See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/333981393

Frequency Response Based Curve Fitting Approximation of Fractional–Order

PID Controllers

Article in International Journal of Applied Mathematics and Computer Science · January 2019
DOI: 10.2478/amcs-2019-0023

CITATIONS READS

22 766

4 authors:

Kishore Bingi Rosdiazli Ibrahim

VIT University Universiti Teknologi PETRONAS
90 PUBLICATIONS 362 CITATIONS 155 PUBLICATIONS 1,110 CITATIONS

SEE PROFILE SEE PROFILE

Mohd Nohd Karsiti Sabo Miya Hassan

Universiti Teknologi PETRONAS Abubakar Tafawa Balewa University
72 PUBLICATIONS 402 CITATIONS 65 PUBLICATIONS 495 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

TNSCST Project View project

Advanced controller design View project

All content following this page was uploaded by Kishore Bingi on 09 July 2019.

The user has requested enhancement of the downloaded file.

Int. J. Appl. Math. Comput. Sci., 2019, Vol. 29, No. 2, 311–326
DOI: 10.2478/amcs-2019-0023

FREQUENCY RESPONSE BASED CURVE FITTING APPROXIMATION OF

FRACTIONAL–ORDER PID CONTROLLERS

K ISHORE BINGI a,∗ , ROSDIAZLI IBRAHIM a , M OHD N OH KARSITI a , S ABO M IYA HASSAM b ,
V IVEKANANDA R AJAH HARINDRAN c

a
Department of Electrical and Electronics Engineering
PETRONAS University of Technology, Seri Iskandar, 32610 Perak, Malaysia
e-mail: {bingi.sai_g03426,rosdiazli,nohka}@utp.edu.my
b
Department of Electrical and Electronics Engineering
Abubakar Tafawa Balewa University, PMB 0248 Bauchi, Nigeria
e-mail: [email protected]
c
Instrumentation and Control
PETRONAS Group Technical Solutions, Jalan Sultan Hishamuddin, 50050 Kuala Lumpur, Malaysia
e-mail: [email protected]

Fractional-order PID (FOPID) controllers have been used extensively in many control applications to achieve robust con-
trol performance. To implement these controllers, curve fitting approximation techniques are widely employed to obtain
integer-order approximation of FOPID. The most popular and widely used approximation techniques include the Oustaloup,
Matsuda and Cheraff approaches. However, these methods are unable to achieve the best approximation due to the limi-
tation in the desired frequency range. Thus, this paper proposes a simple curve fitting based integer-order approximation
method for a fractional-order integrator/differentiator using frequency response. The advantage of this technique is that it
is simple and can fit the entire desired frequency range. Simulation results in the frequency domain show that the proposed
approach produces better parameter approximation for the desired frequency range compared with the Oustaloup, refined
Oustaloup and Matsuda techniques. Furthermore, time domain and stability analyses also validate the frequency domain
results.

Keywords: curve fitting, fractional-order PID controller, frequency response, integer-order approximation, Oustaloup ap-
proximation, Matsuda approximation.

1. Introduction stable performance, especially for higher-order systems.

Moreover, the controller can easily attain the iso-damping
In industrial process control applications, PID controllers property (Monje et al., 2010; Shah and Agashe, 2016).
are most widely employed for low-level control (Bingi It should be noted that seven different configurations
et al., 2018a). This is because they are simple to design can be achieved with the PIλ Dμ controller (i.e., P, PI,
and easy to tune, and their implementation is seamless PIλ , PD, PDμ , PID and PIλ Dμ ) (Xue et al., 2007; Xue,
(Pachauri et al., 2018; Bingi et al., 2018b; Shah and 2017; Monje et al., 2010; Kishore et al., 2018; Shah and
Agashe, 2016). The fractional-order PID (FOPID or Agashe, 2016). However, a key issue with the practical
PIλ Dμ ) controller is a variant of PID realized through realization or equivalent circuit implementation of such
fractional-ordering of the integral and derivative actions controllers in a finite-dimensional integer-order system is
(Monje et al., 2010; Xue et al., 2007; Shah and Agashe, the approximation of the fractional-order parameters. This
2016; Xue, 2017). A key feature of PIλ Dμ is that it has generated a lot of interest among researchers recently.
is robust to system parameter variations and provides
For effective approximation of a fractional-order
∗ Corresponding author integrator and differentiator in the PIλ Dμ controller,

© 2019 K. Bingi et al.

This is an open access article distributed under
Brought to you by | Universiti Teknologi Petronas (UTP)
the Creative Commons Attribution-NonCommercial-NoDerivs license
(http://creativecommons.org/licenses/by-nc-nd/3.0/). Authenticated
Download Date | 7/9/19 9:03 AM
 312 K. Bingi et al.

researchers have proposed several frequency domain method (Du et al., 2017), the time moments approach
approximation techniques (Krishna, 2011; Vinagre et al., (Khanra et al., 2013), the state space approach (Poinot
2000; Li et al., 2017). Despite these proposals, it is and Trigeassou, 2003; Krajewski and Viaro, 2011) and the
very difficult to determine the best method. This is frequency distribution mode (Wei et al., 2014b) have been
because, while considering certain conditions such as proposed, too. A key issue with these methods is that they
the order of approximation or the accuracy of frequency are quite complex and hence difficult to implement.
and time responses, some of these methods can be more Motivated by the discussion above, this paper
advantageous over others (de Oliveira Valério, 2005; proposes a simple curve fitting approximation approach
Djouambi et al., 2007; Deniz et al., 2016). using exact frequency response data of fractional-order
Among the frequency domain approximation operators (differentiator/integrator). The proposed
techniques available in the literature, the Oustaloup approach is expected to achieve better approximation
approximation is most popular and widely used. It compared with the commonly used Oustaloup, refined
is based on a recursive distribution of poles and Oustaloup and Matsuda techniques (Monje et al.,
zeros in the desired frequency range using frequency 2010). To demonstrate the performance of the proposed
response fitting (Merrikh-Bayat, 2012). In certain approach, a simulation study will be conducted on a class
situations, the Oustaloup approximation allows fitting of fractional-order based controllers and systems.
the entire frequency range of interest (Oustaloup The remaining sections of the paper are organized
et al., 2000; Monje et al., 2010). Thus, to overcome as follows: the definitions of the fractional-order
this drawback, a modified or refined version has been differintegral operator and the fractional-order PID
proposed (Merrikh-Bayat, 2012; Xue et al., 2006; Meng controller as well as an overview of the Oustaloup,
and Xue, 2012; Wei et al., 2014a; Krajewski and refined Oustaloup and Matsuda approximation algorithms
Viaro, 2014; Atherton et al., 2014). However, it produces are presented in Section 2. The proposed curve fitting
a higher-order approximated model. approximation using the frequency response and the
Subsequently, to reduce the order of approximation, integer-order approximation table for a fractional-order
Liang et al. (2014) proposed a fixed-pole approximation differentiator are given in Section 3. A simulation
technique. Similarly, Carlson derived a different study on fractional-order based controllers and systems
technique (Tepljakov et al., 2012) using the Newton to demonstrate the performance of the proposed
iterative method for continued fractional expansion (CFE) approximation is given in Section 4. Finally, Section 5
of the fractional-order differentiator. However, this concludes the paper.
technique is limited to fewer values of the fractional-order
parameter (Tepljakov et al., 2012; Sheng et al., 2011).
Later, Matsuda proposed an approximation technique 2. Preliminaries
using the gain of the fractional-order transfer function This section is divided into two parts. The first will briefly
(Valério et al., 2013). However, in this method, if the discuss fractional calculus while the second will give an
order of approximation is chosen as an odd number, overview of some standard approximations.
the approximated transfer function will be improper;
i.e., there will be one more zero than poles (Yüce
et al., 2017; Vinagre et al., 2000; Sheng et al., 2011). 2.1. Fractional calculus. In this subsection, the
Other researchers have also proposed various power definitions of the fractional-order differintegral operator
series expansion (PSE) techniques based on Taylor series, and the fractional-order PID controller are presented.
Maclaurin series, etc. (Valério et al., 2013; Petráš, 2011a;
2011b; Caponetto, 2010). Nevertheless, it is proven that 2.1.1. Fractional-order differintegral. In fractional
CFE methods are converging more rapidly than PSE ones calculus, the fractional-order differintegral operator, a
(Vinagre et al., 2000). combined fractional-order differentiator and integrator,
In the related development, Charef proposed an which generalizes the notation for the differentiator
approximation technique (Das, 2011) where the accuracy (Re(γ) > 0) and the integrator (Re(γ) < 0) for
is determined by properly selecting the maximal the function x(t) (Kaczorek, 2018; Joice Nirmala and
permissible error. However, the order of approximation Balachandran, 2017), is defined as
involves a significant amount of trials and errors ⎧ t
(Mitkowski and Oprzedkiewicz, 2016; Oprzedkiewicz, ⎪
⎪ x(τ ) dτ γ , Re(γ) < 0,
2014). Thus, an extension of this method was proposed by ⎨ 0
γ
a DT x(t) =
x(t), Re(γ) = 0, (1)
Meng and Xue (2012). This extended method is focused ⎪
⎪ γ
on improving the accuracy of the original proposal. ⎩ d x(t) , Re(γ) > 0,
dtγ
Other approximation algorithms based on the stability
boundary locus (Deniz et al., 2016), the vector fitting where

