A review of tuning methods for

fractional PIDs ?
Duarte Valério ∗ José Sá da Costa ∗
∗
IDMEC/IST, Technical University of Lisbon, Portugal
(e-mail: [email protected], [email protected]).

Abstract: This paper presents a review of tuning methods found in the literature for fractional
PIDs. Analytical, numerical and rule-based tuning methods are considered.

Keywords: Fractional PIDs, fractional control, fractional derivatives, tuning rules.

1. INTRODUCTION
 
I λπ cos(−π + ϕm − ∠G(jωgc ))

λ

 P + λ
+ Dω gc cos =
ω 2 |G(jωgc )|

gc
This paper presents a review of tuning methods found in 
I

λπ sin(−π + ϕ m − ∠G(jωgc ))
λ
the literature for fractional PID (also known as PIλ Dµ  − λ + Dωgc sin =


ωgc 2 |G(jωgc )|
or FPID) controllers. A fractional PID is a controller for
(5)
a single-input, single-output plant G in a feedback loop,
1
with a transfer function that is a generalisation of that of and a gain margin gm = |C(jωpc )G(jω pc )|
at phase crossover
well-known and widely used PID controllers, and is given frequency ωpc : ∠[C(jωpc )G(jωpc )] = −π
by 1
I Dsµ+λ + P sλ + I C(jωpc )G(jωpc ) = − ⇔ (6)
C(s) = P + λ + Dsµ = , λ, µ ∈ R+ gm
s sλ 1 jπ
(1) gm e
C(jωpc ) = ⇔ (7)
When λ = µ = 1, C reduces to a PID controller. When |G(jωpc )|ej∠G(jωpc )
D = 0, C reduces to a fractional PI controller. Tuning λπ
Ie−j 2 ej(π−∠G(jωpc ))
λ j λπ
methods for fractional PIDs found in the literature can be P+ λ
+ Dω pc e 2 = ⇔ (8)
ωpc gm |G(jωpc )|
divided into analytical (section 2), numerical (section 3)
and rule-based (section 4). Many tuning methods can be
 
I λπ cos(π − ∠G(jωpc ))

λ
P + + Dωpc cos =

turned into self-tuning methods if combined with properly 
ω λ 2 gm |G(jωpc )|
pc
devised tests (using e.g. a relay feedback); this possibility 
I

λπ sin(π − ∠G(jωpc ))
λ
is addressed by some of the authors reviewed, but was  − λ + Dωpc sin =


not explicitly taken into account below. No examples of ωpc 2 gm |G(jωpc )|
application of the methods reviewed are given; these can (9)
be found in the literature quoted. The four conditions (5) and (9) allow finding the four pa-
rameters P , I, λ and D. Should λ 6= µ, a fifth specification
2. ANALYTICAL METHODS may be fulfilled.
Caponetto et al. (2004) propose choosing freely an order
2.1 Phase and gain margins λ = µ > 1 (since orders λ and dλe have similar effects in
what reference tracking is concerned, this means that ramp
Vinagre (2001) proposes setting λ = µ and imposing a references will be followed without steady-state error) and
phase margin ϕm = 180◦ + ∠[C(jωgc )G(jωgc )] at gain imposing a phase margin ϕm at gain crossover frequency
crossover frequency ωgc : |C(jωgc )G(jωgc )| = 1 ωgc . This allows choosing freely yet another parameter (P ,
I or D) and then determining the other two from (5).
C(jωgc )G(jωgc ) = ej(−π+ϕm ) ⇔ (2)
e j(−π+ϕm ) Zhao et al. (2005) further develop (5) and (9) for the
C(jωgc ) = ⇔ (3) case when the requirement λ = µ is dropped and
|G(jωgc )|ej∠G(jωgc ) G(s) = a1 sα +a12 sβ +a3 .
λπ
Ie−j 2 λ j λπ ej(−π+ϕm −∠G(jωgc ))
P+ λ
+ Dω gc e 2 = ⇔ (4) Maione and Lino (2007) develop a similar idea, based upon
ωgc |G(jωgc )| the symmetrical optimum method, for fractional PIs when
K
? This work is supported by the Portuguese Government under G(s) = s(1+T s) .
project PTDC/EME-CRO/70341/2006, Fundação para a Ciência e
Tecnologia, and by the Portuguese Government and FEDER under 2.2 Flat phase
program “Programa de Financiamento Plurianual das Unidades de
I&D da FCT para as atividades de investigação do laboratório Chen et al. (2006) propose that a fractional PI controller
associado LAETA” (POCTI-SFA-10-46-IDMEC). be tuned combining gain and phase margin requirements
(as seen above) with a flat phase for the open-loop around d
e
the frequency ωc at which the sensitivity circle touches the
F G* G
Nyquist curve; that is to say,
d∠[C(jω)G(jω)]
=0⇔ (10)
dω ω=ωc G'

