Fractional Calculus Applications in Control Systems
Fractional Calculus Applications in Control Systems
Mark Axtell
Veda Incorporated
5200 Springfield Pike, Suite 200
Dayton, OH, 45431
Michael E. Bise
Veda Incorporated
5200 Springfield Pike, Suite 200
Dayton, OH, 45431
where m is some integer (1,2,3,4,5 ...). Typically fm(x) is
referred to as the mth derivative of f(x). The notion of
fractional calculus is to remove the integer restriction on m.
That is, let m be a rational number (fraction) q. Rewriting (2)
in this way yields:
Abstract
Standard control systems can be characterized by type in
the s-domain. Typically these types are of integer order.
This paper explores some of the implications of non-integer
order systems in the s-domian. In order to accomplish this
results from the area of fractional calculus, which defines
mathematics of non-integer order derivative and integration
are utilized.
introduction
In general, fractional calculus analysis techniques have not
been widely incorporated into the engineering sciences.
This is partly because of the conceptually difficult idea of
taking the 1/2 derivative and its notorious lack of
geometrical interpretation, and partly because there are so
few physical applications.
LIf(t)l 3 F(s)
One of the great advantages of the Laplace transform is that
fundamental mathematical operations like convolution,
differentiation and integration, which frequently introduce
computational difficulties in the time domain, are found to
be simple algebraic operations in the sdomain. For this
reason, much of control systems design and analysis is
done exclusively in the s-domain, returning to the time
domain only to verify those system design parameters that
are specified in time( e.g. rise time, peak overshoot, settling
time, etc.). While the s-domain is certainly more analytically
robust, some time domain operations and techniques for
which no Laplace transform exists or is thought to exist,
have remained absent from the control systems
engineering literature. Fractional calculus is one example.
a
,f o r q e o
dx9
R1
= f l (x) ;
g(x) =
So that
&
(6)
Jf(x)dx =
(7)
dd-1)
All of this may seem quite confusing at first but it is actually
quite an intuitively succinct scheme which demonstrates the
inverse relationship between integration and
563
0 1990
IEEE
K n(S-2,)
G(s)H(s)
(10)
=
c =I
9 =0,+1,+2,. . . *
The s4 ODerator,
The system described in (11 ) is characterized by the sq
term in the denominator. Keep in mind that the sign of q
can change (differentiation vs. integration) without
introducing any structural difficulties into the model.
Mathematically s9 is complex and can be expressed as
564
Freauencv ResPonse
The utility of the being able to vary q continuously is
apparent when considering system frequency response. In
this section a conventional Bode analysis will be performed
on a simple system which incorporates a fractional element.
Consider
G(s)
K(s - Z,)(s - z ~
=
S V S -P,)(S - P p )
. . ds - P,)
LG(jw)= L K m + L ( a + j ~ ) - q 9 0 - L ( j % j w )
(18)
(14)
G(s)=
(19)
s9 (s+l)
or equivalently
then
G(j@)=
-
(20)
jd(iW+l)
p=q L s +
Ls-plf
... Ls-pv-
LS-Zl-
... Ls-zw
565
m10
0 -10
-20
-30 -
*YI-
M -70
80-
-90 -1m
hYl"Y(W1
ConclusionlRecommendations
FIGURE 3
Figure 2 shows that as the type of the system ,q, is reduced
from 1 to 0, the effect is to reduce the system gain at a
constant rate, -9w. Figure 3 demonstrates that a fractional
pole can be used to add phase in a linear fashion. In this
figure the bottom trace, -90 degrees at w=O, is the standard
response of a strong pole located at the origin. As q is
reduced, phase is added, and the whole response is shifted
upward. This is an interesting result when contrasted with
conventional lead, lag networks or PID-type controllers
which can have undesirable phase-gain trade offs at certain
frequencies.
1 + T,s)( 1 t T ~ s .) . . R(s)
Here the differintegral operator was introduced in a system theoretic context by making use of a result involving the
Laplace transform. Extensive research is still in required in
order to identify all the impacts that fractional order systems
can have in the controls field.
REFERENCES
1.
D'Azzo, J.J. and C. H. Houpis. Linear Control
Svstem Analvsis and Desian. McGraw-Hill Book Company,
New York, NY, 1975.
T,s)(~+&s)...
(21)
In the limit, s9 goes to zero if q is fractional or integral. Of
course this is dependent on the input R(s) as well. The
following table compares the effects of fractional and
integral s9 on steady state error.
2.
Oldham, K.B. and J. Spanier. The Fractiona!
Calculus. Academic Press, New York, NY, 1974.
3
Churchill, R.V. and J.W. Brown. Comslex Variables
jlnd Asdications. McGraw-Hill Book Company, New York,
NY, 1984.
McDonald, A.C. and H. Lowe. Feedback and Control
Svstems. Reston Publishing Company, Inc., Reston, VA,
1981.
4.
System Type, q
O<q<l
1<q<2
2<q<3
a-Error
Unit step
Ramp
00
Parabolic
00
Unit step
Ramp
Parabolic
00
a-Error
+ KO
Roll
5.
Kreyszig, E. Advanced Enaineerina Mathematics.
Fifth Edition. John Wiley and Sons, New York, NY, 1983.
00
00
Unit step
Ramp
Parabolic
TABLE 1
566