Professional Documents
Culture Documents
http://dx.doi.org/10.1007/s12555-018-0129-3 http://www.springer.com/12555
Abstract: Research on the fractional order system is becoming more and more popular. Most of the fractional order
controller design methods focus on single-input-single-output processes. In this paper, a fractional order internal
model controller with inverted decoupling is proposed to handle non-integer order two-input-two-output systems
with time delay. The fractional order two-input-two-output (FO-TITO) process is decoupled by inverted decoupling
method. The fractional order internal model control (IMC) is then used to simplify the tuning process. Because of
the complexity of multiple time delay, the condition of FO-TITO process with time delay is discussed. In order to
ensure the robustness of the system, a Maximum sensitivity function is used to tune the parameters. Then Lyapunov
stability theory is applied to verify the stability of the system. The proposed controller provides ideal performance
for both set point-tracking and disturbance rejection and is robust to process gain variations. Numerical results
show the performance of the proposed method.
Keywords: Fractional order system, internal model control, inverted decoupling, Lyapunov stability theory,
maximum sensitivity function, time delay, two-input-two-output process.
Manuscript received March 5, 2018; revised September 21, 2018; accepted November 12, 2018. Recommended by Associate Editor Niket
Kaisare under the direction of Editor Young IL Lee. This research has been supported in part by the National Natural Science Foundation of
China (61873022, 61573052), in part by the Beijing Natural Science Foundation (4182045) and in part by the Fundamental Research Funds
for the Central Universities under Grant XK1802-4.
Dazi Li, Xingyu He, Tianheng Song, and Qibing Jin are with the Department of Automation, College of Information Science and Tech-
nology, Beijing University of Chemical Technology, Beijing, 100029, China (e-mails: lidz@mail.buct.edu.cn, hexingyu1991@foxmail.com,
skyer_00@163.com, jinqb@mail.buct.edu.cn).
* Corresponding author.
⃝ICROS,
c KIEE and Springer 2019
Fractional Order IMC Controller Design for Two-input-two-output Fractional Order System 937
the zero of Gm (s) doesn’t lie in the right half section re- y(t) = Cx(t), (13)
gion of the complex plane. What’s more, G+ m (s) must have
a steady-state gain equal to one. And the steady-state gain where the fractional order γ ∈ (0, 1), A is the system ma-
of Gm (s) remains in G−m (s).
trix, B is the input matrix and C is the output matrix. By
Step 2: The controller CFOIMC (s)is specified as using the Laplace transformation to (9) and (10), with zero
initial condition, the state space of the system can be ob-
1 tained as
CFOIMC (s) = f (s), (6)
G−
m (s)
Sγ X(s) = AX(s) + BU(s), (14)
where f (s) is a low pass filter with a steady gain of one.
Typically, it has the form: Y (s) = CX(s). (15)
c2 G p22 = y2 , (22)
d12 (s)
Fig. 3. The block diagram of the inverted decoupling. (b121 sβ121 + b120 )(a112 sα112 + a111 sα111 + a110 )
=− ,
(b111 sβ111 + b110 )(a122 sα112 + a121 sα121 + a120 )
expression. Simple decoupling method will make the gen- d21 (s)
eralized matrix complex. Different from the ideal and sim- (b211 sβ211 + b210 )(a222 sα222 + a221 sα221 + a220 )
=− ,
ple decoupling methods, inverted decoupling has a simple (b221 sβ221 + b220 )(a212 sα112 + a211 sα211 + a210 )
structure of decoupler and generalized matrix. The con- (24)
troller design method is based on IMC which is a model
based method. A simple structure of generalized matrix where βi j and αi j are fractional order, a and b are the co-
can reduce difficulty of controller design. These are why efficients of the transfer function decoupler in (18). The
inverted decoupling is chosen. Inverted decoupling has decoupler in (24) can be the form of a filter with fractional
simple structure which is easy to implement. After in- order [28]. Obviously, the decoupler may not be realized
verted decoupling, the structure of the decoupler and the in some situation. Researches on the realizablation of de-
generalized transfer matrix are very simple. The block di- coupler for integer can be found in [29,30]. The condition
agram of the inverted decoupling is shown in Fig. 3. that can guarantee the realizablation of the inverted decou-
According to the theory of decoupling, the decoupler pling will be discussed later.
