Академический Документы
Профессиональный Документы
Культура Документы
Introduction
714
Fractional-order integrator
Rational approximation
1
sm
KI
1 s=vc m
KI
KI vm
1
c
m GI s
m
s
s=vc
sm
with KI (1/vm
c ); vc is the 23m dB frequency corner of
the FPP,pwhich is obtained from the low frequency vL , as
vc vL (10(1/10m) 2 1), where 1 is the maximum permitted error between the slopes of the fractional-order integrator of (1) and the FPP of (2) in the given frequency band
of interest (vL , vH).
In order to represent the FPP of (2), and consequently the
fractional-order integrator, by a linear time-invariant system
model, it is necessary to approximate its irrational transfer
function by a rational one. The method of approximation
consists of approximating the 220m dB/dec slope on the
Bode plot of the FPP by a number of alternate slopes of
220 and 0 dB/dec corresponding to alternate poles and
zeros on the negative real axis of the s-plane such that
IEE Proc.-Control Theory Appl., Vol. 153, No. 6, November 2006
K
4
Gs
I QN
1 s=vc m
i0 1 s=pi
Using a simple graphical method that began with a specified error y in decibels and frequency band vmax , which can
be 100vH , the parameters a, b, p0 , z0 and N can be calculated as [11]
-10
-20
-30
-40
-50
-60
-70
a 10y=101m ;
z0 ap0
b 10y=10m ; p0 vc 10y=20m ;
logvmax =p0
and N Integer
1
logab
The poles (pi) and the zeros (zi) of (5) are found to be in a
geometric progression form. They can then be derived from
the above parameters as pi p0(ab)i for i 0, 1, . . . , N and
zi z0(ab)i for i 0, 1, . . . , N 2 1. Hence, the fractionalorder integrator can be approximated by a rational function
in a given frequency band of interest as
QN1
i
1
KI
i0 1 s=z0 ab
K
5
GI s m
Q
I
N
i
s
1 s=vc m
i0 1 s=p0 ab
2.2
Illustrative example
1
s0:65
zi 4:33802:7514i
for i 0; 1; . . . ; 15
-90
-100
1
10
10
10
10
10
10
10
where the ki are the residues of the poles that can be easily
calculated as [11]
QN 1
1 abij =a
ki KI QJN0
for i 0; 1; . . . ; N 7
ij
j0;i=j 1 ab
It can be seen that (6) is analogous to the impedance Z(s)
of the N 1 parallel RC cells connected in series as shown
in Fig. 3, which is given as
N
X
Ri
8
Zs
1 sRi Ci
i0
Hence, from (6) and (8), and for i 0, 1, . . . , N
R i ki
and
Ri Ci
1
abi p0
0:6629
1 s=1:88210:65
pi 2:24692:7514i
-80
Ri k i
and Ci
1
abi p0 ki
for i 0; 1; . . . ; N
-50
and
-55
2.3
N
X
i0
ki
1 s=abi p0
-65
1
10
10
10
10
10
10
10
Fig. 2 Phase Bode plot of the integrator s20.65 and of its rational
function approximation
715
Fractional-order differentiator
3.1
3.2
Rational approximation
10
and
The zeros (zi) and the poles (pi) of (13) can then be derived
from the above parameters as zi z0(ab)i for i 0, 1, . . . , N
and pi p0(ab)i for i 0, 1, . . . , N.
Hence, the fractional-order differentiator can be approximated by a rational function in a given frequency band of
practical interest by
QN
s m
1 s=z0 abi
KD QNi0
GD s sm KD 1
i
vc
i0 1 s=p0 ab
14
b 10y=10m ; z0 vc 10y=20m
logvmax =z0
N Integer
1
logab
Illustrative example
and
pi 0:50792:7514i
for i 0; 1; . . . ; 15
Figs. 4 and 5 show the Bode plots of the fractional-order differentiator transfer function and its rational approximation.
It can be easily observed that they are quite overlapping
over the frequency band of interest.
