ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Research article

Modified reduced order observer based linear active disturbance rejection

control for TITO systems
S.N. Pawar, R.H. Chile, B.M. Patre n
Department of Instrumentation Engineering, S. G. G. S. Institute of Engineering and Technology, Nanded 431606, India

art ic l e i nf o a b s t r a c t

Article history: This paper proposes an observer based control approach for two input and two output (TITO) plant
Received 24 August 2016 affected by the lumped disturbance which includes the undesirable effect of cross couplings, parametric
Received in revised form uncertainties, and external disturbances. A modified reduced order extended state observer (ESO) based
9 July 2017
active disturbance rejection control (ADRC) is designed to estimate the lumped disturbance actively as an
Accepted 28 July 2017
extended state and compensate its effect by adding it to the control. The decoupled mechanism has been
used to determine the controller parameters, while the proposed control technique is applied to the TITO
Keywords: coupled plant without using decoupler to show its efficacy. Simulation results show that the proposed
Active disturbance rejection control (ADRC) design is efficiently able to nullify the interactions within the loops in the multivariable process with
Extended state observer (ESO)
better transient performance as compared to the existing proportional-integral-derivative (PID) control
Two input two output (TITO) systems
methods. An experimental application of two tanks multivariable level control system is investigated to
Lumped disturbance
Real time experimentation present the validity of proposed scheme.
& 2017 ISA. Published by Elsevier Ltd. All rights reserved.

1. Introduction as well as satisfactory performances. With these benefits, it was

Many industrial processes are complex and intrinsically mul-
tivariable in nature, and the problems like interactions, couplings,
unmodeled dynamics, parameter uncertainties are always en-
countered in multi-input multi-output (MIMO) processes. It de-
mands multivariable control design with better transient perfor-
mances in presence of undesirable effects. Two-input two-output
(TITO) system is a general class of multivariable process. The
presence of the loop interactions, dynamic coupling, etc. are re-
sponsible for poor control performance and the design of multi-
loop control is more challenging than single loop control. The ef-
fects of decoupling and stability of decoupled systems for TITO
processes are discussed in [1]. Many researchers like Vinante and
Luyben [2], Wood and Berry [3] and Ogunnaike and Ray [4], have
examined a large number of particular multi-loop control systems
which comprise of single input single output (SISO) controllers
acting on a multi-loop design. The extension of SISO controllers in
multiple single loops forms a multivariable control design. Due to
the superiority, structural simplicity, ease of implementation and
easy to tune, proportional-integral-derivative (PID) controllers are
widely used in the majority of control systems as a standard
feedback controller. Also, PID can provide better cost/benefit ratio
n
Corresponding author.
E-mail addresses: sushantnpawar@gmail.com (S.N. Pawar),
rhchile@yahoo.com (R.H. Chile), bmpatre@ieee.org (B.M. Patre).
widely used in industry during the period of 1920–1990s [5], and
around 90% of the practical industrial processes are stabilized and
controlled using SISO-PID-Controllers [6]. In multivariable case,
decentralized PID controller is extensively used due to its simple
structure, and tuning simplification [7–10]. The PID controllers
have some limitations such as error computation, complications
arise due to integral control, and noise degradation in the deri-
vative control, etc. [11]. Therefore, there is a requirement of further
developments in PID framework. The technique active disturbance
rejection control (ADRC), was investigated to overcome the iden-
tified problems associated with PID [11].
Most of the existing passive anti-disturbance control techni-
ques are constructed from the mathematical description of plant
dynamics. Many physical plants in the real world are nonlinear
and time-varying as well as highly uncertain, hence, mathematical
information of physical plants are not obtained accurately in an
industrial context. Thus, an active anti-disturbance control is in-
troduced with linear error feedback controller structure with
compensation components which can automatically detect system
model and the role of external disturbance in real time [11]. The
introduced controller architecture accounts for total uncertainties
including the internal uncertainties and external disturbances.
Such control technique called as an ADRC [11], offers a robust
controller building block where the system model is extended as
new state variable with unknown dynamics and disturbances.
The concept of ADRC and its applications were introduced by
Prof. Jingqing Han first time in Chinese literature in 1998 [12].

http://dx.doi.org/10.1016/j.isatra.2017.07.026
0019-0578/& 2017 ISA. Published by Elsevier Ltd. All rights reserved.