can be obtained:
3.2. Inverted decoupling with time delay
G p12 (s)
d12 (s) = − , Process with no time delay was discussed in the last
G p11 (s) sections. However, most of the processes in industry do
G p21 (s) have time delay.
d21 (s) = − . (19)
G p22 (s) Considering a FO-TITO process with time delay:
[ ]
Because of the special structure of the decoupler, the G p11 (s) G p12 (s)
G p (s) = , (25)
generalized open loop transfer matrix cannot be easily ob- G p21 (s) G p22 (s)
tained by the product of the decoupling matrix and the
transfer function matrix. Based on the structure shown in where each transfer function G pi j (s) has the form of
Fig. 3, the generalized open loop transfer matrix can be
bi j1 sβi j1 + bi j0
obtained. GPi j (s) = α
e−θi j s . (26)
From Fig. (3), the output of controller c1 , c2 and the ai j2 s i j2 + ai j1 sα i j1 + ai j0
input u1 , u2 of Gm (s) have the relationship as By applying the inverted decoupling mentioned in the last
G p12 (s) section, the decoupler can be obtained as
u1 = c1 − u2 ,
G p11 (s) d12 (s)
G p21 (s) (b121 sβ121 + b120 )(a112 sα112 + a111 sα111 + a110 )
u2 = c2 − u1 . (20) =−
G p22 (s)
(b111 sβ111 + b110 )(a122 sα112 + a121 sα121 + a120 )
The input u1 , u2 of Gm (s) and the output of the system y × e−(θ12 −θ11 )s ,
has the relationship as d21 (s)
u1 G p11 (s) + u2 G p12 (s) = y1 , (b211 sβ211 + b210 )(a222 sα222 + a221 sα221 + a220 )
=−
u2 G p22 (s) + u1 G p21 (s) = y2 . (21) (b221 sβ221 + b220 )(a212 sα112 + a211 sα211 + a210 )
× e−(θ21 −θ22 )s . (27)
Substituting (20) into (21), the result will be
It is worth noting that if the time-delayed item θi j has
c1 G p11 = y1 , the relationship such as θ12 < θ11 , θ21 < θ22 , the decoupler
940 Dazi Li, Xingyu He, Tianheng Song, and Qibing Jin
θii ≤ θik ≤ min(θi j ). (28) Usually, the fractional order αi j and βi j range from 0 to 1,
i̸= j
which means k = 1 satisfies (33).
- In order to make sure the decoupler element proper,
the relative degree ri j must be greater than zero which 3.5. Parameters tuning based on maximum sensitivity
means the order of the denominator must be bigger or A common sense in a controller design is that the pa-
equal than that of the numerator in (27), where ri j is rameters tuning is of great importance, especially when
the relative degree of G pi j (s), and rik is the relative de- the controller design method is model based. Because of
gree of row i of the transfer function G p (s). Relative the changeable of the operating conditions and times, ro-
degree ri j must satisfy the condition following bustness against the parameter fluctuations of the system
is important in process controls. To solve this problem,
rii ≤ ri j ≤ min(ri j ). (29) maximum sensitivity (Ms ) is used to tune the parameters
j̸=i
in the fractional controller design. The maximum sensi-
According to (28) and (29), to make the decoupler real- tivity is defined as
izable, the diagonal element G pii should have the smallest
of time delay and relative degree in its row. 1
Ms = max . (34)
ω ∈[0,+∞] 1 +C( jω )G p ( jω )
3.4. Controller design
It is a common sense that gain margin (Am ) and phase
Let us consider a FO-TITO process. Each element of
margin (φm ) show the robustness of the system. Fig. 4
the transfer function matrix has the same form as (18). In
shows the Nyquist curve of the open-loop transfer func-
order to design the controller of the system, the FO-TITO
