You are on page 1of 14

I Encontro de Jovens Investigadores do LAETA

Instituto Superior Tcnico, Lisboa, 15 e 16 de Novembro 2010

FRACTIONAL AND VARIABLE ORDER


CONTROLLERS
Duarte Valério

IDMEC / IST, TULisbon


Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal
e-mail: duarte.valerio@ist.utl.pt

Key words: Fractional derivatives, Fractional Calculus, Fractional control, Fractional PID,
Crone control, Adaptive control
Abstract. This survey paper presents the basic ideas of fractional derivatives, variable order
derivatives, fractional control and variable order control. A list of applications, and the rationale
for using this type of controllers, is included.

1 Introduction
Given a (well-behaved) function f (t), we learn in Calculus to calculate f 0 (t), f 00 (t), and generically
z
f (n) (t) for all natural numbers n. In Fractional Calculus we learn to generalise d dtf z(t) to any order
z, fractional, irrational or even complex. (The name “Fractional Calculus” is a historical reminder
that fractions were the first target of generalisation attempts.) As one might expect, order 0 is
simply function f itself, and, for negative integer orders z, indefinite integrals are retrieved. These
fractional derivatives (that thus generalise both our well-known, usual natural order derivatives
and natural order integrals) can have a smoothly varying order, and thus time-varying derivatives
dz f (t)
dtz are easily conceived.
This paper addresses the theory of fractional derivatives in section 2, the theory of time-varying
derivatives in section 3, and then applies this to dynamic systems (and in particular to the synthesis
of controllers) in section 4. Section 5 concludes the paper with a list of applications found in the
literature.

2 Fractional derivatives
2.1 Simple examples
From

f (t) = eλt (1)


0 λt
f (t) = λe (2)
00 2 λt
f (t) = λ e (3)
(3) 3 λt
f (t) = λ e (4)
..
.
f (n) (t) = λn eλt (5)
Duarte Valério

we are tempted to write


f (z) (t) = λz eλt (6)
even when z ∈
/ N. Something similar happens with

f (t) = tλ (7)
0 λ−1
f (t) = λt (8)
00 λ−2
f (t) = λ(λ − 1)t (9)
(3) λ−3
f (t) = λ(λ − 1)(λ − 2)t (10)
..
.
n−1
Y
f (n) (t) = λ(λ − 1) . . . (λ − (n − 1))tλ−n = tλ−n (λ − k) (11)
k=0

If λ is integer, the last equality can be written as


λ!
f (n) (t) = tλ−n (12)
(λ − n)!

If it is not, we can use the Gamma function instead of the factorial

Γ(λ + 1) λ−n
f (n) (t) = t (13)
Γ(λ − n + 1)

and so we are tempted to write

Γ(λ + 1) λ−z
f (z) (t) = t (14)
Γ(λ − z + 1)

even when z ∈/ N. Fractional Calculus generalises these results systematically.


(These two examples are taken from [11] and are further elaborated below in section 2.3. Notice
that some restrictions on λ, and in the case of f (t) = tλ also on t, must apply. Surely the reader
can find them out.)

2.2 Integer order derivatives


By definition,
f (t) − f (t − h)
f 0 (t) = lim (15)
h→0 h
and so
f 0 (t) − f 0 (t − h)
f 00 (t) = lim
h→0 h
f (t)−f (t−h)
h − f (t−h)−f
h
(t−2h)
= lim
h→0 h
f (t) − 2f (t − h) + f (t − 2h)
= lim (16)
h→0 h2
Duarte Valério

f 00 (t) − f 00 (t − h)
f (3) (t) = lim
h→0 h
f (t)−2f (t−h)+f (t−2h)
h2 − f (t)−2f (t−h)+f
h2
(t−2h)
= lim
h→0 h
f (t) − 3f (t − h) + 3f (t − 2h) − f (t − 3h)
= lim (17)
h→0 h3
..
.
n  
X n
(−1)k f (t − kh)
k
(n) k=0
f (t) = lim (18)
h→0 hn
(A rigourous proof of the last equality can be found by mathematic induction.) Recall that
combinations of a things, b at a time are defined as
 