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
2 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

ADRC strategy is analyzed in the time domain as well as frequency as pre-specified gain-phase margin boundaries. The D-partition
domain for different types of systems have been reported in the technique [44] is used to obtain the stable region for controller
literature. The transfer function model of ADRC has been de- parameters and the Kharitonov theorem is utilized to determine
termined, and frequency domain analysis is carried out to evaluate the robust stability boundary. The performance of proposed
the performance and stability characteristics of systems [13]. The method under parametric uncertainty as well as external dis-
frequency domain analysis of non-linear ADRC using describing turbance and stochastic disturbance has been carried out in a si-
function method is employed for LTI plant [14]. Li et al. [15], uses mulation environment and is experimentally validated for two
the root locus analysis, describing function and extended circle
tanks multivariable level control setup.
criterion to approach frequency domain stability analysis of servo
The remainder of this paper is organized as follows. The pro-
system. In [16], it is shown that time domain analysis of linear
blem and objective is stated in Section 2. Section 3 presents the
ADRC can be used for a broad class of physical processes, when the
linear ADRC and reduced order linear ADRC. D-partition and
model is known and unknown, which is estimated using ESO. A
identification of stability region with gain-phase margin approach
time domain approach of ADRC for non-linear, time-varying plant
by singular perturbation concept has been explained in [17]. An is introduced in Section 4. In Section 5, application of modified
ADRC has particular component as ESO, which plays a vital role in reduced order linear ADRC to TITO process is explained by con-
estimating external disturbance as well as internal dynamics, and sidering a case study of Wood and Berry distillation column, fol-
it is also validated that such uncertainty can be reduced using lowed by its simulation results in Section 6. Modeling of an ex-
ADRC [18]. A review of the methods which improves the overall perimental setup of two-input two-output plant is given in Section
efficiency of extended state observers has been given in [19]. Many 7 succeeded by its experimental results in Section 8. Finally con-
ADRC applications have been noted in the literature. Some prac- cluding remarks are summarized in Section 9.
tical industry-wide applications of ADRC has been introduced in
[20]. The independent model ADRC gives better performance than
PID control for motion control design application [21]. ADRC has
been efficiently implemented on motion control problem [22–25]. 2. Problem statement
As ADRC is the technique that does not depend on the mathe-
matical model of the plant, which makes it a better solution to Consider a two-input two-output system as shown in Fig. 1. The
industry application like MEMS Gyroscope [26–30]. Back-stepping input-output relationships are given by
active disturbance rejection control with reduced order ESO is y1(s ) = G11(s )u1(s ) + G12(s )u2(s ) (1)
designed for integrated missile guidance in [31]. The concept of
ADRC has been implemented successfully in many applications,
e.g. web tension and velocity regulation [32,33], twin rotor multi-
y2 (s ) = G21(s )u1(s ) + G22(s )u2(s ) (2)
input multi-output system [34], chemical-process control [35,36],
boiler-turbine unit [37], Alstom gasifier [38]. where, G11(s ), G12(s ), G21(s ) and G22(s ) are the four transfer functions
The key component of ADRC, extended state observer (ESO) relating to the two-outputs corresponding to the two-inputs. As
which can track the extended state of a class of uncertain plant stated in Eqs. (1)–(2), the change in control signals due to coupling
satisfactorily by choosing proper functions and related parameters phenomenon will affect the controlled outputs y1 and y2
of the observer [39]. ADRC has the strong quality as the error respectively.
driven rather than model based control law. Also, it offers a ESO It is proposed to consider the two-input two-output system
which estimates extended state as a lumped disturbance includes
affected by the lumped disturbance which includes the effect of
parametric uncertainties, external disturbances, and cross cou-
cross couplings within the loops, parametric uncertainties, and
pling effect in multivariable processes. These promising features of
external disturbances. Thus, the objective is to design the control
ESO makes ADRC, a better solution to multivariable systems. A
input u1 and u2 such that the plant outputs y1 and y2 will follow
transformation based nonlinear control approach for uncertain
multivariable systems with time delay has been presented in [40]. desired response in presence of lumped disturbance.
The issue of rejecting the disturbance in industrial processes with In order to explain the basic idea of ESO based ADRC, primarily
transport delay was focused, and the addressed problem was first order plus time delay (FOPTD) system is considered. Fur-
overcome by designing ADRC combined with smith predictor thermore the reduced order ESO based linear ADRC for FOPTD
strategy [41]. Also, a modified active disturbance rejection control plant has discussed. The results are further extended to TITO plant
design has been offered for time delay systems [42]. A state space in Section 5.
based nonlinear control approach has presented in [40] for TITO
system, which is demonstrated in simulation environment with-
out a proper method for tuning of nonlinear feedback parameters
as well as for observer gain. In this paper, a linear ADRC metho-
dology is extended to solve the problem of tuning of controller
parameters for TITO systems. The proposed control technique is
based on decoupled systems obtained from initially coupled plant.
The tuned control law is especially applied to originally coupled
system directly, and yet it handles the interactions efficiently. In
the proposed technique, the control gains are selected from the
boundary region of the controller where boundary region is ob-
tained by considering reduced order model of decoupled systems.
The closed loop stability in presence of time delay of the plant
have been investigated. The D-partition boundaries for closed loop
system are plotted and stability regions are identified to make
proper selection of controller parameters. A gain-phase margin Fig. 1. Block diagram of process with two controlled outputs and two manipulated
approach [43] is employed to find the stability boundaries as well inputs.