tion. In (35), C( jω )G p ( jω ) is the open-loop transfer
process should be decoupled. The block diagram of the
function. The maximum sensitivity Ms means the inverse
system is shown in Fig. 3. After applying inverted de-
of the shortest distance from the Nyquist curve to the crit-
coupling to the FO-TITO process, the interaction between
ical point (−1, j0). In Fig. 4, the point A is the tangent
the two primary loops has been minimized, which means
point. If the value of Ms is given, according to (34), we
the controllers C11 (s) and C22 (s) can be designed sepa-
have
rately. The FOIMC method discussed in Section 2 is used
to achieve the controller with the following expressions: 1 − jθ
C ( jw) G p ( jw) = −1 + e . (35)
Ms
C11 (s) = G−1
m11− (s) f 1 (s),
C22 (s) = −1
Gm22− (s) f2 (s), (30) A is the tangent point of the Nyquist curve and we have
d (C ( jw) G p ( jw)) π
where f1 (s) and f2 (s) are the low pass filters. They have arg = −θ. (36)
the forms: dw 2
ẋ = Ax. (47)
AT P + PA = −Q, (48)
4. NUMERICAL EXAMPLES
4.1. Example 1
Considering a FO-TITO process without time delay as
mentioned in [27]:
[ 1.2 0.6
]
2s0.5 +1 3s 0.7 +1
G p (s) = 0.5 1.5 . (49)
s0.8 +1 3s0.6 +1
Fig. 7. Step response of the proposed method with 30% Fig. 9. Tracking and disturbance rejection performance of
parameter perturbation. the proposed method with white noisy.
4.2. Example 2
In this section, the process with time delay is chosen to
test the performance of the proposed method.
Consider a FO-TITO process with time delay as men-
Fig. 8. Step response of the TITO-FOPID method with tioned in [27]:
30% parameter perturbation. [ −0.2s −0.3s
]
1.2e 0.6e
G p (s) = 2s0.5 +1 3s0.7 +1 . (54)
0.5e−0.4s 1.5e−0.3s
s0.8 +1 3s0.6 +1
In Fig. 9, a Gaussian white noisy with zero mean, a vari-
ance of 0.1 and a frequency equal to 100 Hz is added to the Since θ12 > θ11 and θ21 > θ22 in this example, accord-
input signal. This figure shows that the proposed method ing to (27), the new decoupler d12 (s) and d21 (s) can be
performs well in noisy rejection. The output curve of the obtained as follows:
TITO-FOPID is much noisier than the proposed method. 2s0.5 + 1 −0.1s
As is mentioned in section 3, the second method of Lya- d12 (s) = − e ,
2(3s0.7 + 1)
punov is used to analyze the stability of the system. In (55)
3s0.6 + 1 −0.1s
Example 1, after decoupled, the stability of the two loops d21 (s) = − 0.8 e .
of the FO-TITO system can be verified separately. Both of 3(s + 1)
the two loops are stable, then conclusion that the system is As mentioned in Example 1, the controller is designed
stable can be reached. According to (50), the state space by the method as [7]. The design specification of loop 1
equation of loop 1 can be achieved as are φm1 = 54◦ and ωc1 = 1 rad/s, and the specifications of
1 loop 2 are φm2 = 70◦ and ωc2 = 1 rad/s. The TITO-FOPID
ẋ = − x + u, (52) controller obtained is
λ1
2.3
where λ is the parameter of the controller. A1 = −1/λ1 , a C11 (s) = 1.78(1 +
s0.93
),
positive matrix P1 = 1 is chosen, according to the second (56)
2.67
method of Lyapunov, Q1 = 2/λ1 can satisfy (47) which C22 (s) = 1 +
s0.78
.
944 Dazi Li, Xingyu He, Tianheng Song, and Qibing Jin
Fig. 10. Step response of the TITO-FOPID method and Fig. 11. tep response of the TITO-FOPID method with
the proposed method. 30% parameter perturbation with time delay.
