Professional Documents
Culture Documents
net/publication/281985672
CITATION READS
1 50
3 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Roy Daou on 25 September 2015.
*
University of Bordeaux 1, laboratory IMS, Department LAPS,
electromechanical system, composed of two pumps, a uniform tank and a level sensor.
Both controllers are fractional orders where the first one is the generalized PID controller
and the second one is the CRONE (French acronym: Commande Robuste d’Ordre Non
Entier) controller. In a first time, the transfer function of the plant is presented after
identification (using the gray-box method) and simplification process. Then, the realisation
of both regulators is shown where the first controller (generalized PID) is obtained after
imposing the regulator model whereas the second one is deduced using the open-loop
constraints. At the end, a comparison between the behaviour of both controllers is made in
the frequency domain around some functional points of the plant as its transfer function is
nonlinear.
Fractional calculus.
1. INTRODUCTION
The fractional calculation was introduced in the 19th century by well-known mathematicians as Laplace,
Liouville, Abel, Riemann, Cauchy,... [1] but its application in the physics domain has not started before the
second half of the following century [2-7]. As far as the control systems are concerned, the fractional
calculus was introduced markedly by Manabe in 1961 [8] then by Oustaloup [9] in 1975 when he introduced
the control of laser by a regulator of order 3/2. This experiment had led to the development of the CRONE
command.
As for the PID controller, this regulator is one of the eldest and most used regulators. It was firstly developed
in its integer form by Ziegler and Nichols in 1942 [10]. Many approaches were developed in order to
calculate or to fine tune the optimal values of the regulator parameters [11-12].
As for the fractional or generalized PID controller, it has been implemented in the last part of the 20th
century. Two main approaches of the generalized PID were mainly developed: the cascade form and the
summation form. One of the most recent forms of the generalized PID was introduced by Poldubny [13]
using the summation approach. Hence, the transfer function of this controller is as follow:
C (s ) = k p + k i / s λ + k d s μ , (1)
where λ and η represent the differentiation and integration orders which can be real. When these two
parameters are equal to the unit, the well-known integer PID controller is obtained. Numerous algorithms
were introduced to tune the five parameters of this controller in order to get a more reliable and robust system
One of the essential reasons that had led to the use of the CRONE and the generalized PID controllers is the
huge difference in the synthesis method applied to get the transfer functions for both regulators. These
- the structure of the generalized PID is fixed a priori from traditional PID in cascade form;
- the structure of the CRONE is deduced a posteriori from the loop shaping.
Before starting the calculation of both controllers transfer function, the hydro-electromechanical composed of
a double-direction electropump (it fills and empties the main tank), two tanks and a level sensor is
introduced. A level disturbance is introduced by a second electropump. In order to calculate the plant transfer
function, a grey-box approach [19] is used to derive a model that represents the most the behaviour of this
hydro-electromechanical system. This approach combines physical knowledge and information coming from
test-bench tests and measurements. Once the plant transfer function is defined, both controllers are calculated
In more details, this article is composed as follow: in section 2, the synthesis method for both controllers is
presented. The specifications are first shown then the approaches used to calculate the generalized PID
controller and the CRONE controller are introduced. In section 3, the plant test bench based on the hydro-
electromechanical is presented. The identification method is developed then the plant transfer function is
deduced. Based on the plant transfer function, both the PID and the CRONE controllers are calculated with
respect to the specifications proposed in the previous section. At the end, the simulation results and the test
bench outputs are presented. Finally, section 4 presents a conclusion and some perspectives for this work.