d∠C(jω) d∠G(jω)
=− (11)
dω ω=ωc dω ω=ωc
Notice that, for a fractional PI controller,
I πλ d
λ sin e
∠C(jω) = arctan ω I 2 πλ ⇒ (12) C G
P + ωλ cos 2
d∠C(jω) 1
= I2
×
dω sin2 πλ
1 + 2 I 2 ω2λ2 πλ 22P I πλ Fig. 1. Block diagram for IMC (top) and block diagram
P + cos + cos
ω2λ 2 ωλ 2
equivalent to that above (bottom)
(− ωλ+1
Iλ
sin πλ
2 )(P + ωIλ cos πλ2 )−
( ωIλ sin πλ )( Iλ
− λ+1 cos πλ
)
× 2
2 ω 2
(13) This is equivalent to the second block diagram of figure 1
2P I
P2 + ωI2λ cos2 πλ 2 + ω λ cos 2
πλ
if controller C is given by
I2λ F G∗
− ωPλ+1
Iλ
sin πλ πλ
2 − ω 2λ+1 sin 2 cos 2
πλ
= (14) C= (18)
2
P 2 + ωI2λ cos2 πλ 2P I πλ I2 2 πλ 1 − F G ∗ G0
2 + ω λ cos 2 + ω 2λ sin 2
I2λ C is not, in the general case, a PID or a fractional PID,
− ωPλ+1
Iλ
sin πλ2 − 2ω 2λ+1 sin(πλ) but in some cases it will if
= 2 (15)
P 2 + ωI2λ + 2P I
ω λ cos 2
πλ
K
G= e−Ls (19)
Monje et al. (2006) make
a similar proposal for a controller 1 + sα T
λ
given by C(s)=kP 1+ T1I s (1+TD s)µ , referred to by the au- Firstly, let
thors as a fractional PID, but different from (1).
1
2.3 Dominant poles F= (20)
1 + sTF
1 + sα T
For a plant given by G∗ = (21)
K
−Ls
be K
G(s) = (16) 0
G = (1 − sL) (22)
sα + a 2 sβ + a 3 1 + sα T
Vinagre (2001) proposes making λ = β, µ + λ = α, Notice that the delay of G was neglected in G∗ but not
P = Da2 and I = Da3 , and thus the closed-loop becomes in G0 , where an approximation consisting of a truncated
Dsα +Da2 sλ +Da3 be−Ls McLaurin series has been used. Then (18) becomes
C(s)G(s) sβ sα +a2 sβ +a3
= Dsα +Da2 sλ +Da3 1 T
1 + C(s)G(s) 1+ be−Ls
K(TF +L) K(TF +L)
sβ sα +a2 sβ +a3
C= + (23)
Dbe −Ls s s1−α
= (17) This is a fractional PID controller with the proportional
sβ + Dbe−Ls
D is then chosen to ensure that the closed-loop is as fast part equal to zero.
as desired (and possible), or, in other words, to place the Secondly, let
dominant poles of the closed-loop.