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
Frequency response based curve fitting approximation of fractional-order PID controllers 313

• T and a are the lower and upper bounds of the with the zeros ωk and poles ωk of (4) computed as
operator,  2k−1−γ
2N
ωh
• γ is the order of the operator, and ωk = ωl , (5)
ωl
• Re(γ) is the real part of γ.  2k−1+γ
2N
ωh
ωk = ωl , (6)
The Laplace transform of (1) at zero initial ωl
conditions is given as where

L{0 DTγ x(t); s} = sγ X(s). (2) • γ is the order of the fractional-order derivative,
• N is the order of approximation, and
From (2), the approximation of the fractional-order
differentiator (sγ ) for γ > 0 and the fractional-order • (ωl , ωh ) is the frequency range of interest.
integrator (1/sγ ) for γ < 0 will be made using proposed
curve fitting approach and other approximation techniques The Oustaloup approximation is the most widely
in Section 3. used technique for integer-order approximation of
fractional-order operators. However, for practical
applications, it is frequently found that it cannot fit the
2.1.2. Fractional-order PID controller (FOPID whole expected range of frequency (Monje et al., 2010;
or PIλ D μ ). The generalized transfer function of the Xue et al., 2006; 2007).
PIλ Dμ controller as reported by Shah and Agashe (2016)
is
2.2.2. Refined Oustaloup approximation. The
Ki modified or refined Oustaloup approximation of sγ in the
C(s) = Kp + λ + Kd sμ , 0 < λ, μ < 2, (3)
s desired frequency range (ωl , ωh ) is defined as
 γ
where dωh ds2 + bωh s
sγ ≈
b d(1 − γ)s2 + bωh s + dγ
• Kp , Ki and Kd are the proportional, integral and (7)
derivative constant gains, N
s + ωk
× , 0 < γ < 1,
s + ωk
• λ is the order of integration, and k=−N

with ωk and ωk being the respective zeros and poles

• μ is the order of differentiation.
computed as
From (3), the approximation of the fractional-order  2k−1−γ
2N
integrator (1/sλ ) and the fractional-order differentiator ωh
ωk= ωl , (8)
(sμ ) will be performed using the proposed curve ωl
fitting approach and other approximation techniques in  2k−1+γ
2N
ωh
Section 3. ωk = ωl , (9)
ωl

2.2. Overview of some standard approximation where b and d are constants with values set as 10 and
algorithms. In this subsection, three of the standard 9, respectively, to achieve good approximation (Xue
frequency domain approximation algorithms for the et al., 2007). This modified approximation has very high
fractional-order differentiator (sγ ) are presented. The accuracy in the entire frequency range (Xue et al., 2006).
approximation algorithms considered are the Oustaloup, However, the method results in a very high integer-order
refined Oustaloup and Matsuda ones. The performance transfer function.
of the proposed curve fitting technique will be compared
these algorithms in Sections 3 and 4. 2.2.3. Matsuda approximation. The Matsuda
approximation of sγ will be performed in two steps. First,
2.2.1. Oustaloup approximation. The Oustaloup a rational model of sγ will be obtained using the continued
approximation of the fractional-order differentiator (sγ ) fraction expansions (CFE) method. Then, the fitting of
in the desired frequency range (ωl , ωh ) as defined by the original function at desired frequency points ω0 , ω1 ,
Oustaloup et al. (2000) is . . . , ωn is performed. Thereby, the approximated transfer
function of sγ is

N
s + ωk s − ω0
sγ ≈ ωhγ , 0 < γ < 1, (4) sγ ≈ d0 (ω0 ) + s−ω1 , (10)
s + ωk d1 (ω1 ) + s−ω2
k=1 d2 (ω2 )+ ...

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
 314 K. Bingi et al.

where where τ = 1, . . . , T is the iteration step. From (17), it

should be noted that the unbiased curve fitting is achieved
d0 (ω) = |(jω)γ |, (11) when Qτ −1 (jωk ) approaches Qτ (jωk ). Furthermore, to
ω − ωk obtain a minimal realization of G(s), the matching poles
dk+1 (ω) = , (12)
dk (ω) − dk (ωk ) and zeros will be cancelled.
k = 0, 1, . . . , N. The final stage is to convert the obtained state-space
model given in (18) after pole-zero cancellation into a
In this method, the sum of the total number of transfer function model of the following form:
zeros and poles is known as the order of approximation
N . Here, N should be an even number; otherwise, the sX(s) = AX(s) + BU (s),
approximation technique results in an improper transfer (18)
Y (s) = CX(s) + DU (s),
function (Vinagre et al., 2000; Yüce et al., 2017; Deniz
et al., 2016). Y (s)  −1
G(s) ≈ = C sIN − A B + D, (19)
U (s)
3. Proposed curve fitting approximation where
The first stage of the proposed curve fitting approximation
is obtaining the frequency response data (frd) of sγ . • X(s), Y (s) and U (s) are the state, output and control
This is accomplished by substituting s = jω and then vectors, respectively,
evaluating the function for different values of ω ∈ • A, B, C and D are the state, input, output,
(ωl , ωh ) as follows: feedforward matrices, respectively.
sγ = (jω)γ |ω=(ωl ,...,ωh ) . (13)
Therefore, the above procedure for the proposed
The next stage is to derive an integer-order transfer integer-order approximation of sγ based on the curve
function model G(s) from the data obtained in (13) fitting of frequency response data with the use of
using the Sanathanan–Koerner (SK) least-squares iterative MATLAB built-in commands will be implemented as
method (Shi, 2016). Thus, the transfer function model of follows:
G(s) is defined as follows:
1. Obtain the frequency response data for integer-order
N part of sγ within the desired frequency range ω ∈ (ωl ,
p n sn ωh ) using the MATLAB inbuilt function frd().
P (s) n=0 P ψ(s)
G(s) = ≈ N
≈ , (14) 2. Obtain the exact frequency response data of sγ by
Q(s) 1 + Qφ(s)
n raising the data obtained in the previous step to a
1+ qn s
n=1 power of γ.
where the coefficients P , Q and the monomial functions 3. Choose the approximation order N for the
ψ(s), φ(s) are defined as follows: integer-order model.
⎡ ⎤ ⎡ ⎤
p0 q1 4. Obtain the state space model of exact function
⎢ p1 ⎥ ⎢ q2 ⎥
⎢ ⎥ ⎢ ⎥ response data based on SK’s least-squares iteration
P = ⎢ . ⎥, Q = ⎢ . ⎥, (15)
⎣ . ⎦
. ⎣ .. ⎦ method by using inbuilt MATLAB function
pN qN fitfrd().