2. SYNTHESIS METHODS
A. Performance specifications
• the bandwidth;
In the following, the study is limited in the frequency domain. So, the stability degree can be specified with
respect to the module margin MM, the gain margin MG or the phase margin MФ. The last one will be used in
where β(s) represents the open-loop transfer function and ωu the crossover frequency in the open loop
β ( jωu ) = 1 . (3)
M Φ ≥ M Φmin , (4)
where M Φ min represents the minimal acceptable value of the phase margin. In substituting relation (4) in
equation (2), the constraint of the open-loop argument for the crossover frequency ωu is easily computed as
follow:
The argument of the controller can be deduced for equation (5) in respect to equation (6)
where G(s) represents the plant transfer function. Hence, the reduced form of the controller argument
The bandwidth specification is also computed at the crossover frequency ωu. The objective of this constraint
is to fix the closed-loop dynamics speed. Hence, the frequency range that is concerned by this condition is
ωu ≥ ωu min , (8)
where ωu min represents the minimal acceptable value of ωu. Referring to equation (3) and knowing that:
β ( jω ) = C( jω ) G( jω ) , (9)
a controller gain constraint at the frequency ωu can be deduced from relation (8) as follow:
Concerning the rejection level of the measured noise, it is calculated using a specification applied to the
T ( jω ) = β ( jω ) 1 + β ( jω )
−1
∀ ω ≥ ωT , ≤ A, (11)
where A shows the desired rejected noise level for the given frequency ωT:
T ( jωT ) = A . (12)
If the value of ωT is chosen to be much bigger than ωu, relation (12) can be rewritten as:
∀ ω ≥ ωT , T ( jω ) ≈ β ( jω ) ≤ A . (13)
Hence, a new controller gain constraint around the frequency ωT is deduced from relation (11) as follow:
C ( jω ) ≤ A G ( jωT ) .
−1
∀ ω ≥ ωT , (14)
The rejection level of the output disturbance is used for low frequencies (ωS < ωu) and is computed using a
S ( jω ) = 1 + β ( jω )
−1
∀ ω ≤ ωS , ≤ B, (15)
where B represents the desired rejected output disturbance level for the frequency ωS:
S ( jωS ) = B . (16)
Hence, a new controller gain constraint around the frequency ωS is deduced from relation (15) as follow:
C ( jω ) ≥ (B G ( jω S ) ) .
−1
∀ ω ≤ ωS , (18)
R( jω ) = C ( jω ) (1 + β ( jω ))
−1
∀ ω ≥ ωR , ≤ D, (19)
R( jωR ) = D . (20)
As for the third condition, if the frequency ωR is much bigger than ωu, the relation (19) can be rewritten as:
∀ ω >> ωR , C( jω ) ≤ D . (21)
The constraints (14) and (21) represent almost the same aspect which is the controller gain at high
frequencies. Hence, they can be reduced to one constraint by choosing the lowest value of these two
relations. Thus,
Two forms are the most used for representing the generalized PID controller: the parallel form and the
cascade form. This second one is used in this article. The transfer function can be presented as follow:
C (s ) = C0 C I (s ) CD (s ) , (23)
⎛⎛
λ
⎞ ⎛ ⎞⎞
⎜⎜ s s ⎟⎟
C I (s ) = 1 +
⎟ ⎜
⎜⎜ ⎟ / ⎜ ⎟⎟ (24)
with ⎜⎜
⎜ ⎜⎝
⎝
ω ⎟⎟
i ⎠
⎜⎜
⎝
ω ⎟⎟ ⎟
i ⎠ ⎟⎠
⎛⎛
μ
⎞ ⎛ ⎞⎞
⎜⎜ s s ⎟⎟
CD (s ) = 1 +
⎟ ⎜
and ⎜⎜ ⎟ / 1+
⎜ ⎟⎟
, (25)
⎜⎜
⎜ ⎜⎝
⎝
ω⎟⎟
b⎠
⎜⎜
⎝
ω ⎟⎟ ⎟
h ⎠ ⎟⎠
where ωi, ωb et ωh are the transitional frequencies, C0 is a constant, λ the integration and μ the differentiation
real order.
As previously mentioned, the computation of the PID transfer function is done directly according to the
specifications.
