Past and Present

by Chengbin Ma
and Yoichi Hori

Fractional-Order Control: Theory and Applications in Motion Control

1 + A( jω)β = 0 (1) Manabe [3]–[5]. However, the FOC

concept was not widely incorporated
This article carries on the discussion on where A = Jm /Kd in nominal case and into control engineering mainly due
fractional-order control started by the β = 2 − α. Equation (1) can be rewrit- to the unfamiliar idea of taking frac-
article D. Cafagna, “Fractional calculus: A ten in the form tional order, so few physical applica-
mathematical tool from the past for tions and limited computational
1
present engineers,” IEEE Industrial Elec- ( jω)β = − . (2) power available at that time [6].
tronics Magazine, vol. 1, no. 2, pp. A
In last few decades, researchers
35–40, Summer 2007. The movement of −1/ A can be pointed out that fractional-order differ-
considered to be the locus of the criti- ential equations could model various
cal point (see Figure 2) when the gain materials more adequately than inte-

T
he concept of fractional-order variation occurs. For integer order ger order ones and provide an excel-
control (FOC) means controlled system, when β = 2, the system will lent tool for describing complex
systems and/or controllers are be oscillatory due to its zero phase dynamic features [1], [7]. Especially
described by fractional-order differen- margin. Taking β = 1 leads to poor for the modeling and identification of
tial equations. Expanding derivatives robustness against saturation since flexible structures with increasing
and integrals to fractional orders is by pure D controller will be used. By let- application of lighter materials, frac-
no means new and actually has a firm ting β be fractional between one and tional- order differential equations
and long-standing theoretical founda- two, a better tradeoff between stabili- could provide a natural solution since
tion. Interest in this subject was evi- ty and robustness will be obtained. these structures are essentially distrib-
dent almost as soon as the ideas of Namely, the fractional-order Dα con- uted-parameter systems [8]. Obvious-
classical calculus were known. The ear- troller is naturally introduced whose ly, the fractional-order models need
liest more or less systematic studies order α should be chosen properly fractional-order controllers for more
seem to have been made in the begin- between zero and one. Therefore, nec- effective control of the “real” systems.
ning and middle of the 19th century by essary phase margin can be easily This necessity motivated renewed
Liouville, Holmgren, and Riemann, kept to any desired amount in wide interest in various applications of FOC
although Eular, Lagrange, and others range of frequencies below and in the [10]–[13]. And with the rapid develop-
made contribution even earlier [1]. Par- neighborhood of the critical point. ment of computer performances, real-
allel to these theoretical beginnings This characteristic highlights the ization of FOC systems also became
was the development of applying frac- hopeful aspect of applying FOC to possible and much easier than before.
tional calculus to various problems [1]. control engineering. Generally there are three main
As to fractional calculus’ application Some other pioneering works were advantages for introducing fractional-
in control engineering, FOC was intro- also carried out around 1960s by order calculus to control engineering:
duced by Tustin for the position control
of massive objects (see Figure 1) half a
century ago in 1958, where actuator sat-
θr + e u 1
uration requires sufficient phase margin Kd sα θm
around and below the critical point [2]. − Jms2
The characteristic equation of the Saturation
above close-loop 1/s β system with
variable gain factor A is

Digital Object Identifier 10.1109/MIE.2007.909703 FIGURE 1 — The position control loop with fractional order Dα controller.

6 IEEE INDUSTRIAL ELECTRONICS MAGAZINE ■ WINTER 2007 1932-4529/07/$25.00©2007IEEE

 1) adequate modeling of control very straightforward [7]. The final
plant’s dynamic features expression of the Laplace transform of Ims
2) effective and clear-cut robust con- the fractional-order derivative is
trol design S-Plane
 α
α
3) reasonable realization by approxi- L 0 D t f(t) = s F (s)
−1/A
mation.

n−1
In following sections, detailed − sk0 D α−k−1
t f (0) −1 0 Res
descriptions of the above three advan- k =0
tages will be given. (6)
ω

Mathematical Aspects where n − 1 < α < n again. If all the

initial conditions are zero, the Laplace
β
Mathematic Definitions transform of fractional-order derivative
One of the most frequently encountered is simply
definitions of fractional-order calculus is
 α

called the Grünwald-Letnikov definition: L 0 D t f(t) = sα F (s). (7) FIGURE 2 — Nyquist plots of the fractional
order 1/sβ system.