a a!
= (19)
b b!(a − b)!
We are tempted to use (18) for orders which are not natural, by redefining combinations of a
things, b at a time using function Γ:
 
a Γ(a + 1)
= (20)
b Γ(b + 1)Γ(a − b + 1)
It can be easily shown (again, using mathematical induction) that this expression is equivalent to
 
a (−1)b Γ(b − a)
= (21)
b Γ(b + 1)Γ(−a)
Remember, however, that Γ(z) is defined for all complex arguments, save for z ∈ Z −0 , since the
function has poles at those points. (In other words, lim Γ(z) = ∞ if n ∈ Z−
0 .) This means that
z→n
the applicability range of (20) and (21) is not exactly the same; actually combinations are better
defined joining both expressions together and adding some points otherwise undefined but where
(20) or (21) converge to 0 to get

Γ(a + 1)

 / Z−
, if a, b, a − b ∈
    Γ(b + 1)Γ(a − b + 1)
a
= (−1)b Γ(b − a) (22)
b  , if a ∈ Z− ∧ b ∈ Z+ 0


 Γ(b + 1)Γ(−a)
0, if (b ∈ Z− ∨ b − a ∈ N) ∧ a ∈ / Z−
 
α
(Cases not included are equal to ∞; or, to be more rigourous, lim = ∞ if
(α,β)→(a,b) β
a ∈ Z− ∧ b ∈/ Z+ −
0 , or if a ∈ Z ∧ b − a ∈ N.) So we can indeed write
?  
X
k z
(−1) f (t − kh)
k
(z) k=0
f (t) = lim (23)
h→0 hz
The sole question remaining is what the upper limit of the summation should be.

2.3 Simple examples revisited


Notice that (6) also works for z = −1:
Z t
(−1) 1
f (t) = eλt = eλt dt (24)
λ −∞
Duarte Valério

(14) too:
Z t
(−1) Γ(λ + 1) λ+1 1
f (t) = t = tλ+1 = tλ dt (25)
Γ(λ + 2) λ+1 0
We should likewise set the upper limit of the summation in (26) to retrieve an integral. But, as
the examples above remind us, we should care for what the lower limit of integration will be.

2.4 The Grünwald-Letnikoff definition


Making z = −1 in (26), and using (21),
?  
X
k −1
(−1) f (t − kh)
k
(−1) k=0
f (t) = lim
h→0 h−1
?
X (−1)k Γ(k + 1)
= lim h (−1)k f (t − kh) (26)
h→0 Γ(k + 1)Γ(1)
k=0
Rt  
If this is to be a Riemann integral c f (t) dt, then the upper limit of the summation must be t−ch ,
and h must be restricted to positive values. (If the good behaviour of f includes differentiability,
this will bring no problems of different right derivatives and left derivatives at any point.) We are
thus led to a definition of fractional derivative due to the works of Grünwald and of Letnikoff:
bXh c
t−c
 
k z
(−1) f (t − kh)
k
(z) k=0
f (t) = lim (27)
h→0+ hz
(With some effort it can be shown that z = −2, z = −3 and so on also lead to the corresponding
Riemann integrals.)
 
Notice that t−c h is converging (more precisely, diverging) to +∞. When z is natural, the
summation can be truncated after z terms, because from then on (22) shows that all terms are
zero (as k − z ∈ N). This is the only case the summation can be truncated; in all others there
will be an infinity of terms. What this means is that when z is natural the derivative will have
a value independent of c; otherwise the derivative will depend on c. Or, on other words, natural
order derivatives are local operators; all others are not local, since they depend on c, which for
z = −1 is the lower limit of integration (and is often called terminal). In other words still, if t is
identified with time, the result has a memory of what happens to function f for some time before,
since time c. This might come as a surprise: fractional derivatives always look like integrals in
that respect—whether z is a positive real or a negative real, even whether z is complex—and not
like derivatives.
To account for terminal c, the following notations are employed instead of the too simple f (z) (t):
dz f (t)
z
c Dt f (t) = c It−z f (t) = Dc+
z z
f (t) = (Dc+ z
f )(t) = Dt−c f (t) = (28)
d(t − c)z
(The + is because h → 0+ .) Below we will stick to the first of them.