In order to have a more reduced form of the PID transfer function, let’s introduce some new variables as
follow:
a = ωh / ωb , b = ωu / ωi and ωm = ωb ωh , (26)
where ωm represents the frequency for which the controller phase is maximal (e.g., ϕm = Max[argC(jω)] =
μ
λ ⎛⎜ ⎞
⎟
⎛
1 + b s /ω ⎞
1 + a s /ω
C (s ) = C
⎜ ⎟ ⎜
⎜ u ⎟ u ⎟
⎜ ⎟ . (27)
b s /ω ⎛ ⎞⎟
0 ⎜ ⎟⎟
⎜
⎝ u ⎠
⎜
⎜ 1 + s / ⎜⎜ a ω ⎟
u ⎟⎠ ⎟
⎝ ⎝ ⎠
μ/2
⎧ ⎛ ⎞
λ /2 ⎜
⎪ C ( jω ) =C ⎛⎜1 + (bω / ω ) 2 ⎞⎟ (bω / ω ) − λ ⎜
⎜ 1 + a ⎛⎜
⎝
ω / ω ⎞⎟ 2 ⎟
u⎠ ⎟
⎟
⎪⎪ ⎝ ⎠ −
0 u u ⎜ ⎟
⎜1 + a
1 ⎛⎜
ω / ω ⎞⎟ 2 ⎟
⎨ ⎝ ⎝ u⎠ ⎠ (28)
⎪arg C ( jω )=λ (arctan (bω / ωu ) − π / 2 )
⎪
⎩⎪ ( ( )
+ μ arctan a ω / ωu − arctan ω / a ωu ( ( )))
At the open-loop crossover frequency ωu, relation (28) can be rewritten as follow:
⎧Cu = C ( jωu ) = C0 (1 + b2 )λ / 2 b − λ a μ / 2
⎪
⎪
⎨
⎪ϕ u = arg C ( jωu ) = λ (arctan(b ) − π / 2 )
. (29)
⎪
⎩ ( ( )
+ μ arctan a − arctan 1 / a ( ))
If b >> 1 (minimization of the noise effect caused by the integration due to the phase delay of ωu), system
(29)becomes:
⎧Cu = C ( jωu ) = C0 a μ / 2
⎪
⎨ , (30)
⎪ϕ = arg C ( jω ) = λ (arctan(b ) − π / 2 ) + ϕ
⎩ u u m
After recalculating the controller module and phase at central frequency ωu, we obtain for low frequencies (ω
<< ωu)
Considering the system (28), the constraints of relations (7), (10), (14), (18) and (21), valid independently of
the controller nature, can be rewritten to suit best the generalized PID controller as follow:
λ /2
C0 ⎛⎜1 + b2 ⎞⎟ b − λ a μ / 2 ≥ G ( jωu min ) ,
−1
(35)
⎝ ⎠
C0 a μ ≤ A G ( jωT )
−1
∀ ω ≥ ωT , (36)
C0 (ωu / (b ω )) ≥ (B G ( jωS ) )
λ −1
∀ ω ≤ ωS , (37)
In order to determine the optimal values of the parameters vector θ = [C0, b, a, λ, μ], two phases are required.
The first one depends on the initial values of the vector θ and the second phase involves the search for the
In order to accomplish the first phase, the five following steps must be achieved.
3 – One deduces ϕ m = M Φ − π − Φ 0 ;
4 – One calculates the value of a with respect to the relation (31): a = (1 + sin ϕ m ) / (1 − sin ϕ m ) ;
C0 = (a μ / 2 ρ 0 ) .
−1
5 – Knowing that β ( jωu ) = 1 , one deduces the value of C0 as follow:
The second phase consists on determining the optimal values of the vector θ. Two methods can be used in
this case: the genetic algorithms or the optimization toolbox of Matlab and its function fmincon.
In this article, the second method is used. This function requires a large number of inputs and returns the five
values of the vector θ once all five relations ((34) to (38)) are valid. The initial vector and the lower and
upper bound vectors have a big impact on the output values. Note that, in order to emphasize the fractional
behaviour of the generalized PID controller, the value of the integrator and differentiator orders, λ and η,
C. CRONE Controller
The second approach used to design a controller, presented in this article, is based on defining this regulator
with respect to the open-loop constraints (robust loop shaping). The CRONE controller is designed using this
method.