n  
α α Therefore the Laplace transforms of
a Dt = lim h−α (−1) j
h→0 j fractional ± α order calculus lead to
j=0
nh = t−α the use of fractional-order Laplace models to describe real objects. The
× f(t − jh) (3) operator s ±α . The transfer functions of fractional-order model can provide a
  models and controllers, which are new possibility to acquire more ade-
α
where j
are the binomial coeffi- described by fractional-order differen- quate modeling of dynamic processes.
cients. Obviously, introducing the frac- tial equations, can be derived conve- Fractional-order models have been
tional-order calculus leads to infinite niently using fractional-order Laplace applied to describe reheating furnaces
dimension, while the integral calculus operator s±α . [7], viscoelasticity [1], [7], [8], chemical
is finite dimension. Similarly, the Fourier transform of processes [14], and chaos systems [15].
Another widely known definition is fractional-order derivative is Actually, using a fractional-order
called the Riemann-Liouville definition model for describing distributed-
 α

with an integrodifferential expression. Fe 0 D t f(t) = (jω)α F (jω). (8) parameter systems is quite natural
The definition for the fractional-order since the Laplace transform of partial
integral is The frequency response of FOC sys- differential equations will inevitably
tem can be exactly obtained by substi- introduce the fractional-order s opera-
 t
−α 1 tuting s±α with (jω)±α in its transfer tor. For a simple example, consider a
a D t f(t) = (t − ξ )α−1
(α) a function. This advantage implies fre- torsional model as shown in Figure 3,
× f(ξ )dξ (4) quency-domain analysis of FOC system which consists of a flexible shaft
is as convenient as integer order con- attached to a rigid disk [16]. The rigid
while the definition of fractional-order trol (IOC) systems. The graphical tools body equation of the disk is given as
derivatives is for IOC in frequency domain are still
I1 s 2 θ1 = T1 + T12 . (9)
dγ  −(γ −α) 
available for FOC analysis and design.
α
a D t f(t) = γ a Dt (5)
dt
Modeling and Identification Take a small element of length dx
where (x) is the Gamma function, a Recently, there has been a growing sig- along the shaft axis and observe the
and t are limits, and α (α > 0 and nificant demand for better mathematic cylindrical surface, as shown in
α ∈ R) is the order of the operation. γ is
an integer that satisfies γ − 1 < α ≤ γ .
The Grünwald-Letnikov and Riemann-
Liouville definitions are both a unifica- θ1 T12
T1 θ2
tion of integer order derivatives and
integrals [7].
I1 θ(t, x ) I 2, d, k
Laplace and Fourier Transforms
Fractional-order calculus is quite com-
plicated in time domain, as shown in
l
its two definitions. Fortunately one of
the features most important to control
engineers, its Laplace transform, is FIGURE 3 — The flexible shaft attached to a rigid disk.

WINTER 2007 ■ IEEE INDUSTRIAL ELECTRONICS MAGAZINE 7

Figure 4(a). This element will deform ure 4(b), since this shear stress acts For (14), the Laplace transform in t, a
through a small angle dθ . tangentially, the overall torque at the second-order differential equation, is
Based on the theory of elasticity, shaft cross section is
I2 2 d 2 θ(x)
the corresponding shear stress at the s θ(x) − kl =0 (15)
l dx 2
deformed point at radius r is 
T= r × (τ × 2πrdr) where θ(s, x) is abbreviated as θ(x) for
∂θ(t, x)
τ = Gγ = Gr (10)  simplicity. For the free end of the shaft,
∂x ∂θ(t, x)
=G 2πr 3 dr there is no deformation and the shear
∂x
where G is shear modulus and γ is stress is zero. Therefore, the following
∂θ(t, x)
the shear strain [17]. As shown in Fig- = GJ . (11) boundary conditions can be obtained:
∂x
 
 dθ(x) 
θ(x) = θ1 , = 0.
Apply Newton’s second law for rotato- x =0 dx x=l
ry motion of the small element dx (16)
γ r dθ shown in Figure 4(a), the equation of
T T+ dT
motion is Torque T12 in Figure 3 can be obtained:

dθ(x) 
∂ 2 θ(t, x) T12 (s) = kl
dx ρ Jdx = T + dT − T dx x=0
(a) ∂t 2
∂T(t, x) = −tanh(μls)θ1 (17)
Shear = dx. (12)
Stress
∂x
r + dr where μ 2 = I2 /kl. Finally, substitute
τ
For a uniform shaft segment of length l T12 in (9), the transfer function
r
with associated overall angular defor- between T1 and θ1 can be achieved:
mation θ, the torsional stiffness k is
T1
= I1 s 2 + kl · μs · tanh(μls). (18)
T ∂θ(t, x) 1 GJ θ1
k= = GJ · = . (13)
θ ∂x θ l
However, in a conventional modeling
Based on (11) and (12), the following method, the torsional system in
(b) equation can be obtained: Figure 3 is usually modeled as a rigid
FIGURE 4 — Deformation of the torsional
body system with inertia I = I1 + I2 :
shaft: (a) small element of the torsional shaft I2 ∂ 2 θ(t, x) ∂ 2 θ(t, x)
− kl = 0. (14) T1
and (b) shear stress in a small annular cross l ∂t 2 ∂x 2 = ( I1 + I2 )s 2 . (19)
section. θ1
As shown in the Bode plots of Figure 5,
the fractional-order transfer function
100
model in (18) displays the mechanical
Classical Model
resonance effect naturally. At low fre-
Magnitude (dB)

50
quency range, the two models give sim-
0 ilar frequency responses. At high
frequency range, the fractional model
−50 can describe the distributed nature of
Fractional Order Model
the torsional system; while in conven-
−100
100 101 102 103 104 tional integer order model, this nature
is totally ignored. Fractional-order mod-
200 eling is a useful tool to give a more ade-
quate description of system’s “real”
150
dynamic features.
Phase (°)

100 From the above example, it can be

seen that distributed-parameter sys-
50 tems are naturally described by a set
of partial differential equations. How-
0 ever, these equations will lead to
100 101 102 103 104
transfer functions that are quotients of
Freq. (rad/s)
transcendental functions.
FIGURE 5— Bode plots of the torsional system’s fractional-order model and conventional integer Using a fractional-order transfer func-
order model. tion model, a quotient of polynomials in

8 IEEE INDUSTRIAL ELECTRONICS MAGAZINE ■ WINTER 2007

sα , it is also possible to fit better a set selecting the phenomena that can be A frequency-band, fractional-order
of experimental data. For example, the modeled. For example, when modeling controller can be realized by broken line
frequency-domain identification of a a flexible structure, using α = 2 can approximation in frequency domain. But
flexible structure by a fractional-order not model damping. In α = 1 case, the further discretization is required for this
model can take into account not only damping can be modeled. By further method [18]–[20]. As to direct discretiza-
material damping, but also other vari- taking α = 0.5, other phenomena such tion, various methods have been pro-
eties of physical phenomena such as as viscoelasticity and anomalous posed such as sampling time scaling
viscoelasticity and anomalous relax- relaxation will be described. The other [21], short memory principle [7], Tustin
ation. This fact indicts fractional-order advantage of this model is that exist- Taylor expansion [22], Lagrange function
models can be an appropriate and ing optimization methods can still be interpolation method [10], while all the
hopeful tool to model the dynamic fea- used since only integer order s α is approximation methods need truncation
tures of flexible structure more accu- introduced. of the expansion series. A detailed com-
rately which is becoming more and parison of the above direct discretiza-
more important due to lighter materi- Realization Methods tion methods can be found in [23].
als and faster motions [7], [8]. Though it is not difficult to under-
For fractional-order models like stand the theoretical advantages of Frequency-Band Approximation
(20), frequency-domain identification FOC, especially in frequency domain, Since a fractional-order system’s fre-
methods to determine the coeffi- realization issue kept being some- quency responses can be exactly
cients αk, βk(k = 0, 1, 2, . . . ) and what problematic and perhaps was known, approximating fractional-order
ak, bk(k = 0, 1, 2, . . . ) are as routine one of the most doubtful points for controllers by frequency-domain
as conventional integer order models. the application of FOC. Fractional- approaches is natural. At the same
order systems have an infinite dimen- time, it is neither practicable nor desir-
Y (s) sion; while the conventional integer able to try to make the order be frac-
G(s) =
U (s) order systems are finite dimension. tional in whole frequency range. The
bm sβm + · · · + b1 sβ1 + b0 sβ0 To realize fractional-order controllers frequency-band, fractional-order con-
= .
an sαn + · · · + a1 sα1 + a0 sα0 perfectly, all the past inputs should trollers are required and practical in
(20) be memorized. It is impossible in real most control applications. The broken-
applications. Proper approximation line approximation method can be
Various identification methods for by finite differential or difference introduced to realize frequency-band,
determination of the coefficients were equation must be introduced. fractional-order I α controller. Let
developed [7]–[9], based on minimiza-
tion of the difference between the
measured frequency response F (ω)
and the frequency response of the
model G(jω). For example, the quad-
ratic criterion for the optimization can Magnitude (dB)
ωb
be in following form: ω1'
O logω