2.5 Numerical computation of fractional derivatives


(27) can be easily employed to find a numerical approximation of a fractional derivative of a
function known with a sampling time Ts :
b t−c
Ts c  
X
k z
(−1) f (t − kTs )
k
z k=0
c Dt f (t) ≈ (29)
Tsz
Duarte Valério

Because Γ(z) grows very fast with |z|, if there are


 many
 terms in the summation the last ones will
z
be zero (because Γ(k) will return +∞ and thus will return 0). This is the same as bringing
k
c closer to t, and so care should be taken (e.g. resampling data) lest this will vitiate the result.
Of course, if the terms neglected are indeed neglectable, no problem will arise. In particular, if
|f (t)| < M, ∀ t > c, then the error ε committed by approximation
α
c Dt f (t) ≈ t−L Dtα f (t), t > c + L, α > 0 (30)

is bounded by
M
|ε| < (31)
Lα |Γ(1 − α)|

This result (proven in [22]) is known as the short memory principle, because it allows shortening
the memory of operator D without significantly altering the results.

2.6 The Riemann-Liouville definition


While (27) is fine for numerical purposes, it is very hard to use to find any derivatives analytically.
Fortunately, an equivalent expression can be found:
 Z t
(t − τ )−z−1
f (τ ) dτ, if <(z) ∈ R−





 c Γ(−z)





 f (t), if z = 0


z
c Dt f (t) = (32)

 d z−1
 D f (t), if <(z) = 0 ∧ =[z] 6= 0
dt c t







 d<(z)e
 d
 z−d<(z)e
f (t), if <[a] ∈ R+

D
dtd<(z)e c t
(A lengthy and very dense proof of this is given in [22].) This alternative definition is due to the
works of Riemann and Liouville, and for integer negative values of z reduces to
Z x Z x Z x
−n (x − t)n−1
c Dx f (t) = ··· f (t) dt · · · dt = f (t) dt (33)
(n − 1)!
|c c
{z } c

n integrations

which can be proven (once more) by mathematical induction. And for integer positive values of z
we get a particular case of the law of exponents, which says that c Dtm c Dtn f (t) = c Dtm+n f (t) holds
if m, n ∈ Z+ − +
0 , or if m, n ∈ Z0 , or if m ∈ Z ∧ n ∈ Z .

Using (32) it can be shown that



 (t − max{c, a})−α
α , if t > a
c Dt H(t − a) = Γ(1 − α) , c ∈ R ∨ c = −∞ (34)
 0, if c ≤ t ≤ a

 (t − a)−α−1
α , if t ≥ a ≥ c
c Dt δ(t − a) = Γ(−α) , c ∈ R ∨ c = −∞ (35)
 0, if c ≤ t < a ∨ a < c

Γ(λ + 1) λ−α
α λ
0 Dt t = t / Z−
, λ∈ (36)
Γ(λ − α + 1)
Duarte Valério

α λt
−∞ Dt e = λα eλt , λ 6= 0 (37)
α λt λc −α
c Dt e = e (t − c) E1,1−α (λ(t − c)), λ 6= 0 (38)
α λt −α
0 Dt e =t
E1,1−α (λt), λ 6= 0 (39)
 απ 
α α
D
−∞ t sin(λt) = λ sin λt + , λ 6= 0 (40)
 2 
α α απ
−∞ Dt cos(λt) = λ cos λt + , λ 6= 0 (41)
2
where H(t) is the Heaviside function, and Eα,β (t) is the two-parameter Mittag-Leffler function,
given by
+∞
X tk
Eα,β (t) = (42)
Γ(αk + β)
k=0

(For further results, see the tables in [22, 23, 15].)


Another important characteristic of definition (32) is that is clearly shows that D is a linear
operator of function f (t).