Hence, the first step consists on defining the necessary specifications for the synthesis of the nominal plant
transfer function. The frequency approach is also used in this part as the calibration of the sensitivity
functions is easier.
The second step consists on reassigning the frequency closed-loop specifications into open-loop frequency
specifications for the nominal plant. These new conditions take into account the plant behaviour at:
• middle frequencies, especially around the frequency ωu, to get the stability degree robustness;
• high frequencies to have good input plant sensitivity.
Once the behaviour of the system in the open-loop is defined, a decision should me made concerning the
CRONE generation that will be used. This decision depends on the plant uncertainties. If these uncertainties
a related to the gain while the phase is constant, the second generation CRONE control is sufficient [9]. If
both the gain and the phase vary, the third generation CRONE control must be used. In this paper, we are
interested in the second generation as the third one presents complex differentiation and integration orders,
and the use of special programs is a must to solve such equations [9].
Hence, the open-loop behaviour due to the plant gain uncertainties leads to the following transfer function:
nb n
⎛ 1 + s /ω b ⎞ ⎛ 1 + s /ω h ⎞
β ( s ) = β 0 ⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟ (1 + s /ω h )−nh , (39)
⎝ s /ω b ⎠ ⎝ 1 + s /ω b ⎠
where ωb and ωh represent the low and high transitional frequencies, n is the fractional order varying between
1 and 2, nb and nh are the asymptotic order behaviours for low and high frequencies and β0 is a constant that
assure a unit gain at the frequency ωu. This constant is calculated as follow:
(
β 0 = (ωu /ω b )2 1 + (ωu /ω b )2 )( n −nb ) / 2
(1 + (ω
u )
2 ( nh −n ) / 2
/ω h ) (40)
the CRONE transfer function Ccrone(s) is deduced for the nominal plant value, which is to say:
Ccrone (s ) = β ( s ) / G (s ) . (42)
The test bench, used to test the behaviour of both controllers, is a hydro-electromechanical system constituted
of:
- a double-direction electropump that can fill or empty the main tank,
- a secondary tank,
- a level sensor that measures the water level in the main tank,
- and a data acquisition board that enables the communication between the hydro-electromechanical system
Figure 1 shows the physical test bench. Figure 2 represents the closed-loop functional block diagram of the
hydro-electromechanical system.
Advanced control design requires a model that describes process behaviour adequately. Such a model can be
constructed using physical modelling (white-box model) or statistical identification techniques (black-box
model). Both have their disadvantages: physical models are rigorous and thus often expensive to construct,
while black-box model structures are not necessarily compatible with physical reality. Since both approaches
have undeniable merits as well, their combination seems to be attractive and rewarding.
In this study, a grey-box approach [19] is used to derive a model of the plant. This approach combines
physical knowledge and information embedded in data. More precisely, the model structure is deduced from
physical principles and the values of the unknown parameters of the model structure are estimated from
measured data series around three operating points. The structure model is given by
H ( s ) = k / (s (1 + s / ω0 )) , (43)
The parameters k and ω0 are estimated from “System Identification Toolbox” of Matlab (output error
estimation method). A PRBS (Pseudo Random Binary Signal) is used as the input signal. Its amplitude varies
between ± 2.5V whereas the minimal allowed period is 2 seconds and the number of samples 16. This input
signal is applied to the electropump1 around three operating points (three levels in the tank: 2/4/6 cm) in
The values of k and ω0 for each functional point are given in table 1.
As the transitional frequency values ω0 for each operating points are very close, the two controllers are
[ ]
k ∈ 2.24 × 10−4 , 8.30 × 10−4 , ω0 = 3.36 × 10−3 rad/s. (44)
Figure 3.a presents the Bode plots of H(jω) obtained for the three operating points.