M
Q= W 2 (ωm )|F (ωm ) − G(jωm )|2
m =0
−20αdB/dec
(21)
−20dB/dec
where W (ωm ) is a weighting function −20αdB/dec
and M is the number of measured values
of frequencies ω = (ω0 , ω1 , . . . , ωM ). 0dB/dec
Compared to the general fractional-
order model as in (20), a special model
can be introduced, in which only inte-
ger orders of the fractional-order oper-
ator s α are used:
m α i
i =0 ai (s )
G(s) =  ,
(s α )n + j =0 bj (s α ) j
n−1
ω3
n ≥ m. (22) ωh

It is interesting to note that the selec-

tion of α can actually be seen as FIGURE 6— An example of broken-line approximation (N = 3).

WINTER 2007 ■ IEEE INDUSTRIAL ELECTRONICS MAGAZINE 9

 the interval [t − L, t], where L is the
0 length of “memory”
N=1
α
Magnitude (dB)
Ideal Case 0 D t x(t ) ≈t−L D αt x(t ), (t > L). (25)
−5
N=2 and 3 Based on approximation of the time
−10 increment h through the sampling time
T , the discrete equivalent of the frac-
−15 tional-order α derivative is given by
100 101 102 103 104 105 106

0 ⎛ ⎞
1 m
−10 Z {D [x(t)]} ≈ ⎝ α
α
cjz ⎠ X (z)
−j
T j =0
Phase (°)

−20
(26)
−30
−40 where m = [L/T] and the coefficients
cj are
−50
100 101 102 103 104 105 106
Freq. (rad/s) c0 = 1,
 
α
cj = (−1) j
j
FIGURE 7— Bode plots of ideal case, the first-, second-, and third-order approximations.
j−α −1
= · cj−1 , j ≥ 1.
j
 α
s
+1  ω
N−1 s
+1 tion method. Even taking N = 2 can (27)
ωh
s ≈ s
i
. (23) give a satisfactory accuracy in fre-
ωb +1 i =0 ωi
+1
quency domain. It must be pointed out that the neces-
Based on Figure 6, ωi and ωi can be sary memory length, namely how good
calculated as Direct Discretization the approximation is needed, should be
1 α
The most commonly used discretiza- decided by the demand of specific con-
  i+ 2 − 2 tion method of a fractional-order con- trol problem [23]. Larger memory gives
ωh N
ωi = ωb , troller is called the short memory better performance but also leads to a
ωb
1 α principle method. This discretization longer computation time. However, this
  i+ 2 + 2
ωh N method is based on the observation tradeoff is not restricted in FOC, actually
ωi = ωb . (24)
ωb that for the Grünwald-Letnikov defini- a common problem in digital control.
tion, the values of the binomial coeffi-
Figure 7 shows the Bode plots of ideal cients near “starting point’’ t = 0 are An Example in Motion Control
frequency-band case (α = 0.4, ωb = small enough to be neglected or “for- Here a factional order PID k controller
200 Hz, ωh = 10, 000 Hz) and its gotten’’ for large t. Therefore the prin- is applied to torsional system’s back-
first-, second-, and third-order approx- ciple takes into account the behavior lash vibration suppression control
imations by broken-line approxima- of x(t) only in “recent past,” i.e., in [24]. In PID k controller D’s order can

Load Flywheel
(Changeable) Friction Load Adjustment

Driving Flywheel (Changeable)

Load Servomotor Bearing

Backlash Adjustment
Belt
Torsional Shaft
(Changeable) Encoder
Driving
Servomotor Tacho-Generator

FIGURE 8— Experimental setup of torsional system.