3.3
By decomposition of the rational function of (14), it is possible to represent the fractional-order differentiator by a
45
40
35
30
25
20
15
10
5
0
0
10
10
10
10
10
10
10
32
31
4.1
30
29
28
27
26
25
24
0
10
10
10
10
10
10
10
N
X
i0
ki s
1 s=abi p0
15
with G0 KD and
ki
QN
ij
KD
J 0 1 aab
Q
i
N
ij
p0 ab j0;i=j 1 ab
for i 0; 1; . . . ; N
16
Y s
RP i0 1 sRi Ci
17
Ci ki
and
Ri C i
1
abi p0
1
;
G0
Ri
1
abi p0 ki
Ci ki
and
for i 0; 1; . . . ; N
The poles (pIi), the zeros (zIi) and the parameters KI and NI
of the rational function approximation of the fractionalorder integrator can be calculated as described in Section
2.1. Also, the zeros (zDi), the poles ( pDi) and the parameters
KD and ND of the rational function approximation of the
fractional-order differentiator can be calculated as described
in Section 3.1.
Illustrative example
18
where (TI/s) is a first-order integrator, (1/s mI) is a fractionalorder integrator with 0 , mI , 1, (TDs) is a first-order differentiator and (s mD) is a fractional-order differentiator with
0 , mD, 1.
In order to represent the fractional PIlDm controller of
(19) by a linear time-invariant system model, it is necessary
to approximate its irrational transfer function by a rational
one. Hence, in a given frequency band of practical interest
(vL , vH), the fractional-order integrator can be modelled by
an FPP and the fractional-order differentiator by an FPZ. It
has also been shown how the FPP and the FPZ can be
approximated by rational functions. Hence, (20) becomes
!
QNI 1
TI
i0 1 s=zIi
K I QN I
Cs KP
s
i0 1 s=pIi
!
Q ND
i0 1 s=zDi
TD s KD QND
21
i0 1 s=pDi
4.2
0:18
6s1:7
s1:7
90
80
70
60
50
40
Cs 2:4
20
0:18
6s1:35
s1:35
10
0
-10
-2
10
-1
10
10
10
10
4.3
0:18
Cs 2:4
s
(
)
Q14
i
4
i0 1 s=4:6060 10 2:9936
415:9188 Q15
i
4
i0 1 s=2:1379 10 2:9936
(
)
Q15
i
4
i0 1 s=2:1379 10 2:9936
6s 0:0024 Q15
i
4
i0 1 s=4:6060 10 2:9936
Figs. 7 and 8 show the Bode plots of this fractional PIlDm
controller transfer function and its rational approximation.
It can be easily seen that they are quite overlapping over
the frequency band of interest.
It is also interesting to show the effect of the parameters l
and m on the frequency behaviour of the fractional-order
PIlDm controller without any changes in the parameters
KP , TI and TD . To do so, the above numerical example
was used for l m 1.35. Hence, the transfer function
200
150
U s UP s UI s UD s
Es
Es
Es
Es
KP
TI
TD s m
sl
s=z
I
i
KI Qi0
NI
sl
s
i0 1 s=pIi
22
The rational function of the above equation can be decomposed into simple, elementary functions and this integration
element can then be rewritten as
!
X
NI
TI
TI
kIi
23
sl
s
1 s=pIi
i0
where the kIis can be calculated as shown in Section 2.3.
From (23), the integration element can be realised by the
analogue circuit shown in Fig. 11. Hence
100
50
UI s TI ZI2
l
Es
s
ZI1
0
-50
-100
-150
-200
-2
10
-1
10
10
10
10
Fig. 13
ND
X
1
kDi s
YD1 s GD
ZD1
1
s=pi
i0
1
s LI s
ZI1
TI
ND
X
1
sCDi
RD i0 1 sRDi CDi
Z I2
i0
NI
X
k Ii
R Ii
RD
1
;
GD
RDi
1
;
pDi kDi
and
1
C Ii
p Ii k Ii
for i 0; 1; . . . ; NI
Q ND
TD s TD s
KD QNi0
D
i0
1 s=zDi
1 s=pDi
24
ND
X
i0
kDi s
1 s=pDi
!
25
and
for i 0; 1; . . . ; ND
CDi kDi
Cs
U s
TI
RP2 ZI2 ZD2
K P l T D sm
Es
s
RP1 ZI1 ZD1
Conclusion
References
720