2.4 Internal Model Control (IMC) F =1 (24)

1 + sα T
G∗ = (25)
Valério and Sá da Costa (2006) show that the Internal K
Model Control (IMC) methodology may, in some cases, 0 K 1
be used to obtain PID or fractional PID controllers. IMC G = α
(26)
1 + s T 1 + sL
corresponds to the control scheme at the top of figure 1,
where G∗ is an inverse of G (or at least a transfer function (18) becomes
as close as possible to the inverse of plant G), G0 is a 1 1 T
T
model of G and F is some judiciously chosen filter. If G0 C= + KL + KL + sα (27)
were exact, the error e would be equal to disturbance d. K s s1−α K
If, additionally, G∗ were the exact inverse of G and F If one of the two integral parts is neglectable, (27) will be a
were unity, control would be perfect. Since no models are fractional PID controller. (The price to pay for neglecting a
perfect, e will not be exactly the disturbance. That is also term is some possible slight deterioration in performance.)
exactly why F exists and is usually a low-pass filter: to Finally, if a Padé approximation with one pole and one
reduce the influence of high-frequency modelling errors. It zero is used in G0 ,
also helps ensuring that product F G∗ is realisable.
 F =1 (28) 3. NUMERICAL METHODS
1 + sα T
G∗ = (29) Numerical methods for tuning fractional PIDs rely on
K
K 1 − sL/2 the numerical evaluation of an objective function, that
G0 = (30) measures the extent to which several design specifications
1 + sα T 1 + sL/2 are fulfilled, weighting them as the control designer sees fit.
then (18) becomes A numerical minimisation method is then employed to find
1 T the controller parameters that optimise the value of the
1 T α objective function. For instance, Monje et al. (2004, 2008)
C= + KL + KL + s (31)
2K s s1−α 2K propose that the simplex search minimisation method be
Again, (31) will be a fractional PID if one of the two used to minimise one specification while up to four others
integral parts is neglectable. are considered as constraints; the five requirements (five,
because five are the controller parameters) are to be chosen
among the following suggested six:
2.5 The F-MIGO algorithm
(1) a minimum order for dλe (for the same reason men-
Chen et al. (2008) propose tuning fractional PI controllers tioned in section 2.1);
using the fractional Ms constrained integral gain optimisa- (2) values for the gain crossover frequency ωgc and the
tion (F-MIGO) algorithm, that imposes maximum values phase margin ϕm ;
for the sensitivity and complementary sensitivity functions (3) values for the phase crossover frequency ωpc and the
1 gain margin gm ;
Ms = max (32) (4) phase flatness at ωgc , d∠[C(jω)G(jω)] = 0;
ω 1 + C(jω)G(jω) dω
ω=ωgc
C(jω)G(jω) (5) high frequency noise rejection, ≤ H, C(jω)G(jω)
Mp = max (33) 1+C(jω)G(jω)
ω 1 + C(jω)G(jω)
∀ω ≥ ωh ;
by preventing the Nyquist plot of the open-loop control (6) output disturbance rejection at low frequencies,
function from entering the circle with centre at (−C, 0) 1
and radius R: 1+C(jω)G(jω) ≤ N , ∀ω ≤ ωl .
Ms − Ms Mp − 2Ms Mp2 + Mp2 − 1 Another example is provided by Maiti et al. (2008a,b), who
C= (34)
2Ms (Mp2 − 1) propose the use of particle swarm optimisation to optimise
Ms + M p − 1 the control loop’s time response.
R= (35)
2Ms (Mp2 − 1)
4. TUNING RULES
The algorithm, in short, is as follows:
(1) Choose desired values for Ms and Mp ; usual ranges It should be noticed that all authors acknowledge that
are given as Ms ∈ [1.3, 2.0] and Mp ∈ [1.0, 1.5], and tuning rules provide controller parameters that can often
typical desired values as Ms = 1.4 and Mp = 2. be improved by trial and error fine tuning.
(2) Arbitrate a value for λ.
(3) Solve 4.1 Plants with an S-shaped step response
 
