Active Disturbance Rejection Controller for a

Separation Column

Eva H. Dulf, Roxana Both, Cristina I. Muresan

Department of Automation,
Technical University of Cluj-Napoca
Cluj-Napoca, Romania
[email protected]

Abstract— Isotope separation columns are known as complex, Fractional-order system and control have been studied by
nonlinear systems with large, variable time delays. The control of many researchers in the past decade because fractional calculus
such a process is a difficult task, in specific literature being can model real-world phenomena more precisely [7].
proposed complex control algorithms. The main goal of the
present paper is to design a simple control scheme based on For fractional-order model, fractional-order controller can
active disturbance rejection method. For accurate modeling of be naturally considered as the best controller. However, the
the plant a fractional-order model is used. The active active disturbance rejection control (ADRC), proposed as an
disturbance rejection control (ADRC) considers the gain alternative paradigm for control system design, offers a novel
uncertainty of the fractional-order plant model as a common perspective [8,9]. The original ADRC contains tracking
disturbance. Using a proper observer will reject it. Simulation differentiator, nonlinear PID, and extended state observer
results are provided to demonstrate the effectiveness of the (ESO) [10]. Bandwidth-parameterization method is proposed
proposed method. to improve ADRC for easy tuning [11].
Keywords—isotope separation column; fractional calculus; ADRC has been applied to DSP-based power converter
active disturbance rejection [12], delay system [13], motion control [14], hysteretic system
[15], high- performance turbofan engines [16], coordinated
I. INTRODUCTION robust nonlinear boiler– turbine–generator control system [17],
etc.
The major control challenge of the plant is the time delay
variations caused by the intrinsic behavior of the column. In Motivated by these applications, a novel approach named
addition, the input disturbances could destroy the product. fractional-order dynamics rejection scheme is proposed for the
13
Therefore a robust controller is essential to force the column to C cryogenic isotope separation column, based on active
work with imposed performances despite the presences of disturbance rejection method. Considering the fractional-order
system uncertainties and external disturbance. dynamics as a common disturbance and based on a special
observer, the ADRC method will handle both dynamic
For a proper controller design, an accurate and complete uncertainties of the system and external disturbances including
mathematical model of the plant is required. However, an sensor noise.
accurate mathematical model is impossible to obtain in the case
of isotope separation column, where nonlinearities, system The paper is organized as follows: after a brief introduction
uncertainties, and external disturbances are present. Moreover, in fractional calculus in section III is presented the cryogenic
the cryogenic conditions implied in the column operation and isotope separation plant, the proposed control solution is
the imposed condition for isotope exchange leads to only a few detailed in section IV, with the simulation results compared
intermediary experimental results - only inputs and outputs are with a classical PID control in section V. Concluding remarks
known – being difficult to validate the model. are presented in Section VI.
A major concern of researcher in the field of control is to
II. THE MATHEMATICAL BACKGROUND
develop a control technique without the need of an accurate
mathematical model of a system. Robust controller, presented There are several accepted definitions of the fractional
for example in [1], is designed taking into account small order differential or integrator. The Riemann–Liouville’s
uncertainties. The second main approach in controller design definition treats the notion of fractional order integral of order
for uncertain systems are based on disturbance estimators such ℜ(α) > 0 as a natural consequence of Cauchy’s formula for
as unknown input observer (UIO) [2], disturbance observer repeated integrals, expressed as [18]:
(DOB) [3, 4], perturbation observer (POB) [5], and extended
t
state observer (ESO) [6]. 1
∫ (t − τ)
n −1
D c− n f ( t ) = f (τ)dτ , t > c, n ∈ Z + (1)
(n − 1)! c