ψ(s) = [1, s, . . . , sN ], φ(s) = [s, s2 , . . . , sN ]. (16) 5. Convert the state space model to the transfer function
using the inbuilt MATLAB command ss2tf().
The objective is to identify P and Q in the
integer-order transfer function G(s) in the desired The MATLAB commands for implementing the
frequency range ω ∈ (ωl , ωh ). This is achieved by proposed algorithm are provided in Appendix A.
minimizing the difference between data samples (H(s)) To demonstrate the proposed approach, consider the
obtained from (13) and G(s) using the following Levy fractional-order differentiator s0.1 . Here, the desired
linearized cost function with the SK least-squares iteration frequency range ω is chosen as (10−2 , 102 ). Furthermore,
method: to study the effect of variation in N , the order of
h  2 approximation N is chosen as 4, 5 and 6. The choice
 P (jωk ) Qτ (jωk ) 
arg min  − H(jω ) , for the range of frequency and the order of approximation
P,Q  Qτ −1 (jωk ) Qτ −1 (jωk )
k 
k=l is based on works reported by Yüce et al. (2017), Meng
(17) and Xue (2012), Xue et al. (2007) and Deniz et al.

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
Frequency response based curve fitting approximation of fractional-order PID controllers 315

(2016). Thus, the approximated transfer functions using

the proposed approach for chosen orders are 10 Exact
Oustaloup

Magnitude (dB)
1.714s4 + 75.22s3 + 248.1s2 + 83.38s 5 Refined Oustaloup
Matsuda
Proposed Method
0
+ 1.935
s0.1 ≈ , (20)
s + 58.57s3 + 244.1s2 + 103.5s
4 -5

-10
+ 3.237 10-3 10-2 10-1 100 101 102 103

1.777s5 + 123.9s4 + 873.4s3 + 909.9s2 15

Phase (deg)
+ 137.7s + 1.914 10
s0.1 ≈ 5 4 3 2
, (21)
s + 90.81s + 785.4s + 985s 5
+ 182.9s + 3.335
0
1.831s6 + 183.7s5 + 2279s4 + 5199s3 10-3 10-2 10-1 100 101 102 103
Frequency (rad/sec)
+ 2427s2 + 203s + 1.9
s0.1 ≈ . (22) Fig. 1. Bode plots of the fractional-order differentiator (s0.1 ) for
s6 + 128.7s5 + 1919s4 + 5142s3
N = 4 approximated using various methods.
+ 2818s2 + 283.7s + 3.424
Consequently, the Bode plots of (20)–(22) are 10 Exact
Oustaloup
presented in Figs. 1–3, respectively. The plots are
Magnitude (dB)

Refined Oustaloup
5
Matsuda
compared with the Oustaloup, refined Oustaloup and Proposed Method
0
Matsuda approximations. From the figures, it can be
observed that, for the orders of approximations 5 and 6, -5

the proposed approach is more accurate within the desired -10

10-3 10-2 10-1 100 101 102 103
frequency range than the other methods.
To evaluate the effectiveness of the proposed 15

approach, a time domain comparison was also performed.

Phase (deg)

10
For this purpose , the exact step response of the
fractional-order integrator 1/sγ is obtained from the 5
inverse Laplace transform of the integer-order integrator
1/sn as 0
10-3 10-2 10-1 100 101 102 103
  Frequency (rad/sec)
1 tn−1
L−1 n = , n ∈ N. (23)
s (n − 1)! Fig. 2. Bode plots of the fractional-order differentiator (s0.1 ) for
N = 5 approximated using various methods.
As in the case of (23), the inverse Laplace transform
of the fractional integrator 1/sγ is derived as Exact
10
  Oustaloup
tγ−1
Magnitude (dB)

Refined Oustaloup
−1 1 5
L = , 0 < γ < 1, (24) Matsuda
sγ Γ(γ) 0
Proposed Method

where Γ(γ) = (γ − 1)!. From (24), the step response is -5

  -10
10-3 10-2 10-1 100 101 102 103
1 tγ tγ
L−1 γ+1 = = . (25)
s Γ(γ + 1) γΓ(γ) 15
Phase (deg)

This equation will be used to calculate the exact step 10

response of the fractional order operator. Thus, the exact

step response of s0.1 is given by
5

0
0.1 × Γ(0.1) 10-3 10-2 10-1 100 101 102 103
L−1 (s0.1 ) = . (26)
t0.1 Frequency (rad/sec)

The step responses of the proposed approach in Fig. 3. Bode plots of the fractional-order differentiator (s0.1 ) for
comparison with the Oustaloup, refined Oustaloup and N = 6 approximated using various methods.
Matsuda approximations for the orders 4, 5 and 6, are

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
 316 K. Bingi et al.

presented in Figs. 4, 5 and 6, respectively. For numerical

assessment of the results shown in the figures, the mean 1
absolute error (MAE) is chosen Exact
Oustaloup
n 0.9 Refined Oustaloup
1 Matsuda
MAE = |yi − xi |, (27) Proposed Method
n i=1 0.8

Amplitude
where 0.7

• yi is the approximated frequency response data,

0.6

• xi is the exact frequency response data, and

0.5
• n is the total size of the data.
This type of error function is most suitable for 0.4
0 50 100 150 200 250
non-time based problems (Deniz et al., 2016). Therefore, Time (s)

the numerical assessment of the responses is given in

Table 1. Observing the figures and the table, it can be seen Fig. 4. Step responses of the fractional-order differentiator
that the proposed approach outperformed the compared (s0.1 ) for N = 4 approximated using various methods.
techniques. The numerical analysis also confirmed
that the proposed approach produced less error when 1

compared with other approaches. Exact

Oustaloup
Furthermore, the stability analysis of the 0.9 Refined Oustaloup
Matsuda
approximated transfer function (G(s)) will be made Proposed Method
using H2 and H∞ -norms for the frequency band 0.8

ω ∈ (ωl , ωh ) as defined in (28) and (29), respectively.

Amplitude

From the equations, it can be noted that, for a stable 0.7

system, the H2 -norm is the average system gain over all

frequencies while the H∞ -norm is the peak gain of the 0.6

frequency response. However, for an unstable system,

these norms will be infinite: 0.5

  ∞
1 0.4
||G(s)||2 = trace[G(jω)∗ G(jω)] dω, (28) 0 50 100 150 200 250
2π −∞ Time (s)

||G(s)||∞ = sup |G(jω)|. (29) Fig. 5. Step responses of the fractional-order differentiator
ω∈(ωl ,ωh ) (s0.1 ) for N = 5 approximated using various methods.
Thus, the stability analysis of the proposed approach and
other compared techniques from Table 1 shows that all 1
the approximated transfer functions are stable with a finite Exact
Oustaloup
value of the H∞ -norm. 0.9 Refined Oustaloup
Matsuda
The approximation table for the fractional-order Proposed Method
differentiator sγ (γ = 0.1, 0.2, . . . , 0.9) using the 0.8

proposed approach is given in Table 2. The table can

Amplitude

be used directly to obtain the approximated transfer 0.7

function of fractional-order based systems and controllers.

On the other hand, the numerical assessment of the 0.6

approximation table for the step response of the proposed

approach in comparison with the Oustaloup, refined 0.5

Oustaloup and Matsuda techniques is given in Table 3.