  ( d|G(jω)| ) =[G(jω)]
=[G(jω)] dω λ |G(jω)| This section concerns plants with an S-shaped step re-
2R  C +R − −C =0
|G(jω)| |G(jω)| ω dω sponse, such as the one seen in figure 2 for a unit-step
(36) input, from which a steady-state gain K, an apparent delay
in order to ω. L and a characteristic time-constant T may be determined
(4) Calculate (theoretically, numerically or graphically).

cos πλ
2 C=[G(jω)] cos πλ2 C<[G(jω)]
P =R + 2
−
sin πλ
2
|G(jω)| sin πλ
2
|G(jω)|2 K
nt

(37)
i
po
n
io

λ λ
Rω C=[G(jω)]ω
ct
le

I =− − (38)
nf
ti

|G(jω)| sin πλ |G(jω)| 2 sin πλ

ta
output

2 2
en
ng

If either P or I are negative, go back to step 2 and

ta

choose another value for λ.

(5) Check if desired values for Ms and Mp are verified
•inflection point
and if the loop is stable.
This procedure should be repeated for several different 0
0 L L+T
values of λ and the results compared. time

While the algorithm is analytical, solving (36) is likely

to have to be done numerically, using e.g. the Newton- Fig. 2. S-shaped unit-step response
Raphson method.
 4.2 Plants with a critical gain

Valério and Sá da Costa (2007) propose three sets of

tuning rules for plants with a critical gain, that is to
say, plants that, when in a feedback control loop with
proportional gain, show, for a particular gain (the critical
output

gain Kcr ), oscillations (the critical period Pcr being the

period thereof) with an amplitude that does not decrease
or increase with time, as shown in figure 3. The first
and second sets of rules intend respectively to achieve
0
specifications A e B from table 1, as was the case for the
P
cr rules from section 4.1. The third set of rules was developed
0 to cope with plants with a pole at the origin; they intend
time
to achieve specifications A from table 1 for plants given by
K
Fig. 3. Plant output with critical gain control s(s+τ1 )(s+τ2 ) . Parameters are given in table 2. These rules
may be used when parameters are in the ranges given in
the tables.
Specs. ωcg ϕm ωh ωl H N
A 0.5 rad/s 2/3 rad 10 rad/s 0.01 rad/s −10 dB −20 dB
B 0.5 rad/s 1 rad 10 rad/s 0.01 rad/s −20 dB −20 dB 4.3 Unit time change
Table 1. Specifications from Valério and Sá da
Costa (2006, 2007) It can be easily seen that the tuning rules in this section
do no lead to control actions invariant when the time
unit is changed. (Compare this with e.g. Ziegler-Nichols
Valério and Sá da Costa (2006, 2007) propose two sets
tuning rules.) So it is always possible to develop several
of tuning rules for such plants, respectively to ensure
controllers for a given plant using only one set of rules,
specifications A and B from table 1. The rules were
by reckoning variables with different time units: seconds,
conceived from fractional PIDs tuned as in (Monje et al.,
minutes, hours, or in general, any arbitrary time unit.
2004) for plant
K
e−Ls (39) REFERENCES
1 + sT
The parameters of the polynomials involved are given in Bhambhani, V., Chen, Y., and Xue, D. (2008). Optimal
table 2. This means that fractional order proportional integral controller for vary-
T ing time-delay systems. In IFAC World Congress, 4910–
P = − 0.0048 + 0.2664L + 0.4982 4915.
K
 2 Caponetto, R., Fortuna, L., and Porto, D. (2004). A new
T TL
+ 0.0232L2 − 0.0720 − 0.0348 (40) tuning strategy for a non integer order PID controller.
K K In Fractional Differentiation and its Applications. Bor-
and so on. These rules may be used when parameters are deaux.
in the ranges given in the tables. Chen, Y., Bhaskaran, T., and Xue, D. (2008). Practical
Chen et al. (2008) propose the following rules for fractional tuning rule development for fractional order propor-
PI controllers: tional and integral controllers. Journal of Computa-
L tional and Nonlinear Dynamics, 3, 021403.
τ= (41) Chen, Y., Dou, H., Vinagre, B.M., and Monje, C.A.
 T
L + (2006). A robust tuning method for fractional order
 0.7 if τ < 0.1 pi controllers. In Fractional Differentiation and its
0.9 if 0.1 ≤ τ < 0.4