For the controller synthesis, the nominal model is arbitrary fixed by the third operating point (knom = 2.24 10-4
V/V).
C. User constraints
Once the uncertain model transfer function is defined, the user constraints applied to synthesize both
where QTmax is the maximum of the complementary sensitivity function (resonance peak);
D. Controllers synthesis
Considering the model transfer function and its uncertainties, the user specifications and the synthesis
methods for the generalized PID and the CRONE controllers, the generalized PID transfer function is as
follow:
0.2
⎛ ⎞ 0.698
⎜
1 + s / 8.3 10− 4 ⎟⎟ ⎛
1 + s / 7.95 10− 3 ⎞⎟⎟
CPID (s ) = 50 ⎜ ⎜
⎜ ⎟ ⎜ , (45)
⎜
⎜ s / 8.3 10 −4 ⎟
⎟
⎜⎜
⎝ 1 + s / 1.47 10− 2 ⎟⎟⎠
⎝ ⎠
⎛ ⎞ 0.5435
⎜ 1 + s / 36 10 − 4 ⎟ ⎛⎜ 1 + s / 6.98 10
−4 ⎞
Ccrone (s ) = 2.41 10 5 ⎟
⎜ ⎟ ⎜ ⎟ (46)
⎜ −7 ⎛ ⎟ ⎜
⎜
⎝
s / 4.87 10 ⎜⎝1 + s / 0.5014 ⎟⎠ ⎟⎠ ⎜⎝ 1 + s / 0.5014 ⎟⎟⎠
⎞
E. Simulation results
Figures 4 present the open-loop Bode plots obtained with the generalized PID controller (Fig.4.a) and the
CRONE controller (Fig.4.b) around the first (in green) and third (in blue) operating points.
Figures 5 also show the open-loop Nichols loci obtained with the generalized PID controller (Fig.5.a) and
with the CRONE controller (Fig.5.b) around the first (in green) and third (in blue) operating points.
With the CRONE controller, the phase margin (45°) and the module margin (3 dB) remain constant for the
extreme values of k: illustration of the stability degree robustness versus gain uncertainties in frequency
domain.
Figures 6, 7 and 8 present the different sensitivity functions obtained with the generalized PID controller
(Fig.6.a, Fig.7.a, Fig.8.a) and with the CRONE controller (Fig.6.b, Fig.7.b, Fig.8.b) around the first (in green)
and third (in blue) operating points. For the nominal operating point (3rd point, in blue), all performance
bounds are respected, except for the input sensitivity function obtained with the generalized PID controller
(Fig.8.a). For the first operating point (in green), all performance bounds are also respected, except for the
stability degree (Fig.7.a, in green) and for the input sensitivity function obtained with the generalized PID
controller (Fig.8.a).
Figure 9 shows the closed-loop step responses obtained with the generalized PID controller (Fig.9.a) and
with the CRONE controller (Fig.9.b) around the first (in green) and third (in blue) operating points. With the
CRONE controller, the first overshoot and the damping remain constant for the extreme values of k:
illustration of the stability degree robustness versus gain uncertainties in time domain.
Figures 10 and 11 present the measured level in the main tank and the control signal in rejection mode
obtained with the generalized PID controller (Fig.10.a, Fig.11.a) and with the CRONE controller (Fig.10.b,
Fig.11.b) around the first (in red) and third (in blue) operating points. One notes that:
- the static error is nil with the CRONE controller, it is not the case with the generalized PID controller;
- the control signal remains smaller than the saturation value (±6V).
4. CONCLUSION
In this article, the synthesis methods of two fractional order controllers were presented. These controllers
treated in this work are the generalized PID and the CRONE regulators. The main difference between these
two controllers is the way the structure is defined. The generalized PID controller structure is fixed a priori
whereas the CRONE controller is deduced a posteriori using the loop shaping.