10 IEEE INDUSTRIAL ELECTRONICS MAGAZINE ■ WINTER 2007

be any real number, not necessarily be
Tl + ωl
an integer. The experimental setup of 1/(J ls)
torsional system is shown in Figure 8
+
where gear backlash exists. Tuning
−
fractional-order k can adjust control
systems frequency response directly; Ks /s
therefore a straightforward design can +
be achieved for robust control against −
backlash nonlinearity. 1/(Jgs)
The simplest model of the torsional
+
system with gear backlash is the three-
−
inertia model shown in Figure 9, where −δ
Kg 1/s
Jm , Jg , and Jl are driving motor, gear +δ
(driving flywheel) and load’s inertias, +
Ks shaft elastic coefficient, ωm and ωl −
Tm + ωm
motor and load rotation speeds, Tm 1/(Jms)
input torque, and Tl disturbance
torque. An interesting and more thor-
ough analysis of backlash nonlinearity FIGURE 9— Block diagram of the three-inertia model.
based on the describing function
method can be found in [25], in which
the fractional-order dynamics of the
backlash is illustrated. Tl + ωl
1/(JIs)
Since the gear elastic coefficient K g
is much larger than the shaft elastic +
coefficient Ks (K g >> Ks ), for speed −
control design the two-inertia model is Ks /s
commonly used in which driving +
motor inertia Jm and gear inertia Jg
Tm − ωm
are simplified to a single inertia
1/(Jmgs)
Jmg (= Jm + Jg ) (see Figure 10). +
The PID controller is designed
based on the simplified two-inertia
model. Simulation results with the sim- FIGURE 10— Block diagram of the two-inertia model.
plified two-inertia model show this
integer order PID control system has a
superior performance for suppressing 60
torsion vibration (see Figure 11). 50
For three-inertia plant P 3m (s), the
ωm (rad/s)

40
close-loop transfer function of integer
order PID control system from ωr to 30
ωm is 20
10
G close (s) =
0
C I (s)P 3m (s) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1 + C I (s)P 3m (s) + C PD (s)P 3m (s)
30

(28) 25
20
ωl (rad/s)

where C I (s) is I controller and C PD (s) 15

is the parallel of P and D con-
10
trollers in minor loop; therefore
G close (s) is stable if and only if 5
G l (s) = C I (s)P 3m (s)+ C PD (s)P 3m (s) 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
has positive gain margin and phase
Time (s)
margin. But as shown in Figure 12 the
gain margin of G l (s) is negative. With FIGURE 11— Time responses of the integer order PID two-inertia system by simulation.

WINTER 2007 ■ IEEE INDUSTRIAL ELECTRONICS MAGAZINE 11

 the existence of gear backlash the
100 designed integer order PID control
Magnitude (dB) system will easily be unstable and lead
50
to backlash vibration.
0
To be robust against backlash non-
linearity, several methods have been
−50 proposed, but their design processes
are very complicated. As an example,
−100 for PID control introducing a low-pass
101 102 103 104 105
filter Kd s/(Td s + 1) and redesigning
100 the whole control system with three-
50 inertia model can be a solution [26].
0 Due to the necessity of solving high
Phase (°)

order equations, the design is not

−50
easy to carry out. On the contrary,
−100 fractional-order PID k controller can
−150 achieve a straightforward design of
−200 robust control system against gear
101 102 103 104 105 backlash non-linearity. By changing
Freq. (rad/s) the D k controller’s fractional-order
FIGURE 12— Bode plot of Gl (s) in PID control. k the frequency response of G l (s)

100 100
Magnitude (dB)

Magnitude (dB)

50 50
0 0
−50 −50
−100 1 −100 1
10 102 103 104 105 10 102 103 104 105
100 100
50 50
Phase (°)

Phase (°)

0 0
−50 −50
−100 −100
−150 −150
−200 −200 1
101 102 103 104 105 10 102 103 104 105
Freq. (rad/s) Freq. (rad/s)
(a) (b)

100 100
Magnitude (dB)

Magnitude (dB)

50 50
0 0
−50 −50
−100 −100
101 102 103 104 105 101 102 103 104 105
100 100
50 50
Phase (°)

Phase (°)

0 0
−50 −50
−100 −100
−150 −150
−200 −200
101 102 103 104 105 101 102 103 104 105
Freq. (rad/s) Freq. (rad/s)
(c) (d)

FIGURE 13— Bode plots of Gl (s) in PIDk control: (a) k = 0.85, (b) k = 0.8, (c) k = 0.7, and (d) k = 0.5.