λ= (42) Applications. Porto.
 1.0 if 0.4 ≤ τ < 0.6
 Maione, G. and Lino, P. (2007). New tuning rules for
1.1 if 0.6 ≤ τ fractional PIα controllers. Nonlinear dynamics, 49, 251–
0.2978 257.
P = (43)
K(τ + 0.000307) Maiti, D., Acharya, A., Chakraborty, M., Konar, A., and
P (τ 2 − 3.402τ + 2.405) Janarthanan, R. (2008a). Tuning PID and PIλ Dµ
I= (44) controllers using the integral time absolute error crite-
0.8578T
rion. In Information and Automation for Sustainability.
These rules have been developed from the results of the
Colombo.
numerical method from (Chen et al., 2008) mentioned
Maiti, D., Biswas, S., and Konar, A. (2008b). Design of
above applied to plants given by (39). Using a similar
a fractional order PID controller using particle swarm
method, Bhambhani et al. (2008) reached the following
optimization technique. In 2nd National Conference on
rules:
L Recent Trends in Information Systems.
λ= − 0.04L + 1.2399 (45) Monje, C.A., Vinagre, B.M., Chen, Y.Q., Feliu, V.,
L+T Lanusse, P., and Sabatier, J. (2004). Proposals for
0.2T fractional PIλ Dµ tuning. In Fractional Differentiation
P = + 0.16 (46)
KL and its Applications. Bordeaux.
0.25K 0.19833 Monje, C.A., Vinagre, B.M., Feliu, V., and Chen, Y.
I= + + 0.09 (47)
TL L (2006). On auto-tuning of fractional order PIλ Dµ con-
 Parameters for the first set of rules for plants with an S-shaped step response
T T
Parameters to use when 0.1 ≤ K ≤5∧L≤2 Parameters to use when 5 ≤ K ≤ 50 ∧ L ≤ 2
P I λ D µ P I λ D µ
1 −0.0048 0.3254 1.5766 0.0662 0.8736 2.1187 −0.5201 1.0645 1.1421 1.2902
L 0.2664 0.2478 −0.2098 −0.2528 0.2746 −3.5207 2.6643 −0.3268 −1.3707 −0.5371
T
K
0.4982 0.1429 −0.1313 0.1081 0.1489 −0.1563 0.3453 −0.0229 0.0357 −0.0381
L2 0.0232 −0.1330 0.0713 0.0702 −0.1557 1.5827 −1.0944 0.2018 0.5552 0.2208
T 2


K
−0.0720 0.0258 0.0016 0.0328 −0.0250 0.0025 0.0002 0.0003 −0.0002 0.0007
TL
K
−0.0348 −0.0171 0.0114 0.2202 −0.0323 0.1824 −0.1054 0.0028 0.2630 −0.0014

Parameters for the second set of rules for plants with an S-shaped step response
T
Parameters to use when 0.1 ≤ K ≤ 50 ∧ L ≤ 0.5
P I λ D µ
1 −1.0574 0.6014 1.1851 0.8793 0.2778
L 24.5420 0.4025 −0.3464 −15.0846 −2.1522
T
K
0.3544 0.7921 −0.0492 −0.0771 0.0675
L2 −46.7325 −0.4508 1.7317 28.0388 2.4387
T 2