2s0.5 + 1
C11 (s) = ,
1.2(1 + 0.6885s)
3s0.6 + 1
C22 (s) = . (57)
1.5(1 + 0.6512s) Fig. 12. Step response of the proposed method with 30%
parameter perturbation with time delay.
The closed-loop system step responses of TITO-FOPID
and the proposed method are shown in Fig. 10, when a unit
step input is set for the set-points at t = 0 s and t = 5 s. A proposed method is more robust to the time constant un-
step disturbances of the magnitude 0.1 are added into the certainty than the TITO-FOPID.
two inputs at t = 15 s. It can be seen from Fig. 10 that Fig. 15 is the manipulated variable input. In Fig. 16, a
the decoupling performance of the FO-TITO-IMC con- Gaussian white noisy with zero mean, which is the same
troller is better, because the overshoot of the proposed as Example 1 except the existence of time delay in the
method is smaller than the TITO-FOPID method, and the process, a variance of 0.1 and a frequency of 100Hz is
step responses of the proposed method is faster. The pro- added to the input signal. As is shown in Fig. 16, the
posed method reaches the steady-state from the distur- proposed method performs better in noisy rejection, since
bance faster than the TITO-FOPID method which means the output of the TITO-FOPID method is much noisy.
the proposed method perform better for disturbance rejec- Since the parameters of the controllers have been ob-
tion. tained, the stability of the system can be analyzed based
To test the robustness of the control system for process on the parameters. In Example. 2, According to (55), the
with time delay, a mismatch with a 30% parameter uncer- state space equation of loop 1 can be given as
tainty and a 50% time constants perturbation are added to [ ] [ ]
the system. Results of system gain perturbation are shown 0 1 0
ẋ = x+ u. (58)
in Fig. 11 and Fig. 12. Results of time constant pertur- 0 − (2 + 10λ1 ) 1
bation are shown in Fig. 13 and Fig. 14. As shown in So that, the state matrix A is shown as
Fig. 11 and Fig. 12, both of the methods provide a better
[ ]
robust performance with a 30% parameter perturbation. 0 1
A= . (59)
Comparing Fig. 13 and Fig. 14, it can be found that the 0 − (2 + 10λ1 )
Fractional Order IMC Controller Design for Two-input-two-output Fractional Order System 945
Fig. 13. Step response of TITO-FOPID method with 50% Fig. 15. Manipulated variable input graphs with white
time constant perturbation with time delay. noisy.
Fig. 14. Step response of the proposed method with 50% Fig. 16. Tracking and disturbance rejection performance
time constant perturbation with time delay. of the proposed method with white noisy.
According to the second method of Lyapunov, if there is The form of a positive matrix P is
a semi positive matrix Q which can satisfy (48) then the [ ]
system is stable. According to (48), we have a11 a12
P= . (64)
a21 a22
AT P + PA
[ ] According to the second method of Lyapunov, if there
0 a11 −(2+10λ1 )a12 is a semi positive matrix Q that can satisfy (48), then the
=
a11 −(2+10λ1 )a12 a12 +a21 −2∗(2+10λ1 )a22 system is stable. According to (48), we have
= −Q. (61)
AT P + PA
[ ]
It is obvious that there is a solution that can satisfy P is a 0 a11 − 3+10 λ2
= 1.5λ a12
positive matrix and Q is a semi positive matrix, this means λ2 λ2
a11 − 3+10
1.5λ2 a21 a12 + a21 − 2 ∗ 3+10
1.5λ2 a22
loop 1 is stable.
According to (55), the state space equation of Loop 2 = −Q. (65)
can be given as
There is a solution for (66) that can satisfy P is a pos-
[ ] [ ]
0 1 0 itive matrix and Q is a semi positive matrix. This means
ẋ = λ x + u. (62)
0 − 3+10
1.5λ
1 loop 2 is stable. Then the system is stable.