3 Variable-order derivatives
When order z depends on time t, there are three obvious ways of accounting for that variation:
Grünwald-Letnikoff definition GL1:
b t−c
h c
 
X
k z(t)
(−1) f (t − kh)
k
z(t) k=0
c Dt f (t) = lim+ (43)
h→0 hz(t)
Grünwald-Letnikoff definition GL2:
 
k z(t − kh)
z(t)
X (−1)
b t−c
h c
k
f (t − kh)
c Dt f (t) = lim (44)
h→0+
k=0
hz(t−kh)

Grünwald-Letnikoff definition GL3:


 
k z(kh)
z(t)
X (−1)
b t−c
h c
k
f (t − kh)
c Dt f (t) = lim (45)
h→0+
k=0
hz(kh)

Riemann-Liouville definition RL1:


 Z t
(t − τ )−z(t)−1
f (τ ) dτ, if <(z(t)) ∈ R−





 c Γ(−z(t))







 f (t), if z(t) = 0
z(t)
D
c t f (t) = (46)

 dd<(z(t))e z(t)−d<(z(t))e

 D f (t), if <(z(t)) ∈ R+



 dtd<(z(t))e c t




 d z(t)−1
 D f (t), if <(z(t)) = 0 ∧ z(t) 6= 0
dt c t
Duarte Valério

Riemann-Liouville definition RL2:


 Z t
(t − τ )−z(τ )−1
f (τ ) dτ, if <(z(t)) ∈ R−





 c Γ(−z(τ ))







 f (t), if z(t) = 0
z(t)
c Dt f (t) = (47)

 dd<(z(t))e z(t)−d<(z(t))e


 d<(z(t))e c Dt f (t), if <(z(t)) ∈ R+


 dt




 d z(t)−1
 D f (t), if <(z(t)) = 0 ∧ z(t) 6= 0
dt c t
Riemann-Liouville definition RL3:
 Z t
(t − τ )−z(t−τ )−1
f (τ ) dτ, if <(z(t)) ∈ R−





 c Γ(−z(t − τ ))







 f (t), if z(t) = 0
z(t)
D
c t f (t) = (48)

 dd<(z(t))e z(t)−d<(z(t))e

 D f (t), if <(z(t)) ∈ R+



 dtd<(z(t))e c t




 d z(t)−1
 D f (t), if <(z(t)) = 0 ∧ z(t) 6= 0
dt c t
The argument of z is the current value of t in definitions GL1 and RL1, the same argument
of f in definitions GL2 and RL2, and the difference of the two above in definitions GL3 and RL3.
Different definitions correspond to operators with different memories of past values of z(t). In
short, if z(t) has a negative real part, GL1 and RL1 correspond to an operator without memory
of past values of z(t), GL2 and RL2 to an operator with some memory of past values of z(t), and
GL3 and RL3 to an operator with an even stronger memory of past values of z(t); if z(t) has a
positive real part or is a pure imaginary number, GL1 still provides an operator without memory
of past values of z(t), but RL1 does not, while GL2 and RL2 no longer lead to the same result,
neither do GL3 and RL3. These differences compound if the sign of z(t) changes with time. This
(eventual) memory of past values of z(t) should not be confused with the memory of past values
of f (t). (See [30] for details, or [12] in what negative real orders are concerned.)

4 Fractional and variable order controllers


4.1 Laplace transforms
It can be shown from the definition of Laplace transforms and from the Riemann-Liouville definition
that  z
 s F (s), if <(z) ∈ R−







 F (s), if z = 0



L [0 Dtz f (t)] = sz F (s) − 0 Dtz−1 f (0), if <(z) = 0 ∧ =(z) 6= 0 (49)






 d<(z)e−1

 X


 s z
F (s) − sk 0 Dtz−k−1 f (0), if <(z) ∈ R+
k=0
Duarte Valério

=[σ] =[σ]

απ απ
2 2
<[σ] <[σ]

Figure 1: In grey: regions of the complex plane where the roots of A(σ) may lie in the case of a
stable commensurate transfer function; left: 0 < α < 1; right: 1 < α < 2

4.2 Fractional transfer functions


Laplace transforms of differential equations involving fractional derivatives will, if initial conditions
are all set to zero, lead us to fractional transfer functions such as
m
X
b k s yk
k=1
G(s) = n (50)
X
zk
ak s
k=1

Those where all orders yk and zk are integer multiples of a least common divisor z are called
commensurable: m
X
bk skz
k=0
G(s) = n (51)
X
kz
ak s
k=0