From the table, it can be noted that for longer time periods, 0.4
0 50 100 150 200 250
the proposed approach yields a better approximation than Time (s)

the other methods. Furthermore, observing the stability

analysis in the table, it can also be noted that all the Fig. 6. Step responses of the fractional-order differentiator
approximated transfer functions using various techniques (s0.1 ) for N = 6 approximated using various methods.
are stable with a finite value of the H∞ -norm.

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
Frequency response based curve fitting approximation of fractional-order PID controllers 317

Table 1. Numerical and stability analysis of the fractional-order differentiator (s0.1 ).

Technique Order t1 ∈ (0, 125) t2 ∈ (0, 250) H∞ -norm Stability
Oustaloup 0.0215 0.0441 1.5849 stable
refined Oustaloup 0.0547 0.0809 1.7425 stable
4
Matsuda 0.0144 0.0122 1.8280 stable
proposed 0.0061 0.0199 1.7145 stable
Oustaloup 0.0237 0.0455 1.5849 stable
refined Oustaloup 0.0540 0.0813 1.7425 stable
5
Matsuda 0.0110 0.0120 1.8905 stable
proposed 0.0053 0.0097 1.7773 stable
Oustaloup 0.0247 0.0463 1.5849 stable
refined Oustaloup 0.0538 0.0814 1.7425 stable
6
Matsuda 0.0103 0.0110 1.9462 stable
proposed 0.0071 0.0059 1.8315 stable

Table 2. Approximations of fractional-order differentiators using the proposed curve fitting approach.
sγ Approximated transfer function
0.1 1.777s5 +123.9s4 +873.4s3 +909.9s2 +137.7s+1.914
s s5 +90.81s4 +785.4s3 +985s2 +182.9s+3.335
3.233s5 +223s4 +1624s3 +1762s2 +275.3s+3.725
s0.2 s5 +116.8s4 +1279s3 +2011s2 +473s+11.14
6.048s5 +413.7s4 +3111s3 +3513s2 +565.4s+7.36
s0.3 s5 +151s4 +2089s3 +4115s2 +1224s+37.33
11.7s5 +794.7s4 +6166s3 +7236s2 +1198s+14.71
s0.4 s5 +197.2s4 +3438s3 +8478s2 +3181s+125.9
23.59s5 +1594s4 +12740s3 +15520s2 +2632s+29.63
s0.5 s5 +262.6s4 +5753s3 +17720s2 +8368s+429.6
50.26s5 +3381s4 +27790s3 +35050s2 +6070s+59.93
s0.6 s5 +361.4s4 +9918s3 +38090s2 +22550s+1498
116s5 +7769s4 +65570s3 +85390s2 +15010s+121.2
s0.7 s5 +526.7s4 +18060s3 +86240s2 +63740s+5453
305.7s5 +20410s4 +176400s3 +236400s2 +41890s+244.3
s0.8 s5 +857.9s4 +36660s3 +217000s2 +199500s+21830
1092s5 +72700s4 +642000s3 +881700s2 +156000s+489.3
s0.9 s5 +1852s4 +98540s3 +720500s2 +819500s+113800

4. Simulation study et al., 2013). Thus, sγ is factorized as follows:

sγ = sγ sγ−γ . (30)
In this section, a simulation study will be conducted
on five examples of fractional-order based controllers 4.1. Fractional-order differentiator. This example
and systems. The selected systems are a differentiator, demonstrates the approximation of a fractional-order
integrator, a PID controller, a low-pass filter and a differentiator
higher-order transfer function. The obtained results G(s) = s0.26 (31)
from the proposed approach will be compared with using the proposed approach. A comparison with the
those of the Oustaloup, refined Oustaloup and Matsuda exact solution and other approximation methods will also
approximations. Furthermore, the stability analysis of be given. Thus, the approximated transfer function using
approximated transfer functions will be conducted using the proposed approach is
H-norms. In all the cases, the selected desired frequency
7.147s5 + 1836s4 + 2.493 × 104 s3
range ω and the order of approximation N are chosen as
(10−3 , 103 ) and 5, respectively. As mentioned earlier, + 3.393 × 104 s2 + 4811s + 29.74
G(s) ≈ . (32)
the choice for the range of frequency and order of s5 + 597.2s4 + 1.608 × 104 s3
approximation is based on the works reported by Yüce
+ 3.85 × 104 s2 + 1.022 × 104 s + 163.4
et al. (2017), Meng and Xue (2012), Xue et al. (2007) and
Deniz et al. (2016). In addition, the order of sγ is limited Similarly, the approximated transfer functions using other
to the [−1, 1] range for effective approximation (Valério methods are given in Appendix B.

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
 318 K. Bingi et al.

Table 3. Numerical and stability analysis of approximation Table 2.

sγ Technique t1 ∈ (0, 125) t2 ∈ (0, 250) H∞ -norm Stability
Oustaloup 0.0265 0.0510 2.5119 stable
refined Oustaloup 0.0832 0.1065 3.0744 stable
s0.2
Matsuda 0.0286 0.0268 3.5911 stable
proposed 0.0208 0.0175 2.6816 stable
Oustaloup 0.0262 0.0450 3.9811 stable
refined Oustaloup 0.0943 0.1045 5.5103 stable
s0.3
Matsuda 0.0427 0.0363 6.8910 stable
proposed 0.0345 0.0233 4.4710 stable
Oustaloup 0.0272 0.0378 6.3096 stable
refined Oustaloup 0.0942 0.0916 10.0820 stable
s0.4
Matsuda 0.0511 0.0400 13.4464 stable
proposed 0.0433 0.0263 7.4974 stable
Oustaloup 0.0293 0.0319 10.0000 stable
refined Oustaloup 0.0881 0.0761 18.9737 stable
s0.5
Matsuda 0.0546 0.0398 26.9262 stable
proposed 0.0476 0.0270 12.7730 stable
Oustaloup 0.0314 0.0276 15.8489 stable
refined Oustaloup 0.0794 0.0617 37.1951 stable
s0.6
Matsuda 0.0548 0.0374 56.1234 stable
proposed 0.0488 0.0265 22.0811 stable
Oustaloup 0.0330 0.0256 25.1189 stable
refined Oustaloup 0.0703 0.0496 77.7765 stable
s0.7
Matsuda 0.0529 0.0340 124.8754 stable
proposed 0.0483 0.0245 38.8116 stable
Oustaloup 0.0338 0.0251 39.8107 stable
refined Oustaloup 0.0618 0.0402 182.9633 stable
s0.8
Matsuda 0.0499 0.0305 313.3221 stable
proposed 0.0468 0.0222 74.6020 stable
Oustaloup 0.0340 0.0205 63.0957 stable
refined Oustaloup 0.0543 0.0330 573.8763 stable
s0.9
Matsuda 0.0466 0.0273 1052.5 stable
proposed 0.0451 0.0249 150.9561 stable

The Bode plots of the proposed approach in

comparison with the Oustaloup, refined Oustaloup and 20 Exact
Oustaloup
Matsuda techniques are presented in Fig. 7. From the
Magnitude (dB)

10 Refined Oustaloup
figure, it can be observed that the proposed approach is Matsuda
Proposed Method
0
more accurate within the desired frequency range than for
the other methods. -10

To evaluate the performance of the proposed -20

10-3 10-2 10-1 100 101 102 103
approach, the step responses of all the compared
techniques is shown in Fig. 8, while the numerical 40

assessment of the figure is given in Table 4. In the figure, 30

Phase (deg)

the exact time response of the system G(s) obtained using 20

(25) is
0.26 × Γ(0.26)
10

g(t) = . (33)
t0.26 0
10-3 10-2 10-1 100 101 102 103
Frequency (rad/sec)
From both the figure and the table, it can be observed
that, for the time period between 0 and 125 seconds, Fig. 7. Bode plots of the fractional-order differentiator (G(s))
the Oustaloup technique has the least error of 0.0315 for different methods.
while, for the time period from 0 to 250 seconds, the

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
Frequency response based curve fitting approximation of fractional-order PID controllers 319

Table 4. Numerical and stability analysis of G(s).