946 Dazi Li, Xingyu He, Tianheng Song, and Qibing Jin
[1] Y. Luo and Y. Q. Chen, Fractional Order Motion Controls, [17] D. Z. Li and X. Y. He, “The inverted decoupling based frac-
John Wiley & Sons, 2012. tional order two-input-two-output IMC controller,” Proc.
of the ASME International Design Engineering Technical
[2] Y. Q. Chen, I. Petras, and D. Xue, “Fractional order Conferences & Computers and Information in Engineering
control-A tutorial,” Proc. of the American Control Confer- Conference, Cleaveland, 2017.
ence (ACC), pp. 1397-1411, 2009.
[18] S. G. Samko, A. A. Kilbas, and O. I. Marichev, “Fractional
[3] D. Li, P. Ding, and Z. Gao, “Fractional active disturbance integrals and derivatives,” Theory and Applications, Gor-
rejection control,” ISA Transactions, vol. 62, pp. 109-119, don and Breach, Yverdon, 1993.
2016.
[19] M. Morari and E. Zafiriou, Robust Process Control, Pren-
[4] A. Oustaloup, La Commande CRONE: Commande Ro-
tice Hall, Englewood Cliffs, NJ, 1989.
buste d’Ordre Non Entier, Hermes, 1991.
[20] D. E. Rivera, M. Morari, and S. Skogestad, “Internal model
[5] I. Polubny, “Fractional-order systems and PIλ Dµ con-
control: PID controller design,” Industrial & Engineering
troller,” IEEE Transactions on Automatic Control, vol. 44,
Chemistry Process Design and Development, vol. 25, no.
pp. 208-214, 1999.
1, pp. 252-265, 1986.
[6] D. Valério, and J. S. da Costa, “Tuning of fractional PID
controllers with ZieglerNichols-type rules,” Signal Pro- [21] B. J. West, “Fractional calculus in bioengineering,” Journal
cessing, vol. 86, no. 10, pp. 2771-2784, 2006. of Statistical Physic, vol. 126, no. 6, pp. 1285-1286, 2007.
[7] C. A. Monje, B. M. Vinagre, and V. Feliu, “Tuning and [22] L. Sun, J. Dong, and D. Li, “A practical multivariable con-
auto-tuning of fractional order controllers for industry ap- trol approach based on inverted decoupling and decentral-
plications,” Control Engineering Practice, vol. 16, no. 7, ized active disturbance rejection control,” Industrial & En-
pp. 798-812, 2008. gineering Chemistry Research, vol. 55, no. 7, pp. 2008-
2019, 2016.
[8] M. Zamani, M. Karimi-Ghartemani, and N. Sadati, “De-
sign of a fractional order PID controller for an AVR using [23] Z. Li and Y. Q. Chen, “Ideal, simplified and inverted decou-
particle swarm optimization,” Control Engineering Prac- pling of fractional order TITO processes,” IFAC Proceed-
tice, vol. 17, no. 12, pp. 1380-1387, 2009. ings Volumes 2014, vol. 47, no. 3, pp. 2897-2902, 2014.
[9] Y. Luo and Y. Q. Chen, “Fractional order [proportional [24] H. Sheng, Y. Q Chen, and T. S. Qiu, Fractional Processes
derivative] controller for a class of fractional order sys- and Fractional-order Signal Processing: Techniques and
tems,” Automatica, vol. 45, no. 10, pp. 2446-2450, 2009. Applications, Springer Science & Business Media, 2011.
Fractional Order IMC Controller Design for Two-input-two-output Fractional Order System 947
[25] Y. Li, Y. Q. Chen, and Y. Cao, “Fractional order univer- Dazi Li received the Ph.D. degree in en-
sal adaptive stabilization,” Proc. of the 3rd IFAC Workshop gineering from the Department of Electri-
on Fractional Differentiation and Its Applications, Ankara, cal and Electronic Systems, Kyushu Uni-
Turkey. 2008. versity, Fukuoka, Japan, in 2004. She is
[26] L. Sun, J. Dong, and D. Li, “A practical multivariable con- currently a Full Professor and Chair of De-
trol approach based on inverted decoupling and decentral- partment of Automation, College of Infor-
ized active disturbance rejection control,” Industrial & En- mation Science and Technology, Beijing
gineering Chemistry Research, vol. 55, no. 7, pp. 2008- University of Chemical Technology, Bei-
2019, 2016. jing, China. Her current research interests
include machine learning and artificial intelligence, advanced
[27] Z. Li and Y. Q. Chen, “Ideal, simplified and inverted decou- process control, complex system modeling and optimization. Dr.
pling of fractional order TITO processes,” IFAC Proceed- Li is currently an Associate Editor of ISA Transactions.
ings Volumes 2014, vol. 47, no. 3, pp. 2897-2902, 2014.