In other words, commensurable transfer functions are the ratio of two polynomials in s z .
Commensurable
Pn transfer functions are stable if and only if the roots σk , k = 1 . . . n of polyno-
mial A(σ) = k=0 ak σ k verify
π
|∠σk | > α , ∀ k (52)
2
(restricting ∠σk to [−π, +π] rad). This theorem is due to Matignon (and is proved in [14]) and
gives us the regions where the σ may lie in the case of a stable commensurable transfer function,
as shown in figure 1. (Notice that for integer systems we reach the usual criterion of stability—all
poles must have a negative real part—and that orders α > 2 make the system necessarily unstable.)
Time responses of fractional transfer functions can be found from the following cases—established
analytically from the definition—using the convolution theorem:
 
−1 1 tα−1
L L [δ(t)] = (53)
sα Γ(α)
 
1 tα
L −1 α L [H(t)] = (54)
s Γ(α + 1)
 
1 tα+1
L −1 α L [t] = (55)
s Γ(α + 2)
Duarte Valério

 
1
L −1 α L [δ(t)] = tα−1 Eα,α (±atα ) (56)
s ∓a
 
1
L −1 α L [H(t)] = tα Eα,α+1 (±atα ) (57)
s ∓a
 
−1 1
L L [t] = tα+1 Eα,α+2 (±atα ) (58)
sα ∓ a
 
1 tα(k+1)−1 dk Eα,α (±atα )
L −1 α k+1
L [δ(t)] = , k ∈ Z+0 (59)
(s ∓ a) Γ(k + 1) d(±atα )k
 
1 tα(k+1) dk Eα,α+1 (±atα )
L −1 α k+1
L [H(t)] = , k ∈ Z+0 (60)
(s ∓ a) Γ(k + 1) d(±atα )k
 
1 tα(k+1)+1 dk Eα,α+2 (±atα )
L −1 L [t] = , k ∈ Z+0 (61)
(sα ∓ a)k+1 Γ(k + 1) d(±atα )k
Frequency responses of fractional transfer functions can be shown to result from replacing s by
jω, as for usual transfer functions. In particular, the frequency response of G(s) = s z is

G(jω) = (jω)z = (jω)<(z)+j=(z) = j <(z) ω <(z) j j=(z) ω j=(z) (62)


   
<(z)π <(z)π =(z)π
= cos + j sin ω <(z) e− 2 cos(=(z) log ω) + j sin(=(z) log ω)
2 2
 =(z)π
 =(z)π
20 log10 |G(jω)| = 20 log10 ω <(z) e− 2 = 20<(z) log10 ω + 20 log10 e− 2 (63)
 h i 
<(z)π <(z)π
∠G(jω) = ∠ cos + j sin cos(=(z) log ω) + j sin(=(z) log ω)
2 2
<(z)π
= + =(z) log(10) log10 ω (64)
2
and the frequency response of G(s) = s 1α +1 , α ∈ R+ is given by G(jω) = ( jω )1α +1 , which means
 α (a) a

a
that when ω  a, G(jω) ≈ jω ; when ω  a, G(jω) ≈ 1; and when ω = a
απ
1 1 + cos απ2 + j sin 2 1
|G(ja)| =

απ απ
=
απ =p (65)
cos 2 + j sin 2 + 1 2 + 2 cos 2 2 + 2 cos απ
2
απ απ απ
− sin 2 −2 sin 4 cos 4 π
∠G(ja) = arctan = arctan απ = −α (66)
1 + cos απ
2 2 cos απ
4 cos 4 4
These frequency responses are shown in figures 2 and 3 respectively. The latter has a gain plot which
is asymptotically horizontal for low frequencies and asymptotically linear for high frequencies, and
a phase plot constant and equal to 0◦ for low frequencies and constant and equal to α90◦ for high
frequencies.