The performance specifications have been firstly translated into open-loop constraints in order to deduce the
CRONE controller transfer function. Then, they were translated into controller constraints for the calculation
of the generalized PID transfer function. In order to determine the five optimal parameters of the generalized
PID, a method with two steps has been proposed:
- in the first step, the three optimal parameters of the traditional PID controller were calculated after
supposing that the integrator and differentiator orders are integers and equal to the unit. These three
- in the second step, an optimization process is simulated under MatLab is used in order to determine the five
optimal parameters.
The last part of this article contains a comparative study of a nonlinear hydro-electromechanical plant
controlled by these two regulators that were calculated based on the same specification. The results obtained
in simulation and with the test bench have shown the interest of this study.
Plant : control tension->sensor tension 100
20
Generalized PID
0 80 CRONE
Gain (dB)
-20 60
Gain (dB)
-40
40
-60
20
-80 -5 -4 -3 -2 -1 0
10 10 10 10 10 10
-100
-90
50
1st point
2nd point
Phase (deg.)
3rd point
Phase (deg)
0
-135
-50
-180 -100
-4 -3 -2 -1 -5 -4 -3 -2 -1 0
10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/s)
(a) (b)
Fig. 3 – Frequency responses of the plant for the three functional points (blue: first point; green: second
point; red: third point) (a) and Bode plots of the PID (in blue) and the CRONE (in red) controllers (b)
100 100
3rd point 3rd point
50 50 1st point
1st point
Gain (dB)
Gain (dB)
0 0
-50 -50
-100 -100
-5 -4 -3 -2 -1 0 -5 -4 -3 -2 -1 0
10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/s)
-90 -90
Phase (deg)
Phase (deg.)
-135 -135
-180 -180
-225 -225
-5 -4 -3 -2 -1 0 -5 -4 -3 -2 -1 0
10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/s) Frequency (rad/s)
(a) (b)
Fig. 4 – Open-loop Bode plots for the generalized PID controller (a) and for the CRONE controller (b)
20 1 dB 20 1 dB
Open-Loop Gain (dB)
Open-Loop Gain (dB)
10 3 dB 10 3 dB
6 dB 6 dB
0 0
-10 -10
-20 -20
-30 -30
-40 -40
-270 -225 -180 -135 -90 -45 0 -270 -225 -180 -135 -90 -45 0
Open-Loop Phase (deg) Open-Loop Phase (deg)
(a) (b)
Fig. 5 – 0pen-loop Nichols loci obtained with the generalized PID controller (a)
0 0
-10 -10
-20 dB -20 dB
-20 -20
3rd point
1st point
-30 -30
-40 -40
-50 -50
-4 -3 -2 -1 0 -4 -3 -2 -1 0
10 10ω S = 0.00108 rad/s10 10 10 10 10ω S = 0.00108 rad/s10 10 10
Frequency (rad/s) Frequency (rad/s)
(a) (b)
Fig. 6 – Sensitivity functions with: (a) the generalized PID controller, (b) the CRONE controller
10 10
QTmax = 3 dB 3rd point QTmax = 3 dB 3rd point
1st point 1st point
0 0
- 3 dB - 3 dB
Complementary Sensitivity function (dB)
-10 -10
-20 dB -20 dB
-20 -20
-30 -30
-40 -40
-50 -50
-60 -60
-3 -2 -1 0 -3 -2 -1 0
10 10 ω umin = 0.0108 rad/s 10 10 10 10 ω umin = 0.0108 rad/s 10 10
Frequency (rad/s) ω T = 0.108 rad/s Frequency (rad/s) ω T = 0.108 rad/s
(a) (b)
Fig. 7 – Complementary sensitivity functions with: (a) the generalized PID controller, (b) the CRONE
controller
60 60
50 50
Input Sensitivity function (dB)
30 30
20 20
10 10
3rd point 3rd point
1st point 1st point
0 0
-3 -2 -1 0 1 2 -3 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10 10 10
ω R = 10.8 rad/s ω R = 10.8 rad/s
Frequency (rad/s) Frequency (rad/s)
(a) (b)