K
−0.0021 0.0018 0.0006 −0.0000 −0.0013
TL
K
−0.3106 −1.2050 0.0380 1.6711 0.0021

Parameters for the first set of rules for plants with a critical gain
Parameters to use when Pcr ≤ 8 ∧ Kcr Pcr ≤ 64 Parameters to use when Pcr ≤ 8 ∧ 64 ≤ Kcr Pcr ≤ 640
P I λ D µ P I λ D µ
1 0.4139 0.7067 1.3240 0.2293 0.8804 −1.4405 5.7800 0.4712 1.3190 0.5425
Kcr 0.0145 0.0101 −0.0081 0.0153 −0.0048 0.0000 0.0238 −0.0003 −0.0024 −0.0023
Pcr 0.1584 −0.0049 −0.0163 0.0936 0.0061 0.4795 0.2783 −0.0029 2.6251 −0.0281
1/Kcr −0.4384 −0.2951 0.1393 −0.5293 0.0749 32.2516 −56.2373 7.0519 −138.9333 5.0073
1/Pcr −0.0855 −0.1001 0.0791 −0.0440 0.0810 0.6893 −2.5917 0.1355 0.1941 0.2873

Parameters for the second set of rules for plants with a critical gain
Parameters to use when Pcr ≤ 2
P I λ D µ
1 −25.8467 10.5528 0.6213 15.7620 1.0101
Kcr −0.0119 0.2352 −0.0034 −0.1771 0.0024
Pcr 40.4266 −17.0426 0.2257 −23.0396 −0.8606
Pcr2 −14.5136 6.3144 0.1069 8.2724 0.1991
Kcr Pcr 0.0147 −0.0617 0.0008 0.1987 −0.0005
1/Kcr 1.6841 −0.9399 1.1809 −0.8892 −0.9300
1/Pcr 4.7503 −1.5547 0.0904 −2.9981 −0.1609
Kcr /Pcr 0.0144 −0.0687 0.0010 0.0389 −0.0009
Pcr /Kcr −7.0200 3.4357 −0.8139 2.8619 0.5846

Parameters for the third set of rules for plants with a critical gain
Parameters to use when 0.2 ≤ Pcr ≤ 5 and 1 ≤ Kcr ≤ 200
P I λ D µ
1 −1.6403 −92.5612 0.7381 −8.6771 0.6688
Kcr 0.0046 0.0071 −0.0004 −0.0636 0.0000
Pcr −1.6769 −33.0655 −0.1907 −1.0487 0.4765
Kcr Pcr 0.0002 −0.0020 0.0000 0.0529 −0.0002
1/Kcr 0.8615 −1.0680 −0.0167 −2.1166 0.3695
1/Pcr 2.9089 133.7959 0.0360 8.4563 −0.4083
Kcr /Pcr −0.0012 −0.0011 0.0000 0.0113 −0.0001
Pcr /Kcr −0.7635 −5.6721 0.0792 2.3350 0.0639
log10 (Kcr ) 0.4049 −0.9487 0.0164 −0.0002 0.1714
log10 (Pcr ) 12.6948 336.1220 0.4636 16.6034 −3.6738
Table 2. Tuning rule parameters from Valério and Sá da Costa (2006, 2007)

trollers. In Fractional Differentiation and its Applica- O. Agrawal (eds.), Fractional calculus: theoretical devel-
tions. Porto. opments and applications in Physics and Engineering,
Monje, C.A., Vinagre, B.M., Feliu, V., and Chen, Y. 463–476. Springer, Dordrecht.
(2008). Tuning and auto-tuning of fractional order con- Vinagre, B. (2001). Modelado y control de sis-
trollers for industry applications. Control Engineering temas dinámicos caracterizados por ecuaciones ı́ntegro-
Practice, 16(798–812). diferenciales de orden fraccional. Ph.D. thesis, Univer-
Valério, D. and Sá da Costa, J. (2006). Tuning of fractional sidad Nacional de Educación a Distancia, Madrid.
PID controllers with Ziegler-Nichols type rules. Signal Zhao, C., Xue, D., and Chen, Y. (2005). A fractional
Processing, 86(10), 2771–2784. order PID tuning algorithm for a class of fractional order
Valério, D. and Sá da Costa, J. (2007). Tuning rules for plants. In International Conference on Mechatronics &
fractional PIDs. In J.A.T. Machado, J. Sabatier, and Automation. Niagara Falls.