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

3. Linear active disturbance rejection control
This section presents a brief overview of linear case of ADRC for
first order plant which is further extended to multivariable TITO
plant. Although ADRC was presented for a relatively straightfor-
ward first order plant [45], the same idea can be applied to FOPTD
plant using modified methodology.
K
considering b0 ≈ b ≈ T . Modeling errors and varying process para-
meters represented by Δb, are handled as an internal uncertainty.
For the sake of simplicity, let us consider f ̇ (t ) = h(t ) as an ex-
tended state for the design of ADRC. Under this, the Eq. (11) can be
transformed into state space form as,
⎧ ẋ (t ) = x (t ) + b u(t )
⎪

⎨ 1 2 0
The nominal model of FOPTD plant is, ⎩ x2̇ (t ) = h(t )
⎪
(12)
y(s ) K
G(s ) = = e−τs where, x1(t ) = y(t ), and x2(t ) = f (t ).
u(s ) Ts + 1 (3) A luenberger observer is employed for designing of linear
where y, u, K, T and τ are denoted as output, control input, process ADRC. To estimate extended state x2(t ), an ESO is designed as,
gain, time lag and time delay respectively. ⎧ e1 = z1 − y
The plant with time delay faces challenges to control design ⎪
⎨ z1̇ (t ) = z2(t ) − l1g1(e1) + b0u(t )
methods, hence many a times transfer function approximation can ⎪
be made in the term e−τs by using Pade approximation or any other ⎩ z2̇ (t ) = − l2g2(e1) (13)
approximation tool [11]. However, from the literature [11], ADRC
where, z1, z2 are the estimated states of x1 and x2, respectively and
does not require an accurate mathematical information of the
plant. Therefore, a simple approximation of e−τs ≈ 1/(τs + 1) may l1 and l2 are observer gain parameters.
prove to be good enough [11]. Thus Eq. (3) is approximated as, Since the virtual input f ̇ (t ) is immeasurable, yet an estimated
state z2 provides an approximate value of f(t). Here, it is taken as
y(s ) K 1 gi(e1) = e1, i ¼ 1,2, referred as linear extended state observer. The
G(s ) = =
u(s ) (Ts + 1) (τs + 1) (4) controller is chosen as,
Considering an added pole due to delay approximation is not a u(t ) = (v(t ) − f (t )/b0) (14)
dominant pole as compared to time constant (T) of the process, the
Eq. (4) can be simplified as, where, v(t) is the output of the proportional controller and given as,

y(s ) K v(t ) = Kp( r (t ) − y(t )) (15)

