractional 34. Merrikh-Bayat F and Karimi-Ghartemani M. An effi-
order proportional integral controllers for fractional cient numerical algorithm for stability testing of
order systems. J Process Contr 2010; 20: 823–831. fractional-delay systems. ISA T 2009; 48: 32–37.
21. Bettayeb M and Mansouri R. Fractional IMC-PID-filter 35. Shi M and Wang Z. An efficient analytical criterion for
controllers design for non-integer order systems. J Pro- stability testing of fractional-delay systems. Automatica
cess Contr 2014; 24: 261–271. 2011; 47: 2001–2005.
22. Tepljakov A, Petlenkov E and Belikov J. Gain and 36. Yu YJ and Wang ZH. A graphical test for the interval
order scheduled fractional-order PID control of fluid stability of fractional-delay systems. Comput Math Appl
level in a multi-tank system. In: Proceedings of 2014 2011; 62: 1501–1509.
IEEE international conference on fractional differentia- 37. Kheirizad I, Jalali AA and Khandani K. Stabilization of
tion and its applications, Catania, 23–25 June 2014. fractional-order unstable delay system by fractional order
New York: IEEE. controllers. J Syst Contr Eng 2012; 226: 1166–1173.
23. Yumuk E, Guzelkaya M, Eksin I, et al. Design of an 38. Hamamci SE. An algorithm for stabilization of fractional-
integer order PID controller for single fractional order order time delay systems using fractional-order PID con-
pole model. In: Proceedings of 2015 IEEE conference on trollers. IEEE T Automat Contr 2007; 52(10): 1964–1969.
systems, process and control (ICSPC), Bandar Sunway, 39. Valerio D and da Costa JS. Introduction to single-input,
Malaysia, 18–20 December 2015. New York: IEEE. single-output fractional control. IET Control Theory A
24. Yumuk E, Guzelkaya M and Eksin I. Reduced integer 2011; 5(8): 1033–1057.
order inverse controller design for single fractional pole 40. Loverro A. Fractional calculus: history, definitions and
model. In: Proceedings of MED’16: the 24th Mediterra- applications for the engineering, 2004, http://citeseer-
nean conference on control and automation, Athens, 21–24 x.ist.psu.edu/viewdoc/download;jsessionid=304F1A4581
June 2016. New York: IEEE. D9686D683E3FBE04E60DDD?doi=10.1.1.363.9101&
25. Wei Y, Sun Z, Hu Y, et al. On fractional order composite rep=rep1&type=pdf
model reference adaptive control. Int J Syst Sci 2016; 47: 41. Tepljakov A, Petlenkov E, Belikov J, et al. Design and
2521–2531. implementation of fractional-order PID controllers for
26. Ladaci S, Loiseau JJ and Charef A. Fractional order a fluid tank system. In: Proceedings of 2013 American
adaptive high-gain controllers for a class of linear sys- control conference, Washington, DC, 17–19 June 2013.
tems. Commun Nonlinear Sci 2008; 13: 707–714. New York: IEEE.
27. Zhang B, Pi Y and Luo Y. Fractional order sliding-mode 42. Tepljakov A, Petlenkov E and Belikov J. FOMCON: a
control based on parameter auto-tuning for velocity con- MATLAB toolbox for fractional-order system identifica-
trol of permanent magnet synchronous motor. ISA T tion and control. Int J Microelectron Comput Sci 2011;
2012; 51: 649–656. 2(2): 51–62.
28. Dadras S and Momeni HR. Control of fractional-order 43. Taijudin M, Arshad NM and Adnan R. A design of
economical system via sliding mode. Physica A 2010; 389: fractional-order PI controller with error compensation.
2434–2442. Eng Technol Int J Comput 2013; 7(6): 727–735.
29. Delavari H, Ghaderi R, Ranjbar A, et al. Fuzzy frac- 44. Holmes GC. The use of hyperbolic cosines in solving
tional order sliding mode controller for nonlinear sys- cubic polynomials. Math Gaz 2002; 86: 473–477.
tems. Commun Nonlinear Sci 2010; 15: 963–978. 45. Mitchell DW. Solving cubic by solving triangles. Math
30. Shah P and Agashe S. Review of fractional PID control- Gaz 2007; 91: 514–516.
ler. Mechatronics 2016; 38: 29–41. 46. Tepljakov A, Petlenkov E and Belikov J. FOMCON tool-
31. Tavazoei MS and Haeri M. A note on the stability of box [Online], 2011, http://www.fomcon.net/
fractional order systems. Math Comput Simulat 2009; 79: 47. Erol OK and Eksin I. A new optimization method: big
1566–1576. bang-big crunch. Adv Eng Softw 2006; 37: 106–111.
32. Jiao Z and Chen YQ. Stability of fractional-order linear 48. Tavazoei MS. Notes on integral performance indices in
time-invariant system with multiple noncommensurate fractional-order control system. J Process Contr 2010; 20:
orders. Comput Math Appl 2012; 64: 3053–3058. 285–291.
33. Hwangi C and Cheng Y. A numerical algorithm for stabi- 49. Wang FS, Juang WS and Chan CT. Optimal tuning of
lity testing of fractional delay systems. Automatica 2006; PID controllers for single and cascade control loops.
42: 825–831. Chem Eng Commun 1995; 132: 15–34.