Technique t1 ∈ (0, 125) t2 ∈ (0, 250) H∞ -norm Stability
Oustaloup 0.0315 0.0213 6.0256 stable
refined Oustaloup 0.0493 0.0439 7.9227 stable
Matsuda 0.0363 0.0266 8.2832 stable
proposed 0.0368 0.0212 7.1467 stable

1 50 Exact
Oustaloup
Exact

Magnitude (dB)
Refined Oustaloup
0.9 Oustaloup Matsuda
Refined Oustaloup Proposed Method
0.8 Matsuda 0
Proposed Method

0.7
Amplitude

-50
0.6
10-3 10-2 10-1 100 101 102 103

0.5
0

0.4
Phase (deg)
-20
0.3
-40
0.2
-60
0.1
0 50 100 150 200 250 10-3 10-2 10-1 100 101 102 103
Time (s) Frequency (rad/sec)

Fig. 8. Step responses of the fractional-order differentiator Fig. 9. Bode plots of the fractional-order integrator (P (s)) for
(G(s)) for different methods. different methods.

proposed approach has the least error of 0.0212. This is in comparison with the Oustaloup, refined Oustaloup and
an indication that, for longer time periods, the proposed Matsuda approaches are presented in Fig. 9. From the
approach yields a better approximation than the other figure, it can be observed that, compared with the other
approaches. Furthermore, the stability analysis of G(s) three techniques, the proposed method is more accurate.
for different methods given in Table 4 shows that all the Furthermore, in the time domain, the step response of
approximation techniques are stable with a finite value of all the compared approaches is shown in Fig. 10 while
the H∞ -norm. the numerical assessment is given in Table 5. It can
be observed from both the response and the table that,
for the longest time range t ∈ (0, 250), the proposed
4.2. Fractional-order integrator. In this example, the
approach has the least error of 0.1146 while, for the
proposed approach is demonstrated for a fractional-order
shorter time range t ∈ (0, 125), the Matsuda technique
integrator given by
has the least error of 0.1280. This indicates that the
1 proposed approach produces a better approximation for
P (s) = . (34) longer time periods than the other approaches. This
s0.6
is in agreement with the case of the fractional-order
Here, a similar comparison is made as to the differentiator given in Section 4.1. Furthermore, the
differentiator example of Section 4.1. Hence, the stability analysis of various methods also given in Table
approximated transfer function model using the proposed 5 shows that the approximation transfer functions of the
approach is Oustaloup, Matsuda and proposed approach are stable,
with the proposed technique having the least H∞ -norm
s5 + 2134s4 + 2.094 × 105 s3 of 37.7568. From the table, it can also be seen that
+ 1.582 × 106 s2 + 1.3 × 106 s + 8.884 × 104 the approximated transfer function using the refined
P (s) ≈ . Oustaloup is unstable.
148.8s5 + 4.646 × 104 s4 + 9.294 × 105 s3
+ 1.828 × 106 s2 + 3.731 × 105 s + 2353
(35) 4.3. Fractional-order PID controller (PIλ Dμ ). In
while those using other methods are given in Appendix this example, the proposed approach is demonstrated
B. The frequency response plots of the proposed approach for the PIλ Dμ controller, which consists of both the

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
 320 K. Bingi et al.

Table 5. Numerical and stability analysis of P (s).

Technique t1 ∈ (0, 125) t2 ∈ (0, 250) H∞ -norm Stability
Oustaloup 0.2531 1.0345 63.0957 stable
refined Oustaloup 1.2426 2.1651 inf unstable
Matsuda 0.1280 0.3347 156.4489 stable
proposed 0.8510 0.1146 37.7568 stable

40 50
Exact

Magnitude (dB)
Oustaloup 40
35
Refined Oustaloup
Matsuda 30
30 Proposed Method
20
25
Amplitude

10
10-3 10-2 10-1 100 101 102 103
20

50
15

Phase (deg) 0
10 Exact
Oustaloup
5 Refined Oustaloup
-50 Matsuda
Proposed Method
0
0 50 100 150 200 250 10-3 10-2 10-1 100 101 102 103
Time (s) Frequency (rad/sec)

Fig. 10. Step responses of the fractional-order integrator (P (s)) Fig. 11. Bode plots of the fractional-order PID controller
for different methods. (C(s)) for different methods.

fractional-order differentiator and the integrator given in transfer functions, the overall approximation of C(s) is
Sections 4.1 and 4.2, respectively. The transfer function done by combining this integral with differential terms,
of the PIλ Dμ controller used for demonstration is which will lead to the transfer function of C(s) given
in (37).
1
C(s) = 5 + + 2s0.5 , (36) Therefore, the frequency plot of the approximated
s0.8
C(s) using the proposed approach in comparison with
1.64 × 105 s10 + 1.122 × 108 s9 the Oustaloup, refined Oustaloup and Matsuda techniques
is presented in Fig. 11, while the stability analysis of
+ 2.207 × 1010 s8 + 9.906 × 1011 s7
the figure is given in Table 6. From both the figure
+ 1.546 × 1013 s6 + 7.506 × 1013 s5 and the table, it can be observed that, compared with
+ 1.384 × 1014 s4 + 9.075 × 1013 s3 the other three techniques, the proposed method is more
accurate. Furthermore, the stability analysis shows that
+ 2.321 × 1013 s2 + 1.953 × 1012 s the approximation transfer functions of the Oustaloup,
+ 5.153 × 1010 Matsuda and proposed approach are stable and that of the
C(s) ≈ . (37)
1379s10 + 2.47 × 106 s9 refined Oustaloup technique is unstable. From the table, it
can also be seen that the proposed technique has the least
+ 8.25 × 108 s8 + 6.329 × 1010 s7 H∞ -norm of 133.1226.
+ 1.405 × 1012 s6 + 8.811 × 1012 s5 The result obtained here shows an improvement
+ 1.842 × 1013 s4 + 1.202 × 1013 s3 regarding the separate integrator and differentiator cases.
This indicates that combining the two using the proposed
+ 2.418 × 1012 s2 + 1.009 × 1011 s
approach yields an overall better result. This will, in turn,
+ 3.871 × 108 lead to an overall improvement in system performance.
For this example, approximation will be performed
in two stages. First, the fractional-order differentiator
term (s0.5 ) and the fractional-order integrator term 4.4. Fractional-order low-pass filter. Apart from
(1/s0.8 ) will be approximated using the approximation fractional-order controllers, the proposed approach can
table given in Table 2. Then, using the approximated also be used to approximate the other fractional-order

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
Frequency response based curve fitting approximation of fractional-order PID controllers 321