G(s ) = =
u(s ) (Ts + 1) (5)
where r is the reference signal.
The Eqs. (13)–(15) gives the structure of linear ADRC as shown
Remark 1. The Eq. (5) is only used to design second order ob- in Fig. 2.
server based ADRC and further for reduced order observer based It can be noted that the observer gain parameters l1 and l2 are
ADRC. It is to be noted that, the neglected time delay has been chosen such way that the observer dynamics will be fast enough.
considered in stability analysis to determine the controller The chosen observer gains must follow the Hurwitz criteria for the
parameters. characteristic polynomial s2 + l1s + l2 . For simplification of tuning,
all the observer poles are placed at -ω0 which results in the
3.1. Second order ADRC characteristic polynomial of Eq. (13) to be,

The plant in Eq. (5) can be represented in time domain as, λ(s ) = s 2 + l1s + l2 = (s + ω0)2 (16)
Tẏ (t ) + y(t ) = Ku(t ) (6) where, ω0 is the observer bandwidth.
From Eq. (16), the observer gain parameters are chosen as,
By considering the effect of input disturbance, d(t) on the plant
dynamics Eq. (6) becomes, ⎡⎣ l1 l2⎤⎦ = ⎡⎣ 2ω0 ω02⎤⎦
Tẏ (t ) + y(t ) = Ku(t ) + d(t ) (7)

Now, taking b =
K
and rearranging Eq. (7), leads to 3.2. Design of reduced order ESO based ADRC
T
1 1 The ADRC as discussed above is taking into account of second
ẏ (t ) = bu(t ) − y(t ) + d(t )
T T (8) order ESO to get an extended state estimation for compensation;
By putting b = b0 + Δb in Eq. (8), gives however, there is an excess when the first part of the state may be
measured correctly. A reduced order ESO is utilized so as to minimize
⎛ 1 1 ⎞ complications due to higher order plant and perform in-depth step
ẏ (t ) = ⎜ Δbu(t ) − y(t ) + d(t )⎟ + b0u(t )
⎝ T T ⎠ (9) as shown in Fig. 3.
K From Eq. (13), a reduced order extended state for x2 is,
where, b0 may represent known part of b = T , and Δb be an un-
known, considering as a modeling error. z2̇ (t ) = −ω0z2(t )−ω0b0u(t ) + ω0ẏ (t ) (17)
The lumped disturbance say, f(t) is considered as,
⎛ 1 1 ⎞
f (t ) = ⎜ Δbu(t ) − y(t ) + d(t )⎟
⎝ T T ⎠ (10)

From Eq. (9) and (10),

ẏ (t ) = f (t ) + b0u(t ) (11)

The primary task of ADRC is to implement ESO which can provide

estimation of f(t) such that ADRC can compensate its effect on the
model by means of disturbance rejection. The ADRC is designed by Fig. 2. Second order ADRC topology.

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
4 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