978-1-4799-3732-5/14/$31.00 ©2014 IEEE

 This equation can be used introducing the Euler’s Gamma concentration (Nf), Nf=N0=1.11%at [20]. The isotope of
function which is a generalization of a factorial: interest (13C), accumulated in the liquid carbon-monoxide, is
withdrawn as final product at the flow rate (P) and
t
1 concentration (Np). The gaseous carbon monoxide with lower
∫ (t − τ)
α −1
Dc− αf ( t ) = f (τ)dτ , t > c, α ∈ R + (2) (13C) concentration is evacuated as “waste” at the top side, by
Γ(α )
c flow rate (Qout) and the concentration (Nw). This “waste” is in
where fact the secondary product of the separation column. This
carbon monoxide with high-purity in (12C) is used in
∞ electronics technology to produce synthetic diamond crystals
∫
Γ(n ) = t n −1 e − t dt
0
(3) with 50% better thermal conductivity than that of the “usual”
synthetic crystals [20]. The column needs to be thermal
insulated by an external vacuum jacket. If cryogenic distillation
and n ∈ R + is the fractional order integral. temperature is used, the vapor pressure of carbon monoxide
with (12C) is greater than the pressure of carbon monoxide with
The definition for the fractional-order integral used for (13C), so that the isotope (13C) accumulates in the liquid phase.
dynamic systems, where it is usual that f(t) is a causal function This process is called the “rectifying process”. Simultaneously
of t, is [18]: decrease the concentration of (13C) in gaseous phase, in the
t “stripping process”. In a static steady-state liquid-gas contact,
1
∫ (t − τ) the accumulation of the isotope of interest is very small, the
α −1
D−αf ( t ) = f (τ)dτ , t > 0, α ∈ R + (4)
Γ(α ) concentration rising with the elementary separation ratio (α),
0
very close to the unity: α≈1.011 [20]. The (13C) isotope
The definition (4) cannot be used for the fractional-order concentration can be increased up to a desired value, if a
derivative by direct substitution of α by –α. The following permanent counter-current of the liquid-gaseous phases is
criteria must be met: a) guarantee the convergence of the ensured. In order to achieve the permanent counter-current the
integrals involved in the definition, and b) preserve the column is equipped by an electrically heated boiler (B) at the
properties of the ordinary derivative of integer-order. The bottom and a condenser (K), cooled with liquid nitrogen, at the
Riemann–Liouville definition for the fractional-order top of the column. For a proper operation of the column the
derivative of order α ∈ R+ has the following form [18]: liquid nitrogen level (hc) in condenser and the liquid carbon-
monoxide level in boiler (hb) must be maintained at constant
dm ⎡ ⎤
t
1 f (τ) values.
∫
α
R D f (t) = ⎢ d τ ⎥ , m −1 < α < m ,
dt m ⎢⎣ Γ(m − α ) 0 (t − τ)α − m +1 ⎥⎦
m∈N (5)
An alternative definition for the fractional-order derivative
was introduced by Caputo as [18]:
t
1 f ( m ) (τ)
∫ (t − τ)
α
C D f (t) = dτ , m − 1 < α < m , m ∈ N
Γ(m − α ) α − m +1
0
(6)
Another definition of the fractional-order derivative is the
Grünwald–Letnikov’s definition. It is based on the backward
difference generalization and is of high importance in
applications [19]:
m
f ( k ) (0 + ) t k − α
GL D
α
f (t) = ∑
k =0
Γ(m + 1 − α )
+

m−α , m > α −1 (7)

t
1
∫ (t − τ) (τ)dτ
( m +1
+ f
Γ(m + 1 − α ) Fig. 1. The cryogenic isotope separation column simplified scheme
0