FOTF in polynomial form with fractional powers is

Table 6. Stability analysis of C(s) for various methods.
Technique H∞ -norm Stability s+1
R(s) = . (40)
Oustaloup 256.2519 stable 10s3.2 + 185s2.5 + 288s0.7 + 1
refined Oustaloup inf unstable
Matsuda 1.2269×103 stable Like in the previous case of F (s), approximation will
proposed 133.1226 stable be performed in three stages. First, based on (30), the
fractional-order derivatives s3.2 and s2.5 will be factorized
as s3 × s0.2 and s2 × s0.5 , respectively. Then s0.2 , s0.5
Table 7. Stability analysis of F (s) for various methods. and s0.7 will be approximated using Table 2. Finally, by
H-norms substituting all these in (40), the overall approximated
Technique Stability transfer function of R(s) is determined. Thus, the
H2 -norm H∞ -norm
Oustaloup 0.9773 1.0273 stable approximated transfer function of R(s) is
refined Oustaloup 0.9841 1.0276 stable
Matsuda 0.9772 1.0275 stable
proposed 0.9784 1.0267 stable s16 + 5206s15 + 7.573 × 106 s14
+ 3.565 × 109 s13 + 5.807 × 1011 s12
systems such as the fractional-order low-pass filter + 3.69 × 1013 s11 + 9.463 × 1014 s10
+ 1.085 × 1016 s9 + 5.936 × 1016 s8
1
F (s) =
0.5s1.15 + 1
. (38) + 1.626 × 1017 s7 + 2.361 × 1017 s6
+ 1.854 × 1017 s5 + 7.647 × 1016 s4
The details of the filter transfer function are available
in the work of Kishore et al. (2017). Consequently, + 1.53 × 1016 s3 + 1.359 × 1015 s2
the approximated transfer function using the proposed + 4.618 × 1013 s + 4.001 × 1011
approach is R(s) ≈ .
44.31s18 + 2.31 × 105 s17 + 3.299 × 108 s16
s5 + 387.7s4 + 6767s3 + 1.119 × 104 s2 + 1.355 × 1011 s15 + 2.312 × 1013 s14
+ 2004s + 19.27 + 1.738 × 1015 s13 + 5.26 × 1016 s12
F (s) ≈ 3.
1.504s6 + 355.3s5 + 4579s4 + 1.178 × 104 s + 7.555 × 1017 s11 + 5.299 × 1018 s10
+ 1.181 × 104 s2 + 2008s + 19.27 + 1.788 × 1019 s9 + 3.458 × 1019 s8
(39) + 4.593 × 1019 s7 + 4.372 × 1019 s6
Here, the approximation of F (s) will be performed
in three stages. First, based on (30), the fractional-order + 2.484 × 1019 s5 + 6.997 × 1018 s4
derivative s1.15 is divided into s × s0.15 . Then, s0.15 is + 8.796 × 1017 s3 + 3.93 × 1016 s2
approximated using the proposed approach. Finally, by
+ 5.396 × 1014 s + 2.125 × 1012
substituting this in (38), the overall approximated transfer (41)
function of F (s) is determined. The frequency plots of R(s) using the proposed approach
The frequency plots of the approximated filter using in comparison with the Oustaloup, refined the Oustaloup
the proposed approach in comparison with the Oustaloup, and Matsuda techniques are presented in Fig. 13, while
refined Oustaloup and Matsuda approaches are presented the stability analysis related to the figure is given
in Fig. 12, while the stability analysis related to the figure in Table 8. From the results, it can be seen that
is given in Table 7. From the results, it can be seen that, all the compared techniques have produced the best
just as in the case of a PIλ Dμ controller, the proposed approximation. Furthermore, the stability analysis from
method is more accurate compared with the other three the table also shows that all the approximated transfer
techniques. Furthermore, the stability analysis also shows functions are stable with the proposed approach having
that all the compared techniques performed better and the least H2 and H∞ -norms of 0.0138 and 0.1883,
stable with H2 -and H∞ -norms of around 0.98 and 1.02 respectively.
respectively. To further evaluate the performance of the proposed
approach, comparison with the power series expansion
4.5. Fractional-order transfer function (FOTF). To technique is made. An example is a finite impulse
further demonstrate the effectiveness of the proposed response (FIR) filter for a discretized fractional-order
approach, a fractional-order transfer function (FOTF) differentiator based on PSE, given in Appendix C.
reported by Khanra et al. (2011; 2013) is considered. The Thus, the Bode plot of the proposed approach in

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
 322 K. Bingi et al.

Table 8. Stability analysis of R(s) for various methods.

0 H-norms
Magnitude (dB)

Technique Stability
H2 -norm H∞ -norm
-50 Oustaloup 0.0141 0.3042 stable
Refined Oustaloup 0.0225 1.2395 stable
-100 Matsuda 0.0164 0.5885 stable
10-4 10-2 100 102 104
Proposed 0.0138 0.1883 stable

0 Exact
Oustaloup
Phase (deg)

Refined Oustaloup Table 9. Stability analysis of R(s) with PSE-FIR for various
Matsuda
-50
Proposed Method values of n.
H-norms
-100 Technique Order Stability
H2 -norm H∞ -norm
-4 -2 0 2 4
10 10 10 10 10 10 inf inf unstable
Frequency (rad/sec)
20 0.0024 7.0452 stable
Fig. 12. Bode plots of the fractional-order low-pass filter (F (s)) PSE-FIR 50 0.0002 0.0508 stable
for different methods. 100 0.0002 0.0301 stable
500 0.0002 0.0176 stable
Proposed 5 0.0138 0.1883 stable
0
Magnitude (dB)

-50

-100 comparison with the FIR filter for orders of truncation

-150
or approximation of 10, 20, 50, 100 and 500 is shown
in Fig. 14. The stability analysis of the figure is given in
-200
10-4 10-2 100 102 104 Table 9. From the figure and the table, it can be seen that
0
the proposed approach performs better compared with the
Exact
Oustaloup
PSE based FIR filter. Observing the responses, it can also
-50
Phase (deg)

Refined Oustaloup be seen that, for orders less than 10, the transfer function
Matsuda
-100
Proposed Method using PSE-FIR is unstable and, for an order greater
-150 then 50, PSE-FIR approaches the exact response, thereby
-200
generating a very high integer-order transfer function.
10-4 10-2 100 102 104
Frequency (rad/sec)

5. Conclusion
Fig. 13. Bode plots of FOTF (R(s)) for different methods.
In this paper, a simple curve fitting approximation
technique for the fractional-order differintegral operator
using the frequency response was proposed. With the
approach, an approximation table for the fractional-order
differentiator was obtained. The table can be used
directly to generate approximated transfer functions
of fractional-order based controllers and systems.
Results from the simulation study show that the
proposed approach produced better approximation of the
fractional-order parameters within the desired frequency
range when compared with the Oustaloup, refined
Oustaloup and Matsuda approximations. Furthermore,
time domain analysis of the results shows that the
proposed approach gives better approximation for longer
time periods than the Oustaloup, refined Oustaloup and
Matsuda approximations. The stability analysis in terms
of the H2 and H∞ -norms also confirms that the proposed
approach is better and stable.
As part of future studies, an attempt will be made to
Fig. 14. Bode plots of the proposed approach compared with implement a fractional-order controller designed using the
PSE-FIR for T = 0.001 and various values of n. proposed approach on a real-time plant.

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
Frequency response based curve fitting approximation of fractional-order PID controllers 323