12 IEEE INDUSTRIAL ELECTRONICS MAGAZINE ■ WINTER 2007

can be directly adjusted (see Figure
13). As shown in Figure 14, letting k 50
be fractional order can improve
PID k control system’s gain margin
40
continuously. When k < 0.84 the PID k
control system will be stable; there-
fore with proper selected fractional- 30
order k the backlash vibration can be

Gain Margin (dB)

suppressed.
20
At the same time, for better back-
lash vibration suppression perform-
ance higher Dk controller’s order is 10
more preferable. As shown in open-
loop gain plots of 0.85, 0.8, 0.7 and 0.5 0
order PID k control systems (see
Figure 15), higher the D controller’s
order is taken lower the gain near gear −10
backlash vibration mode is. Based on
the tradeoff between robustness and −20
vibration suppression performance, 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
fractional-order 0.7 is chosen as D k Fractional Order k
controller’s best order. Here the short
memory principle method is used to FIGURE 14— Gain margin versus fractional order k.
realize the discrete D k controller.
As to the experimental results, first,
integer order PID speed control experi- 60
ment is carried out. As shown in Figure
16 the PID control system can achieve 40
satisfactory response when the back-
lash angle is adjusted to zero degree 20
(δ = 0), while severe vibration occurs
due to the existence of backlash non- 0
Magnitude (dB)

linearity (see δ = 0.6 case). This exper-

Plant
imental result is consistent with the −20
above analysis.
Figure 17 shows the experimental −40
results of fractional-order PID k control
0.85 0.5 0.7
with 0.7 and 0.5 order Dk controllers.
−60 0.8
Severe backlash vibration in the inte-
ger order PID control case is effective-
−80
ly suppressed. The control system’s
stability and robustness against gear
−100
backlash nonlinearity can be greatly 102 103 104
improved by the FOC approach. PID0.7 Freq. (rad/s)
control system has a good robustness
against backlash nonlinearity, while
FIGURE 15— Gain plots of the PIDk control systems and three-inertia plant.
the error in the experimental response
curves is for encoders’ coarse quanti-
zation. The intermittent tiny vibra- shown in Figure 18, PID1 control has providing more flexibility for designing
tions in lower order 0.5 case are due the most severe backlash vibration, robust control systems.
to its relatively high gain near gear while PID0.85 is on the verge of instabili-
backlash vibration mode in open-loop ty. PID0.95 and PID0.9 have intermediate Conclusions
frequency response. time responses. This experimental FOC opened a new dimension for con-
It is interesting to find the vibration result is natural since these orders are trol theory. The highly developed con-
suppression performance of fractional- continuous. The “interpolation’’ charac- trol theory based on integer order
order PID k control system shows some- teristic is one of the main points to differential equations shows quite dif-
what “interpolation’’ characteristic. As understand the superiority of FOC as ferent characteristics when it is

WINTER 2007 ■ IEEE INDUSTRIAL ELECTRONICS MAGAZINE 13

 60 60
50 50
ωm (rad/s)

ωm (rad/s)
40 40
30 30
20 20
10 10
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
30 30
25 25
ωl (rad/s)

ωl (rad/s)
20 20
15 15
10 10
5 5
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s) Time (s)
(a) (b)

FIGURE 16— Time responses of the integer order PID control: (a) δ = 0◦ and (b) δ = 0.6◦ .

60 60
50 50
ωm (rad/s)

ωm (rad/s)

40 40
30 30
20 20
10 10
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

30 30
25 25
ωl (rad /s)

ωl (rad /s)

20 20
15 15
10 10
5 5
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s) Time (s)
(a) (b)

FIGURE 17— Time responses of PIDk control: (a) k = 0.7 and (b) k = 0.5.

expanded into a fractional-order field. vide more flexibility and insight in con- and anomalous relaxation should be
At the same time, FOC is actually a trol design and thus give a clear-cut taken into account. This demand nat-
nice generalization of IOC theory. This approach for designing robust control urally needs fractional-order models
generalization gives huge space for system. The authors do believe some and hence fractional-order con-
researchers to see conventional IOC well-designed IOC system might in fact trollers, which are hopeful tools for
theory in a fresh light and find new and be a unconscious approximation of modeling and controlling complex
interesting things. FOC system. dynamic features.
From a practice viewpoint, the ideal And the dynamic features of “real” Finally, the authors would like to
fractional-order controllers can only be systems can be described more ade- end this introduction of FOC with the
realized by proper approximation with quately by fractional-order models. following expressive quotation:
finite differential or difference equa- Especially for light materials and flex- “… all systems need a fractional
tions. Namely, “design by FOC and real- ible structures, not only damping, time derivative in the equations that
ize by IOC” are inevitable. The but also other variety of physical describe them … systems have memo-
practical advantages for FOC is to pro- phenomena such as viscoelasticity ry of all earlier events. It is necessary