An excessive boiling of the liquid carbon monoxide may

III. PROCESS DESCRIPTION lead to the flooding of the column, compromising the entire
The (13C) isotope separation column of the National separation process. After several experimental trials and
Institute for Research and Development of Isotopic and analysis of experimental data it was considered that the main
Molecular Technologies (NIRDIMT) from Cluj-Napoca is a manipulated variable chosen in orders to maintain the imposed
steel packed pipe, Fig. 1, feed with pure gaseous carbon- mole fraction of 13CO is the electrical power supplied to the
monoxide by the flow-rate (Qin) and (13C) isotope
boiler. By experimental identification methods this transfer Introducing the proportional (KP) and derivative effect
function is obtained as a fractional order one with the form: (KD), as in [21], it can be obtained the estimation-based state
feedback controller, with:
k
G (s ) = , α, β ∈ ℜ (8)
as α + bs β + 1 u 0 (t ) − f̂ (t )
u (t ) = ; u 0 = K P [r (t ) − ŷ(t )] − K D ŷ (t ) (15)
K0
IV. DESIGN OF THE ACTIVE DISTURBANCE REJECTION
The controller’s gain can be tuned to obtain the desired closed
CONTROLLER
loop performances.
In [21] is detailed the ADRC design method for a second
order system. The present paper extends this method to a
V. SIMULATION RESULTS
generalized second order system, a system described by a
transfer function with denominator presenting three terms, the Applying the above described method to the isotope
first two terms having non-integer order, as in (8). separation column having the fractional order mathematical
model
The mathematical expression that relates system input u(t),
output y(t) and an external disturbance d(t) can be generalized 0.58
G (s ) = (15)
as: 12s 2.4
+ 7s1.5 + 1
(α) ⎛ b (β ) 1 1 k ⎞ and imposing the observer eigenvalues l1=22.5, l2=168.75,
y (t ) = ⎜⎜ − y (t ) − y(t ) + d(t ) + ⋅ u (t )⎟⎟ (9) l3=421.875, results the controller parameters - computed
⎝ a a a a ⎠ according to a desired settling time - KP=0.5625, KD=1.5.
Abbreviating K = k and adding a gain uncertainty (ΔK) to Given the explicit feature of ADRC to cope with modeling
a
errors, the first goal of the paper is to provide visual insights
the nominal value (K0), K = K 0 + ΔK results: into the control loop behavior under variations of process
parameters. A series of simulations were run with fixed ADRC
(α) ⎛ b (β ) 1 1 ⎞
y (t ) = ⎜⎜ − y (t ) − y(t ) + d (t ) + ΔK ⋅ u (t )⎟⎟ + K 0 u (t ) (10) parameters as given above and the plant with varying both gain
⎝ a a a ⎠ and fractional order.

All terms being considered into a generalized disturbance, In Fig.3, the closed loop step responses of the process are
f(t), all that remains of the process model is a double integrator: presented considering +/- 50 variations of the process gain. The
initial value of the output is 1.1, the natural abundance of the
13
C isotope.
⎛ (α ) b (β ) 1 1 ⎞
y(t ) = ⎜⎜ − y (t ) − y(t ) − y (t ) − y(t ) + d(t ) + ΔK ⋅ u (t )⎟⎟ + K 0u (t ) 2.4

⎝ a a a ⎠
(11) 2.2

The generalized disturbance f(t) became: 2

mole fraction of (13CO)

(α)
b (β ) 1 1
f (t ) = − y (t ) − y(t ) − y (t ) − y(t ) + d (t ) + ΔK ⋅ u (t ) , (12)
1.8

a a a
1.6 K nominal
and contains the fractional order dynamics, the external 50%K

disturbance and the unknown internal dynamics, the process 1.4

150%K

model being only the double integrator.

1.2
The state equation form of the generalized disturbance:
1
D q x̂ f = Ax̂ f + Bu + L(f − f̂ ) 0 5 10
Time [min]
15 20 25

, (13)
f̂ = Cx̂ f Fig.3. (13CO)Mole fraction evolution considering +/-50% variations of K

with (L) the proportional observer gain and q = [α β] the Fig.4. presents the evolution of the control variable, the
electrical power supplied to the boiler of the column for the
fractional order. The output observer is: same process gain variations.
 = AX + Bu + Ef
X The simulation results considering the variation of the
(14)
ŷ = CX fractional order β of the (13CO) mole fraction and the electrical
power supplied to the boiler evolutions are presented in Fig.5
Using the bandwidth-parameterization method [21], all and Fig.6.
observer poles can be placed at ω0, which is the observer
bandwidth.
 In theory, ideal behavior (i.e., almost complete ignorance of
parameter variations) can be obtained by placing the observer
14

12
poles far enough to the left of the closed loop poles. To
demonstrate this statement, in Fig. 7 and Fig.8 are presented
the closed loop simulation results with different observer poles
Boiler electric power supply [W]

10 K nominal
50%K locations.
150%K
8

2.4
6

2.2
4

2 2

mole fraction of (13CO)

0 1.8 ESO CL
s =10*s
0 5 10 15 20 25
ESO CL
Time [min] s =100*s
ESO CL
s =5*s
Fig.4 Boiler electrical power supply evolution considering +/-50% variations 1.6

of K
1.4

2.4
1.2

2.2
1
0 5 10 15 20 25
2 Time [min]
mole fraction of (13CO)