4.3 The Crone approximation


It is possible to find usual transfer functions (i.e. commensurate ones given by (51) with z = 1) that
approximate the behaviour of fractional transfer functions. The most widely used approximation is
the Crone approximation—Crone being the acronym of Commande Robuste d’Ordre Non-Entier —,
which has N poles and N zeros within a frequency range [ωl , ωh ] and is given by [18, 20]
N 1+ s
Y ωζ,m
z
s ≈C s (67)
m=1
1+ ωπ,m
  2m−1−z
ωh 2N
ωζ,m = ωl (68)
ωl
Duarte Valério

<(z) > 0, =(z) = 0 <(z) < 0, =(z) = 0 <(z) > 0, =(z) > 0
dB de dB 20< dB ade
/d eca (z )
dB dec
dB /de d B/
<(z ) ω ca de ω (z ) ω
20 20<

◦ ◦ ◦
cade
<(z)90◦ 1 0 ) rad/de
log(
ω ω =(z ) ω
<(z)90◦

<(z) > 0, =(z) < 0 <(z) < 0, =(z) > 0 <(z) < 0, =(z) < 0
dB ad e dB 20<
(z )
dB 20<
(z )
dec dB dB
dB/ /de /de
(z ) ω cad ω ca de ω
20< e

◦ ◦ de ◦
/deca
=(z ) 0 ) rad =(z )
log log(1 log
(10) r
ω =(z ) ω
(10) r
cade ω
ad/d ecade ad/de

Figure 2: Bode diagram of sz

dB
a 20α
dB
ω
/de
c ade

a
ω
α45◦
α90◦

1
Figure 3: Bode diagram of
( as )α +1

  2m−1+z
ωh 2N
ωπ,m = ωl (69)
ωl
If z is real, the correct gain at 1 rad/s, which is |(jω)α | = 1, ∀ α, must be set by adjusting C; if z
is complex, both gain and phase at 1 rad/s must be set by adjusting C. (Doing this for frequency
1 rad/s makes calculations easier, but if 1 rad/s is outside [ωl , ωh ] any other suitable frequency
will have to be used instead.)
As is clear from (68) and (69), if z is real poles and zeros are real too, and are recursively placed
alternating on the negative real semi-axis:
  N1
ωζ,m+1 ωπ,m+1 ωh
= = (70)
ωζ,m ωπ,m ωl
This constant ratio implies that both poles and zeros are equidistant in a logarithmic scale of
frequencies, such as that found in a Bode diagram. Each pole lowers the phase 90◦ ; each zero
raises it 90◦ : the superposition of these effects results in a phase oscillating around α90◦ . Each pole
bends the gain down 20 dB per decade; each zero bends it up 20 dB per decade: the superposition
of these effects results in a gain with a slope of 20z dB per decade.  
In practice, if |z| < 1, (67) is valid in the frequency range 10ωl , ω10h . Both gain and phase
have ripples, which decrease as N increases. Typically N should be at least equal to the number
Duarte Valério

of decades in [ωl , ωh ] for acceptable results. When |z| > 1, the frequency range where the approx-
imation behaves acceptably becomes narrower, and approximations such as s z = sd<(z)e sz−d<(z)e
or sz = sb<(z)c sz−b<(z)c (for which only the last term needs to be approximated) are employed.
To approximate a general fractional transfer function given by (50), m+n Crone approximations
of all the syk and szk are built, and then linearly combined.

4.4 Fractional controllers


Fractional transfer functions can be employed to model dynamic systems, and to devise controllers
for dynamic systems. The plants to which fractional controllers are applied may or may not be
fractional themselves. Among other possibilities, the following fractional controllers for single-
input, single-output plants are the most usual:
I
• Fractional PID controllers, given by P + sλ
+ Dsµ , generalise widely used PID controllers,
given by P + Is + Ds. [22, 27, 28]
• First generation Crone controllers, given by sα , where α is a positive real number, increase
the plant’s phase (and thus its phase margin) by a constant value of α90◦ . [18, 19]
• Second generation Crone controllers achieve an open loop given by sα , where α is real, and
which thus has a constant phase margin. [18, 19]
• Third generation Crone controllers achieve an open loop given by sz , where z is complex,
and is chosen with an optimisation method so that the open loop’s frequency response avoids
values leading to poor performances, especially in the presence of plant uncertainties. [20, 19]
These controllers may need to employ time varying orders if the plant or the specifications change
with time. [29] Variable order controllers are thus particular cases of adaptive control.