[28] H. Sheng, Y. Q Chen, and T. S. Qiu, Fractional Processes
Xingyu He received the B.S. degree in au-
and Fractional-order Signal Processing: Techniques and
tomation from the Department of Automa-
Applications, Springer Science & Business Media, 2011.
tion, Beijing University of Chemical Tech-
[29] K. Vladimír, “Optimal decoupling controllers for singu- nology, Beijing, China, in 2011, where he
lar systems,” Proc. of the European Control Conference is currently pursuing the M.S. degree with
(ECC), Zürich, pp. 306-311, 2013. the College of Information Science and
[30] C. A. Lin, “Necessary and sufficient conditions for exis- Technology. His current research interests
tence of decoupling controllers,” IEEE Transactions on Au- include fractional order control.
tomatic Control, vol. 42, no. 8, pp. 1157-1161, 1997.
[31] Q. Jin, X. Du, and Q. Wang, “Analytical design 2 DOF
IMC control based on inverted decoupling for nonsquare Tianheng Song received the M.S. and
systems with time delay,” The Canadian Journal of Chem- Ph.D. degrees in control science and engi-
ical Engineering, vol. 94, no. 7, pp. 1354-1367, 2016. neering from the Department of Automa-
tion, Beijing University of Chemical Tech-
[32] M. A. Pakzad, S. Pakzad, and M. A. Nekoui, “Stability
nology, Beijing, China, in 2011 and 2016,
analysis of time-delayed linear fractional-order systems,”
respectively, where he is currently a post-
International Journal of Control, Automation and Systems,
doctoral fellow with the College of Infor-
vol. 11, no. 3, pp. 519-525, 2013.
mation Science and Technology. His cur-
[33] M. A. Pakzad and M. A. Nekoui, “Stability map of mul- rent research interests include reinforce-
tiple time delayed fractional order systems,” International ment learning.
Journal of Control, Automation and Systems, vol. 12, no.
1, pp. 37-43, 2014.
Qibing jin received the Ph.D. degree
[34] D. Valério, and J. S. da Costa, “A review of tuning meth- in control theory and engineering from
ods for fractional PIDs,” Proc. of the 4th IFAC Workshop the Northeastern University, Shenyang,
on Fractional Differentiation and Its Applications, vol. 10, Liaoning, China, in 1999. He is currently
2011. a Full Professor with the college of In-
[35] B. Sundara Vadivoo, Raja Ramachandran, Jinde Cao, Hai formation Science and Technology, the
Zhang, and Xiaodi Li, “Controllability analysis of nonlin- Director of the Institute of Automation
ear neutral-type fractional-order differential systems with with the Beijing University of Chemical
state delay and impulsive effects,” International Journal of Technology, Beijing, China. His main re-
Control, Automation and Systems, vol. 16, no. 2, pp. 659- search interests include advanced control, intelligent instrument,
669, 2018. Intelligent optimization algorithm, multivariable system identi-
fication and control theory. He has rich experience in control
[36] H. Zhang, M. Ye, J. Cao, and A. Alsaedi, “Synchronization
engineering, and his many research results have been applied in
control of Riemann-Liouville fractional competitive net-
petroleum and chemical industry.
work systems with time-varying delay and different time
scales,” International Journal of Control, Automation and
Systems, vol. 16, no. 3, pp. 1404-1414, 2018. Publisher’s Note Springer Nature remains neutral with regard
to jurisdictional claims in published maps and institutional affil-
iations.