1.8
Fig.7 (13CO)Mole fraction evolution for different observer pole locations
Betanominal
50%Beta
1.6 150%beta 12
ESO CL
s =10*s
11 ESO CL
s =100*s
1.4 ESO CL
s =5*s
10
Boiler electric power supply [W]

1.2 9

8
1
0 5 10 15 20 25
Time [min] 7

6
Fig.5 (13CO)Mole fraction evolution considering +/-50% variations of β
5

12 4

3
Betanominal
10
50%Beta 2
150%beta 0 5 10 15 20 25
Time [min]
Boiler electric power supply [W]

Fig.8 Boiler electrical power supply evolution for different observer pole
locations
6

Given the ubiquity of PID-type controllers in industrial

4
practice, the authors was concerned about advantages of
ADRC against classical approach. For this, a PI controller was
designed for quite similar performances than ADRC and was
compared the ADRC results to the PI controller. The
2

simulation results are presented in Fig.9 and Fig .10.

0
0 5 10
Time [min]
15 20 25
These results clearly demonstrate the ability of the ADRC
approach to keep the closed loop dynamics similar, even
Fig.6 Boiler electrical power supply evolution considering +/-50% variations including actuator saturation. The limitation of electrical power
of β supplied to the boiler is given by the construction of the plant
and are between 0 and 8 W.
The results prove the advantage of the method. One can see
that the closed loop step response remains similar or nearly
identical to the desired behavior. Similar results are obtained
varying the other fractional order α or other parameters of the
transfer function.
 much faster by ADRC compared to the PI controller. The price
2.4
ADRC is a high value of the control signal, Fig.12.
conventional control
2.2
50

2
mole fraction of (13 CO)

40
1.8

Boiler electric power supply [W]

1.6 30 ADRC
conventional control

1.4
20

1.2
10

1
0 5 10 15 20 25
Time [min]
0

Fig.9 (13CO)Mole fraction evolution for different observer pole locations

-10
0 10 20 30 40 50 60
8
Time [min]
ADRC
7 conventional control
Fig.12 Effect of disturbance rejection of ADRC in comparison with
conventional PI control - boiler electrical power supply evolution
Boiler elec tric power supply [W]

2.4
ADRC
5
conventional control
2.2
4

2
3
mole fraction of (13CO)

1.8
2

1.6
1
0 5 10 15 20 25
Time [min]
1.4

Fig.10 Boiler electrical power supply evolution for different observer pole
locations 1.2

As a final experiment for the ADRC, the authors examined 1

the disturbance rejection abilities by adding an input 0 10 20 30
Time [min]
40 50 60

disturbance into the process for both ADRC and the PI

controller. The input disturbance is effective at t=40 min, Fig.13 Effect of disturbance rejection with actuator saturation- (13CO)Mole
fraction evolution
Fig.11.
Including actuator saturation, Fig.13, the ADRC still
2.4 performs better than the conventional controller. In order to
obtain similar results with the classical control, one would need
2.2
to tune the controller for more degree of freedom, which
implies complex design algorithms.
2
mole fraction of (13CO)