5 Applications
Fractional order derivatives and fractional transfer functions can be used to model many systems,
such as:
• heat fluxes in the soil, found from surface soil temperature measurements; [32]
• the propagation of sound in porous materials; [9]
• the transport of substances by water in soils; [3]
• diffusion phenomena in plasmas; [6]
• prices of financial derivatives in stock markets; [4]
• viscoelasticity phenomena; [2]
• optical systems; [24]
• the human respiratory system. [10]
There are many instances of applications of fractional controllers. Their major advantage is
their robustness in the presence of model uncertainties. The following are a few examples:
• A flexible belt transmission was controlled using Crone controllers. The objective was to
keep performance irrespective of variations in the mass conveyed. [21, 25]
• An active suspension system was devised based upon Crone control. The objective was to
keep vehicle security and passenger comfort irrespective of the state of the road and the
travel velocity. [1, 16, 19]
Duarte Valério

• Irrigation canals were controlled using fractional PID controllers. The objective was to keep
water levels irrespective of unscheduled water demands. [8, 7]
• Flexible robots, that have high carried mass to robot mass ratios to increase energy use
efficiency, were controlled using fractional PIDs. The objective was to keep robot movements
precise irrespective of the vibrations unavoidable in a flexible structure. [31, 26]
There are yet other applications of fractional derivatives, such as:
• image processing; [13]
• processing of geological data (e.g. magnetic and gravity data); [5]
• path planning for robots. [17]

Acknowledgments
Research for this paper was partially supported by grant PTDC/EME-CRO/70341/2006 of FCT,
funded by POCI 2010, POS C, FSE and MCTES; and by the Portuguese Government and FEDER
under program “Programa de Financiamento Plurianual das Unidades de I&D da FCT para as
atividades de investigação do laboratório associado LAETA” (POCTI-SFA-10-46-IDMEC).

References
[1] O. Altet, X. Moreau, M. Moze, P. Lanusse, and A. Oustaloup. Principles and synthesis of
hydractive CRONE suspension. Nonlinear Dynamics, 38(1–2):435–459, 2004.
[2] Ronald Bagley and Peter Torvik. Fractional calculus—a different approach to the analysis of
viscoelastically damped structures. AIAA Journal, 21(5), 1983.
[3] David Benson, Charles Tadjeran, Mark M. Meerschaert, Irene Farnham, and Greg Pohll.
Radial fractional-order dispersion through fractured rock. Water resources research, 40:1–9,
2004.
[4] Álvaro Cartea and Diego del-Castillo-Negrete. Fractional diffusion models of option prices in
markets with jumps. Physica A: Statistical Mechanics and its Applications, 374(2):749–763,
2007.
[5] Gordon Cooper and Duncan Cowan. The application of fractional calculus to potential field
data. Exploration Geophysics, 34:51–56, 2003.
[6] D. del-Castillo-Negrete, B. A. Carreras, and V. E. Lynch. Fractional diffusion in plasma
turbulence. Physics of plasmas, 11(8):35–84, 2004.
[7] Jorge Domingues, Duarte Valério, and José Sá da Costa. Rule-based fractional control of an
irrigation canal. ASME Journal of Computational and Nonlinear Dynamics, 2010. Accepted.
[8] V. Feliu-Batlle, R. Pérez, and L. Rodrı́guez. Fractional robust control of main irrigation canals
with variable dynamic parameters. Control Engineering Practice, 15:673–686, 2007.
[9] Z. E. A. Fellah, S. Berger, W. Lauriks, and C. Depollier. Verification of kramers–kronig rela-
tionship in porous materials having a rigid frame. Journal of Sound and Vibration, 270:865–
885, 2004.
[10] Clara-Mihaela Ionescu. Fractional Order Models of the Human Respiratory System. PhD
thesis, Universiteit Gent, 2009.
Duarte Valério