Fig. 8 – Input sensitivity functions with: (a) the generalized PID controller, (b) the CRONE controller
Step reponses with PID controller Step reponses with CRONE controller
1.6 1.6
3rd point 3rd point
1st point 1st point
1.4 1.4
1.2 1.2
1 1
vm(t) (volt)
vm(t) (volt)
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 1400
Time (s) Time (s)
(a) (b)
Fig. 9 – Step responses with: (a) the generalized PID controller, (b) the CRONE controller
0.005 0.005
0 0
-0.005 -0.005
Measured level (V)
-0.015 -0.015
-0.025 -0.025
0 500 1000 1500 2000 2500 3000 3500 0 500 1000 1500 2000 2500 3000 3500
Time (s) Time (s)
(a) (b)
Fig. 10 – Output responses in rejection mode with: (a) the generalized PID controller,
6 6
control signal for 3rd point control input for 3rd point
5 control signal for 1st point 5 control input for 1st point
input disturbance input disturbance
4 4
Control signal (V)
3 3
2 2
1 1
0 0
-1 -1
-2 -2
0 500 1000 1500 2000 2500 3000 3500 0 500 1000 1500 2000 2500 3000 3500
Time (s) Time (s)
(a) (b)
Fig. 11 – Control signal responses in rejection mode with: (a) the generalized PID controller,
[2] Cois O. Systèmes linéaires non entiers et identification par modèle non entier : application en thermique,
[3] Lin J. Modélisation et identification de systèmes d'ordre non entier, Thèse de Doctorat, Université de
Poitiers, 2001.
[4] Miller K.S. and Ross B. An introduction to the fractional calculus and fractional differential equations, A
[5] Oldham K.B. and Spanier J. The fractional calculus, Academic Press, New-York and London, 1974.
[6] Oustaloup A. La dérivation non entière : théorie, synthèse et applications, Edition Hermès, Paris, 1995.
[7] Samko S. G., Kilbas A. A. and Marichev O. I. Fractional integrals and derivatives: theory and
[8] Manabe S. The non-integer integral and its application to control systems. ETJ of Japan, Vol. 6, pp.83-87,
1961.
[9] Oustaloup A. Etude et réalisation d'un système d'asservissement d'ordre 3/2 de la fréquence d'un laser à
[10] Ziegler J. G. and Nichols N. B. Optimum settings for automatic controllers. Transactions of ASME,
[11] Asröm K. J. and Hägglund T. The future of PID control. IFAC Workshop on Digital Control, Past,
Present and Future of PID Control, pp. 19-30, Terressa, Spain, 2000.
[12] Chen Y. Q., Hu C. H. and Moore K. L. Relay feedback tuning of robust PID controllers with iso-
damping property. 42nd IEEE Conference on Decision and Control, Maui, Hawaii, USA, 2003.
[13] Podlubny I. Fractional-order systems and PID-controllers. IEEE Transaction on Automatic Control,
[14] Vinagre B. M., Podlubny I., Dorcak L. and Feliu V. On fractional PID controllers: a frequency domain
approach. IFAC Workshop on Digital Control, Past, Present and Future of PID Control, pp. 53-58,
[15] Caponetto R., Fortuna L. and D. Porto. Parameter tuning of a non integer order PID controller.
Proceedings of 5th International Symposium on Mathematical Theory of Networks and Systems, Notre
[16] Leu J. F., Tsay S. Y. and Hwang C. Design of optimal fractional-order PID controllers. Journal of the
[17] Monje C. A., Calderon A. J. and Vinagre B. M. PI vs fractional DI control : first results. Controlo
2002, 5th Portuguese Conference on Automatic Control, pp. 359-364, Aveiro, Portugal, 2002.
[18] Chen Y. Q., Hu C. H., Vinagre B. M. and Monje C. A. Robust PID controller tuning rule with iso-
[19] Tulleken H.J.A.F. Grey-box modelling and identification using physical knowledge and bayesian