14 IEEE INDUSTRIAL ELECTRONICS MAGAZINE ■ WINTER 2007

 60 60
ωm (rad/s) 50 50

ωm (rad/s)
40 40
30 30
20 20
10 10
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

30 30
25 25
ωl (rad/s)

ωl (rad/s)
20 20
15 15
10 10
5 5
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s) Time (s)
(a) (b)

60 60
50 50
ωm (rad/s)

ωm (rad/s)
40 40
30 30
20 20
10 10
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

30 30
25 25
ωl (rad/s)

ωl (rad/s)

20 20
15 15
10 10
5 5
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s) Time (s)
(c) (d)

FIGURE 18— Continuity of PIDk control’s vibration suppression performance: (a) k = 1, (b) k = 0.95, (c) k = 0.9, and (d) k = 0.85.

to include this record of earlier events his Ph.D. degree, he joined Fanuc Ltd. and its industrial application to motion
to predict the future … and worked as a control engineer. In control, mechatronics, robotics, and
The conclusion is obvious and 2006, he moved to the Department of electric vehicles. He began and directed
unavoidable, Dead matter has memory. Mechanical and Aeronautical Engineer- the research of fractional-order control
Expressed differently, we may say that ing, University of California Davis as a in his laboratory from 1997. He is Fellow
nature works with fractional time postdoctoral researcher. He is a mem- of the IEEE and a member of IEE-Japan,
derivatives.”—S. Westerlund, “Dead ber of IEEE and IEE-Japan. JSME, SICE, RSJ, SAEJ, and JSST.
matter has memory!,” Physics Scripta, Yoichi Hori received the B.S., M.S.
vol. 43, pp. 174–179, 1991. and Ph.D. degrees in electrical engineer- References
With fractional-order calculus and ing from the University of Tokyo in 1978, [1] K.B. Oldham and J. Spanier, The Fractional Calcu-
lus. New York: Academic, 1974.
control, we may be able to extend a lot 1980, and 1983, respectively. He joined
[2] A. Tustin, J.T. Allanson, J.M. Layton, and R.J. Jake-
of new things. the Department of Electrical Engineer- ways, “The design of systems for automatic con-
ing, University of Tokyo, as a research trol of the position of massive objects,” Proc. IEE,
vol. 105-C, no. 1, pp. 1–57, 1958.
Biographies associate in 1983. He is a professor in
[3] S. Manabe, “The non-integer integral and its appli-
Chengbin Ma received the M.S. and the Electronics Division, Electrical Con- cation to control systems,” J. IEE Japan, vol. 80,
Ph.D. degrees in electrical engineering trol System Engineering, Institute of no. 860, pp. 589–597, 1960.
[4] S. Manabe, “The system design by the use of non-
from the University of Tokyo in 2001 Industrial Science, University of Tokyo. integer integrals and transport delay,” J. IEE
and 2004, respectively. After receiving His research interests are control theory Japan, vol. 81, no. 878, pp. 1803–1812, 1962.