Acknowledgment Khanra, M., Pal, J. and Biswas, K. (2013). Rational

approximation and analog realization of fractional order
This work was supported by Universiti Teknologi transfer function with multiple fractional powered terms,
PETRONAS through the Yayasan UTP Fundamental Asian Journal of Control 15(3): 723–735.
Research Grant 0153AA-H16. Kishore, B., Ibrahim, R., Karsiti, M.N. and Hassan, S.M. (2017).
Fractional-order filter design for set-point weighted PID
References controlled unstable systems, International Journal of Me-
chanical & Mechatronics Engineering 17(5): 173–179.
Atherton, D.P., Tan, N. and Yüce, A. (2014). Methods for
computing the time response of fractional-order systems, Kishore, B., Ibrahim, R., Karsiti, M.N. and Hassan, S.M.
IET Control Theory & Applications 9(6): 817–830. (2018). Fractional order set-point weighted PID controller
for pH neutralization process using accelerated PSO
Balas, G., Chiang, R., Packard, A. and Safonov, M. (2007). Ro-
algorithm, Arabian Journal for Science and Engineering
bust Control Toolbox 3: User’ Guide, MathWorks, Natick,
43(6): 2687–2701.
MA.
Krajewski, W. and Viaro, U. (2011). On the rational
Bingi, K., Ibrahim, R., Karsiti, M.N., Hassan, S.M. and
approximation of fractional order systems, 16th Inter-
Harindran, V.R. (2018a). A comparative study of 2DOF
national Conference on Methods and Models in Au-
PID and 2DOF fractional order PID controllers on a
tomation and Robotics (MMAR), Mi˛edzyzdroje, Poland,
class of unstable systems, Archives of Control Sciences
pp. 132–136.
28(4): 635–682.
Krajewski, W. and Viaro, U. (2014). A method for the
Bingi, K., Ibrahim, R., Karsiti, M.N., Hassan, S.M. and
integer-order approximation of fractional-order systems,
Harindran, V.R. (2018b). Real-time control of pressure
Journal of the Franklin Institute 351(1): 555–564.
plant using 2DOF fractional-order PID controller, Arabian
Journal for Science and Engineering 44(3): 2091–2102. Krishna, B. (2011). Studies on fractional order differentiators
and integrators: A survey, Signal Processing
Caponetto, R. (2010). Fractional Order Systems: Modeling and 91(3): 386–426.
Control Applications, World Scientific, Singapore.
Li, Z., Liu, L., Dehghan, S., Chen, Y. and Xue, D. (2017).
Das, S. (2011). Functional Fractional Calculus, Springer, A review and evaluation of numerical tools for fractional
Berlin/Heidelberg. calculus and fractional order controls, International Jour-
de Oliveira Valério, D.P.M. (2005). Fractional Robust System nal of Control 90(6): 1165–1181.
Control, PhD thesis, Universidade Técnica de Lisboa, Liang, S., Peng, C., Liao, Z. and Wang, Y. (2014).
Lisboa. State space approximation for general fractional order
Deniz, F.N., Alagoz, B.B., Tan, N. and Atherton, D.P. (2016). dynamic systems, International Journal of Systems Science
An integer order approximation method based on stability 45(10): 2203–2212.
boundary locus for fractional order derivative/integrator Meng, L. and Xue, D. (2012). A new approximation algorithm
operators, ISA Transactions 62: 154–163. of fractional order system models based optimization,
Djouambi, A., Charef, A. and Besançon, A.V. (2007). Optimal Journal of Dynamic Systems, Measurement, and Control
approximation, simulation and analog realization of the 134(4): 044504.
fundamental fractional order transfer function, Interna- Merrikh-Bayat, F. (2012). Rules for selecting the parameters
tional Journal of Applied Mathematics and Computer Sci- of Oustaloup recursive approximation for the simulation
ence 17(4): 455–462, DOI: 10.2478/v10006-007-0037-9. of linear feedback systems containing PIλDμ controller,
Du, B., Wei, Y., Liang, S. and Wang, Y. (2017). Rational Communications in Nonlinear Science and Numerical Sim-
approximation of fractional order systems by vector fitting ulation 17(4): 1852–1861.
method, International Journal of Control, Automation and Mitkowski, W. and Oprzedkiewicz, K. (2016). An estimation
Systems 15(1): 186–195. of accuracy of Charef approximation, in S. Domek
Joice Nirmala, R. and Balachandran, K. (2017). The and P. Dworak (Eds.), Theoretical Developments and
controllability of nonlinear implicit fractional delay Applications of Non-Integer Order Systems, Springer,
dynamical systems, International Journal of Applied Berlin/Heidelberg, pp. 71–80.
Mathematics and Computer Science 27(3): 501–513, DOI: Monje, C.A., Chen, Y., Vinagre, B.M., Xue, D. and Feliu-Batlle,
10.1515/amcs-2017-0035. V. (2010). Fractional-Order Systems and Controls: Fun-
Kaczorek, T. (2018). Decentralized stabilization of fractional damentals and Applications, Springer, Berlin/Heidelberg.
positive descriptor continuous-time linear systems, Inter- Oprzedkiewicz, K. (2014). Approximation method for a
national Journal of Applied Mathematics and Computer fractional order transfer function with zero and pole,
Science 28(1): 135–140, DOI: 10.2478/amcs-2018-0010. Archives of Control Sciences 24(4): 447–463.
Khanra, M., Pal, J. and Biswas, K. (2011). Rational Oustaloup, A., Levron, F., Mathieu, B. and Nanot, F.M. (2000).
approximation and analog realization of fractional order Frequency-band complex noninteger differentiator:
differentiator, 2011 International Conference on Process characterization and synthesis, IEEE Transactions on Cir-
Automation, Control and Computing (PACC), Coimbatore, cuits and Systems I: Fundamental Theory and Applications
India, pp. 1–6. 47(1): 25–39.

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
 324 K. Bingi et al.

Pachauri, N., Singh, V. and Rani, A. (2018). Yüce, A., Deniz, F.N. and Tan, N. (2017). A new integer
Two degrees-of-freedom fractional-order order approximation table for fractional order derivative
proportional-integral-derivative-based temperature control operators, IFAC-PapersOnLine 50(1): 9736–9741.
of fermentation process, Journal of Dynamic Systems,
Measurement, and Control 140(7): 071006.
Kishore Bingi received the BTech degree (Hons)
Petráš, I. (2011a). Fractional derivatives, fractional integrals, in electrical and electronics engineering from
and fractional differential equations in Matlab, in A. Assi Bapatla Engineering College, Bapatla, India, in
(Ed.), Engineering Education and Research Using MAT- 2012, and the MTech degree (Hons) in instru-
mentation and control systems from the National
LAB, InTech, London, pp. 239–264.
Institute of Technology Calicut, Kerala, India, in
Petráš, I. (2011b). Fractional-Order Nonlinear Systems: Model- 2014. He is currently pursuing the PhD degree
ing, Analysis and Simulation, Springer, Berlin/Heidelberg. with the Electrical and Electronic Engineering
Department, Universiti Teknologi PETRONAS,
Poinot, T. and Trigeassou, J.-C. (2003). A method for modelling Perak, Malaysia. He was with Tata Consultancy
and simulation of fractional systems, Signal processing Service as an assistant systems engineer from 2015 to 2016. His current
83(11): 2319–2333. research interests include process modeling, control, and optimization.