1.8 VI. CONCLUSIONS

1.6
In this paper a novel approach based on ADRC has been
ADRC
conventional control
successfully applied on the 13C cryogenic distillation column,
1.4 where fractional-order dynamics are treated as a common
disturbance and actively rejected. Also the external
1.2
disturbance, sensor noise, and parameter disturbance are
estimated and rejected. The simulation results compared with
1
0 10 20 30
Time [min]
40 50 60 the results using classical control scheme proves the
advantages of the method. Using simple tuning parameters,
Fig.11 Effect of disturbance rejection of ADRC in comparison with ADRC can control easily fractional-order systems.
conventional PI control - (13CO)Mole fraction evolution
Future works will focus on adding dynamical uncertainty to
While both controllers were tuned for the same reaction on each parameter from the system.
setpoint changes, the impact of the disturbance is compensated
 ACKNOWLEDGMENT [10] J. Han, “From PID to active disturbance rejection control”, IEEE
Transactions on Industrial Electronics 2009;56(3):900–6.
This work was supported by a grant of the Romanian [11] Z. Gao, “Scaling and bandwidth-parameterization based controller
National Authority for Scientific Research, CNDI– UEFISCDI, tuning”. In: Proceedings of the 2003 American control conference;
project number 155/2012 PN-II-PT-PCCA-2011-3.2-0591. 2003. 4989–96.
[12] Sun B,Gao ZA., “DSP-based active disturbance rejection control design
for a 1-kW H-bridge DC–DC power converter, IEEE Transactions on
Industrial Electronics 2005; 52(5), pp.1271–1277.
REFERENCES [13] Y. Xia, P. Shi, G.P. Liu, D. Rees, J. Han, “Active disturbance rejection
[1] D.-W. Gu, P. Hr. Petkov and M. M. Konstantinov, Robust Control control for uncertain multivariable systems with time-delay”, IET
Design with MATLAB®, Springer-Verlag, 2013 Control Theory Application 2007; 1(1), pp.75–81.
[2] S. Hui, S. H. Zak, “Observer design for systems with unknown inputs”, [14] Q. Zheng, L.Q. Gao, Z. Gao, “On validation of extended state observer
International Jornal of Applied Math and Computer Science, vol. 15, no. through analysis and experimentation”, Journal of Dynamic Systems,
4, 2005, pp. 431-446 Measurement and Control 2012; 134(2), pp. 024505 (6pages).
[3] A. Radke and Z. Gao, “A survey of state and disturbance observers for [15] F.J. Goforth, Q. Zheng, Z. Gao, “A novel practical control approach for
practitioners,” in Proc. of the 2006 American Control Conference, rate independent hysteretic systems”, ISA Transactions 2012;51(3),
Minneapolis, Minnesota, USA, Jun 14-16, 2006 pp. 5183-5188. pp.477–484.
[4] J. Yang, W. H. Chen, S. Li, “Non-linear disturbance observer-based [16] R. Miklosovic, Z. Gao Z, “A dynamic decoupling method for controlling
robust control for systems with mismatched disturbances/uncertainties”, high performance turbofan engines”, In: Proceeding of the 16th IFAC
IET Control Theory and Applications, vol. 5, no. 18, 2011, pp. 2053- world congress; 2005. 16 (1).
2062. [17] T. Yu, K.W. Chan, J.P. Tong, B. Zhou, D.H. Li, “Coordinated robust
[5] S. Kwon and W. K. Chung, “A discrete-time design and analysis of nonlinear boiler– turbine–generator control systems via approximate
perturbation observer for motion control applications,” IEEE dynamic feedback linearization”, Journal of Process Control 2010;
Transactions on Control Systems Technology, 2003, vol. 11, no. 3, pp. 20(4), pp.365–374.
399-407. [18] C. A. Monje, Y.Q. Chen, B. M. Vinagre, D. Xue, V. Feliu, “Fractional-
[6] Zheng Zhu, Dong Xu, Jingmeng Liu, and Yuanqing Xia, “Missile order Systems and Controls”, DOI 10.1007/978-1-84996-335-0,
Guidance Law Based on Extended State Observer”, IEEE Transactions Springer Verlag, 2010
on Industrial Electronics, 2013, Vol. 60, No. 12, pp. 5882-5891. [19] I. Podlubny, “Fractional Differential Equations, Mathematics in Science
[7] R. Hilfer, “Application of fractional calculus in physics”, World and Engineering”, volume 198. San Diego: Academic Press, 1999
Scientific, Singapore, 2000. [20] D. Axente, M. Abrudean, A. Baldea, “Isotope Separation of 15N, 18O, 10B,
13
[8] Z. Gao, “Active disturbance rejection control: a paradigm shift in C by isotopic exchange”, (in Romanian), Casa Cartii de Stiinta, Cluj-
feedback control system design”. In: Proceedings of the American Napoca, 1994
control conference; 2006. p. 2399–405. [21] G. Herbst, “A Simulative Study on Active Disturbance Rejection
[9] Z. Gao, Y. Huang, J. Han, “An alternative paradigm for control system Control (ADRC) as a Control Tool for Practitioners”, Electronics 2013,
design”, In: Proceedings of IEEE conference on decision and control; 2, doi:10.3390/electronics2030246, pp. 246-279.
2001. p. 4578–85.