WINTER 2007 ■ IEEE INDUSTRIAL ELECTRONICS MAGAZINE 15

[5] S. Manabe, “The system design by the use of a digital fractional-order controller to temperature [20] A. Charef, H. Sun, Y. Tsao, and B. Onaral, “Fractal
model consisting of a saturation and non- control,” in Proc. Acta Montanistica Slovaca, vol. 7, system as represented by singularity function,”
integer integrals,” J. IEE Japan, vol. 82, no. 890, no. 2, pp. 131–137, 2002. IEEE Tran. Automat. Contr., vol. 37, no. 9, pp.
pp. 1731–1740, 1962. [13] W. Li and Y. Hori, “Vibration suppression using 1465–1470, 1992.
[6] M. Axtelland and M.E. Bise, “Fractional calculus neuron-based pi fuzzy controller and fractional- [21] C. Ma and Y. Hori, “The time-scaled trape-
applications in control systems,” in Proc. IEEE order disturbance observer,” IEEE Trans. Ind. Elec- zoidal rule for discrete fractional order con-
1990 Nat. Aerospace and Electronics Conf., New tron., vol. 54, no. 1, pp. 117–126, 2007. trollers,” Nonlinear Dynam., vol. 38, pp.
York, USA, pp. 563–566, 1990. [14] I. Petras, B.M. Vinagre, L. Dorcak, and V. Feliú, 171–180, 2004.
[7] I. Podlubny, “Fractional differential equations,” in “Fractional digital control of a heat solid, experi- [22] J.A. Tenreiro Machado, “Discrete-time fractional-
Mathematics in Science and Engineering, vol. 198, mental results,” in Proc. 3rd Int. Carpatian Control order controllers,” Fractional Calculus Applied
New York: Academic, 1999. Conf., Malenovice, Czech Republic, pp. 365–370, Anal., vol. 4, no. 1, pp. 47–66, 2001.
[8] B.M. Vinagre, V. Feliú, and J.J. Feliú, “Frequency 2002.
domain identification of a flexible structure with [15] I. Petras, “Control of fractional-order Chua’s sys- [23] C. Ma, Y. Hori, “Time-domain evaluation of frac-
piezoelectric actuators using irrational transfer tem,” J. Elec. Eng., vol. 53, no. 7–8, pp. 219–222, 2002. tional order controllers’ direct discretization
function,” in Proc. 37th IEEE Conf. Decision and methods,” IEE J. Trans. Ind. Applicat., vol. 124, no.
[16] S. Manabe, “A suggestion of fractional-order 8, pp. 837–842, 2004.
Control, Tampa, FL, pp. 1278–1280, 1998. controller for flexible spacecraft attitude con-
[9] L.L. Ludovic, A. Oustaloup, J.C. Trigeassou, and F. trol,” Nonlinear Dynam., vol. 29, no. 1–4, pp. [24] C. Ma, Y. Hori, “Backlash vibration suppression
Levron, “Frequency identification by non-integer 251–268, 2002. control of torsional system by novel fractional
model,” in Proc. IFAC System Structure and Control [17] C.W. de Silva, Vibration Fundamentals and Prac- order P I D k controller,” IEE J. Trans. Ind.
Conf., Nantes, France, 1998, pp. 297–302. tice. London and New York: CRC, 2000. Applicat., vol. 124, no. 3, pp. 312–317, 2004.
[10] J.A. Tenreiro Machado, “Theory analysis and [18] A. Oustaloup, F. Levron, and F.M. Nanot, “Fre- [25] R.S. Barbosa and J.A. Tenreiro Machado,
design of fractional-order digital control sys- quency-band complex noninteger differentiator, “Describing function analysis of systems with
tems,” J. Syst. Anal. Modeling Simulation, vol. 27, characterization and synthesis,” IEEE Trans. Circ. impacts and backlash,” Nonlinear Dynam., vol.
no. 2–3, pp. 107–122, 1997. Sys. I, vol. 47, no. 1, pp. 25–39, 2000. 29, no. 1–4, pp. 235–250, 2002.
[11] A. Oustaloup, J. Sabatier, and X. Moreau, “From frac- [19] H. Sun, A. Abdelwahab, and B. Onaral, “Linear [26] H. Kobayashi, Y. Nakayama, and K. Fujikawa,
tal robustness to the CRONE approach,” Euro. Series approximation of transfer function with a pole of “Speed control of multi-inertia system only by a
Applied Ind. Math., vol. 5, pp. 177–192, Dec. 1998. fractional power,” IEEE Trans. Automat. Contr., PID controller,” IEE J. Trans. Ind. Applicat., vol.
[12] I. Petras, B.M. Vinagre, “Practical application of vol. 29, no. 5, pp. 441–444, 1984. 122-D, no. 3, pp. 260–265, 2002.

Introducing

BETA

Exclusive programming for technology professionals.

FEATURING:
Conference Highlights Reports from IEEE Spectrum
Author Profiles Special Features
Careers in Technology and more!

New programs every month!

Tune in. www.ieee.org/ieeetv
access via

IEEE.tv is made possible by the members of IEEE

887-Qb IEEEtv 7x4.75.indd 1 8/16/06 11:13:37

16 IEEE INDUSTRIAL ELECTRONICS MAGAZINE ■ WINTER 2007