[11] Marcia Kleinz and Thomas J. Osler. A child’s garden of fractional derivatives. The College
Mathematics Journal, 31(2):82–88, 2000.
[12] Carl F. Lorenzo and Tom. T. Hartley. Variable order and distributed order fractional opera-
tors. Nonlinear dynamics, 29:57–98, 2002.
[13] B. Mathieu, P. Melchior, A. Oustaloup, and Ch. Ceyral. Fractional differentiation for edge
detection. Signal processing, 83:2421–2432, 2003.
[14] Denis Matignon. Stability properties for generalized fractional differential systems. ESAIM
Proceedings, 5:145–158, 1998.
[15] K. S. Miller and B. Ross. An introduction to the fractional calculus and fractional differential
equations. John Wiley and Sons, New York, 1993.
[16] X. Moreau, O. Altet, and A. Oustaloup. The CRONE suspension: Management of the dilemma
comfort-road holding. Nonlinear Dynamics, 38(1–2):461–484, 2004.
[17] A. Oustaloup, B. Orsoni, P. Melchior, and H. Linarès. Path planning by fractional differenti-
ation. Robotica, 21:59–69, 2003.
[18] Alain Oustaloup. La commande CRONE : Commande Robuste d’Ordre Non Entier. Hermès,
Paris, 1991.
[19] Alain Oustaloup, Olivier Cois, Patrick Lanusse, Pierre Melchior, Xavier Moreau, Jocelyn
Sabatier, and Jean-Luc Thomas. A survey on the CRONE approach. In Alain Le Méhauté,
J. A. Tenreiro Machado, Jean Claude Trigeassou, and Jocelyn Sabatier, editors, Fractional
differentiation and its applications, pages 735–780. LAPS, Bordeaux, 2005.
[20] Alain Oustaloup, François Levron, Benoı̂t Matthieu, and Florence M. Nanot. Frequency-band
complex noninteger differentiator: characterization and synthesis. IEEE Transactions on
Circuits and Systems—I: Fundamental Theory and Applications, 47(1):25–39, January 2000.
[21] Alain Oustaloup, B. Mathieu, and P. Lanusse. The CRONE control of resonant plants: ap-
plication to a flexible transmission. European Journal of Control, 1:113–121, 1995.
[22] I. Podlubny. Fractional differential equations: an introduction to fractional derivatives, frac-
tional differential equations, to methods of their solution and some of their applications. Aca-
demic Press, San Diego, 1999.
[23] S. G. Samko, A. A. Kilbas, and O. I. Marichev. Fractional integrals and derivatives. Gordon
and Breach, Yverdon, 1993.
[24] Enrique Tajahuerce, Genaro Saavedra, Walter D. Furlan, Enrique E. Sicre, and Pedro Andrés.
White-light optical implementation of the fractional fourier transform with adjustable order
control. Applied Optics, 39(2):238–245, 2000.
[25] Duarte Valério. Non-integer order robust control: an application. In European Control Con-
ference (Student Forum), Porto, 2001.
[26] Duarte Valério and José Sá da Costa. Fractional control of a flexible robot. In Alain Le
Méhauté, J. A. Tenreiro Machado, Jean Claude Trigeassou, and Jocelyn Sabatier, editors,
Fractional differentiation and its applications, pages 649–660. LAPS, Bordeaux, 2005.
[27] Duarte Valério and José Sá da Costa. Tuning of fractional PID controllers with Ziegler-Nichols
type rules. Signal Processing, 86(10):2771–2784, 2006.
Duarte Valério

[28] Duarte Valério and José Sá da Costa. Tuning rules for fractional PIDs. In J. A. Tenreiro
Machado, Jocelyn Sabatier, and Om Agrawal, editors, Fractional calculus: theoretical devel-
opments and applications in Physics and Engineering, pages 463–476. Springer, Dordrecht,
2007.
[29] Duarte Valério and José Sá da Costa. Variable order fractional controllers. In Fractional
Differentiation and its Applications, Badajoz, 2010.
[30] Duarte Valério and José Sá da Costa. Variable-order fractional derivatives and their numerical
approximations. Signal Processing, 2010. http://dx.doi.org/10.1016/j.sigpro.2010.04.006.
[31] Blas Vinagre. Modelado y control de sistemas dinámicos caracterizados por ecuaciones ı́ntegro-
diferenciales de orden fraccional. PhD thesis, Universidad Nacional de Educación a Distancia,
Madrid, 2001.
[32] J. Wang and R. L. Bras. Ground heat flux estimated from surface soil temperature. Journal
of Hydrology, 216:214–226, 1999.

You might also like