Shah, P. and Agashe, S. (2016). Review of fractional PID

controller, Mechatronics 38: 29–41. Rosdiazli Ibrahim received the BEng degree
in electrical engineering from Universiti Putra
Sheng, H., Chen, Y. and Qiu, T. (2011). Fractional Processes Malaysia, Kembangan, Malaysia, in 1996, the
and Fractional-Order Signal Processing: Techniques and MSc degree in automation and control from
Applications, Springer, Berlin/Heidelberg. Newcastle University, UK, in 2000, and the PhD
degree in electrical and electronic engineering
Shi, G. (2016). On the nonconvergence of the vector fitting from the University of Glasgow, UK, in 2008. He
algorithm, IEEE Transactions on Circuits and Systems II: is an associate professor with the Department of
Express Briefs 63(8): 718–722. Electrical and Electronics Engineering at Univer-
siti Teknologi PETRONAS (UTP), Seri Iskandar,
Tepljakov, A., Petlenkov, E. and Belikov, J. (2012). Application Perak, Malaysia. He is currently the dean at the Centre for Graduate
of Newton’s method to analog and digital realization of Studies, UTP. His present research interests include intelligent control
fractional-order controllers, International Journal of Mi- and non-linear multi-variable process modeling for control application.
croelectronics and Computer Science 2(2): 45–52.
Valério, D., Trujillo, J.J., Rivero, M., Machado, J.T. and Mohd Noh Karsiti received the BEng degree in
Baleanu, D. (2013). Fractional calculus: A survey of useful electrical engineering from California State Uni-
versity, Long Beach, CA, in 1985, the MSc de-
formulas, The European Physical Journal Special Topics
gree in electrical engineering (control) from Cal-
222(8): 1827–1846. ifornia State University in 1987, and the PhD de-
Vinagre, B., Podlubny, I., Hernandez, A. and Feliu, V. (2000). gree in electrical and computer engineering (hi-
erarchical control) from the University of Cali-
Some approximations of fractional order operators used in fornia, Irvine, CA, in 1991. He joined Univer-
control theory and applications, Fractional Calculus and siti Teknologi PETRONAS, Perak, Malaysia, in
Applied Analysis 3(3): 231–248. 1999, where he is currently an associate profes-
sor. His present research interests include advanced process control,
Wei, Y., Gao, Q., Peng, C. and Wang, Y. (2014a). A rational robotics systems manipulator design and control, control of highly non-
approximate method to fractional order systems, Inter- linear systems and automatic compensation of control systems.
national Journal of Control, Automation and Systems
12(6): 1180–1186.
Hassan Sabo Miya received the BEng de-
Wei, Y., Gao, Q., Peng, C. and Wang, Y. (2014b). A rational gree (Hons) in electrical and electronic engi-
approximate method to fractional order systems, Inter- neering from Abubakar Tafawa Balewa Univer-
national Journal of Control, Automation and Systems sity, Bauchi, Nigeria, in 2008, the MSc degree
12(6): 1180–1186. (Hons) in control systems from the University of
Sheffield, UK, in 2011, and the PhD degree at
Xue, D. (2017). Fractional-order Control Systems: Fundamen- the Electrical and Electronic Engineering Depart-
tals and Numerical Implementations, Walter de Gruyter ment, Universiti Teknologi PETRONAS, Perak,
GmbH, Berlin. Malaysia. He is currently a lecturer in the De-
partment of Electrical and Electronic Engineer-
Xue, D., Chen, Y. and Attherton, D.P. (2007). Linear Feed- ing, Abubakar Tafawa Balewa University, Bauchi, Nigeria. His present
back Control: Analysis and Design with MATLAB, SIAM, research interests include wireless networked control systems, intelligent
Philadelphia, PA. control, soft computing, and optimization.

Xue, D., Zhao, C. and Chen, Y. (2006). A modified

approximation method of fractional order system, Pro-
ceedings of the 2006 IEEE International Conference
on Mechatronics and Automation, Luoyang, China
pp. 1043–1048.

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
Frequency response based curve fitting approximation of fractional-order PID controllers 325
13 4 12
Vivekananda Rajah Harindran received the 7.923s + 1.029 × 10 s
BSc degree in electrical and electronics engineer-
ing from Huddersfield University, UK, in 1984
+ 1.664 × 106 s11 + 1.832 × 107 s10
and the MSc degree in technology management + 1.279 × 107 s9 + 5.634 × 105 s8
from Staffordshire University, UK, in 2010. He
joined PETRONAS Group Technical Solutions in + 1566s7 + 0.2747s6
1992, where he is currently the custodian engi-
neer in the Division of Instrumentation and Con- + 3.039 × 10−6 s5 + 2.122 × 10−12 s4
trol. He has 27 years’ experience in petrochem- + 9.347 × 10−20 s3 + 2.588 × 10−28 s2
ical, oil and gas industries, with skills in project
design, improving plant production, reliability and availability. His cur- + 4.252 × 10−38 s
rent research interests include the design and implementation of instru- GR (s) ≈ ,
mentation systems and control. s13 + 1885s12 + 5.853 × 105 s11
+ 1.315 × 107 s10 + 1.883 × 107 s9
+ 1.705 × 106 s8 + 1.01 × 104 s7
Appendix A + 5.758s6 + 0.0008956s5

MATLAB code for the proposed curve fitting + 1.864 × 10−8 s4 + 2.655 × 10−14 s3
based approximation + 2.398 × 10−21 s2 + 1.362 × 10−29 s
% Curve Fitting Approach for sγ % + 4.589 × 10−39
(B2)
function Gp=curveFitting(gam,N,wl,wh)
s=tf(’s’); 8.283s5 + 2347s4 + 2.872 × 104 s3
FRD=frd(s,(logspace(log10(wl),... + 1.982 × 104 s2 + 749.1s + 1
log10(wh)))); GM (s) ≈ 5 , (B3)
s + 749.1s4 + 1.982 × 104 s3
FRD.ResponseData=FRD.ResponseData.^gam;
+ 2.872 × 104 s2 + 2347s + 8.283
Gp=fitfrd(FRD,N);
[num,den]=ss2tf(Gp.A,Gp.B,Gp.C,Gp.D);
Gp=tf(num,den); end
s5 + 614.2s4 + 2.239 × 104 s3
To compute approximation with the help of
+ 5.129e04s2 + 7384s + 63.1
above-proposed curve fitting approximation, the MATLAB PO (s) ≈ , (B4)
Robust Control Toolbox is required. The user 63.1s5 + 7384s4 + 5.129 × 104 s3
guide of the toolbox is available in the work of Balas et al. + 2.239e04s2 + 614.2s + 1
(2007).

s13 + 3392s12 + 1.728 × 106 s11

Appendix B
+ 6.224 × 107 s10 + 1.426 × 108 s9
Results from other approximation methods + 2.068 × 107 s8 + 1.99 × 105 s7
+ 209.6s6 + 0.0623s5
The approximated transfer function of the fractional-order + 2.129 × 10−6 s4 + 4.857 × 10−12 s3
differentiator given in Section 4.1 using the Oustaloup
+ 7.017 × 10−19 s2 + 6.374 × 10−27 s
GO (s), refined Oustaloup GR (s) and Matsuda GM (s)
approximation techniques is presented in (B1), (B2) and + 3.436 × 10−36
PR (s) ≈ ,
(B3), respectively. Similarly, for the fractional-order 148.1s13 + 1.819 × 105 s12
integrator given in Section 4.2, the approximated transfer
+ 1.938 × 107 s11 + 1.338 × 108 s10
functions using the Oustaloup PO (s), refined Oustaloup
PR (s) and Matsuda PM (s) approximation techniques are + 5.84 × 107 s9 + 1.608 × 106 s8
+ 2795s7 + 0.3065s6
5 4 4 3
6.026s + 1128s + 1.253 × 10 s + 2.12 × 10−6 s5 + 9.255 × 10−13 s4
2
GO (s) ≈
+ 8750s + 384s + 1
, (B1) + 2.548 × 10−20 s3 + 4.411 × 10−29 s2
5 4 3
s + 384s + 8750s + 4.532 × 10−39 s
4 2
+ 1.253 × 10 s + 1128s + 6.026 (B5)

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
 326 K. Bingi et al.

s5 + 1807s4 + 8.141 × 104 s3 where

+ 1.922 × 105 s2 + 2.565 × 104 s • n is the order of approximation or truncation,
+ 156.4 • T is the sampling period,
PM (s) ≈ . (B6)
156.4s5 + 2.565 × 104 s4 • R is the polynomial in the variable z −1 of order, n
5 3 4 2
+ 1.922 × 10 s + 8.141 × 10 s and
 γ
+ 1807s + 1 • PSE{ 1 − z −1 } denotes the power series
 γ
expansion of the function 1 − z −1 .
Appendix C From (C1), it can be noted that
γ the approximated
Approximation of fractional-order differ- power series expansion of 1 − z −1 in the form of the
entiator using finite impulse response (FIR) FIR filter has only zeros.
based on PSE
According to Petráš (2011b; 2011a) and Caponetto Received: 17 June 2018
(2010), the approximation of the discretized Revised: 31 October 2018
fractional-order differentiator in the form of the FIR Re-revised: 5 December 2018
filter based on PSE is as follows: Accepted: 8 January 2019

1
γ
 γ 
D
0 T
γ
X(z) = P SE 1 − z −1 n
T (C1)
−γ −1
≈ T Rn (z ),

Brought to you by | Universiti Teknologi Petronas (UTP)

Authenticated
Download Date | 7/9/19 9:03 AM
View publication stats