4. Stability analysis

4.1. Gain-phase margin approach

For the analysis and design of practical control systems, gain

margin and phase margin are the two important specifications. The
frequency domain approaches such as Nyquist, Bode and Nichols,
gives gain margin and phase margin in simple manner. But these
aforementioned approaches are not suitable for the systems with
Fig. 3. Reduced order ESO based ADRC.
two or more adjustable parameters. The gain margin and phase
margin is used for plotting the boundaries of constant gain margin
Though deepening measuring noises can be avoided by direct
and phase margin with the effects of adjustable parameters.
numerical differentiation in Eq. (17), by defining a new state as
Consider the system with open loop transfer function is
z = z 2 − ω0 y
G(s ) = num(s)/den(s ) (26)
Therefore, Eq. (17) can be rearranged as
Let s = jω ; then, Eq. (26) becomes
z ̇ = − ω0z − ω02y − ω0b0u (18) G( jω) = num( jω)/den( jω) (27)
forms reduced order ESO. Eq. (27) can be expressed in terms of its magnitude and phase,
From Eq. (18), extended state z2 results as, such as
ω0(sy − b0u)
z2 = G( jω) = |G( jω)|e jθ (28)
(s + ω0) (19)
where,
and from Eq. (14), the control input corresponds to,
Kp( s + ω0)( r − y) G( jω) = Re(G( jω))2 + Im(G( jω))2 (29)
ω0 y
u= − .
b0s b0 (20) and
The closed loop transfer function becomes, θ = ∠G( jω) = tan−1{ Im(G( jω))/Re(G( jω))} (30)
KKp( s + ω0)
GrLADRC = . From Eqs. (27) and (28),
( )
b0Ts 2 + b0 + Kω0 + KKp s + KKpω0 (21)
den( jω) + Ae−jϕ num( jω) = 0 (31)
Without loss of generality, consider a stable system
where,

G(s ) =
( 1 + τs ) 1/|G( jω)| = A (32)
( 1 + T1s)( 1 + T2s) (22)

where, z0 = − 1/τ < 0, p1 = − 1/T1 < 0, p2 = − 1/T2 < 0. The step

θ + 180deg = ϕ (33)
response of system (22) is monotonically increasing, if and only if,
T1 ≥ τ Eq. (31) is the closed-loop characteristic equation with the gain-
phase margin tester (Ae−jϕ ).
Proposition 1. [45] The step response of system (21) is mono-
The system has considered with controller parameters:
tonically increasing without overshoot, if and only if following con- b0 , Kp, ω0… and hence, Eq. (31) can be written as,
ditions hold:
F ( jω) = F (b0 , Kp, ω0, …, A, ϕ, jω) = den( jω) + Ae−jϕ num( jω) (34)
(i) For 0 < ω0 ≤ 1/T , there is
After some algebraic manipulations and decomposing the char-
b0 > 0, Kp > 0 (23) acteristic equation into two stability equations,
(ii) For ω0 > 1/T , there is Fr (b0 , Kp, ω0, …, A, ϕ, jω) = 0 (35)
Kω0
0 < b0 ≤ , Kp > 0 and
Tω0 − 1 (24)
Fi(b0 , Kp, ω0, …, A, ϕ, jω) = 0 (36)
or Assume that Eqs. (35) and (36) are linear functions of b0 , Kp . Then,
⎧ Kω0 Fr (b0 , Kp, ω0, …, A, ϕ, jω) = b0B1 + KpC1 + D1 (37)
⎪ b0 >
⎪ Tω0 − 1
⎨
⎪ (2 Tω0 − 1)b0 − Kω0 − 2 c
⎪
⎩
0 < Kp < Fi(b0 , Kp, ω0, …, A, ϕ, jω) = b0B2 + KpC2 + D2 (38)
K (25)

where, c = Tω0b0(Tω0b0 − b0 − Kω0) where B1, B2, C1, C2, D1 and D2 are functions of ω0, … , A , ϕ and ω.
After solving Eqs. (37) and (38) for b0 and Kp, one has
For better control performance, the boundary region for re-
duced order linear ADRC is obtained from Eq. (25). By keeping ω0 C1·D2 − C2·D1
b0 =
fixed, concerning b0 and Kp, the feasible region is confined by the B1·C2 − B2·C1 (39)
two axes of coordinates with a boundary curve.

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

D1·B2 − D2 ·B1
Kp =
B1·C2 − B2·C1 (40)

Consider A ¼ 1 and ϕ = 0, and set ω0, … equal to constants: then

for various values of ω, a locus that contains the stability boundary of
the system without the gain-phase margin tester can be plotted in
the b0 vs. Kp plane. If A is assumed equal to constant value and ϕ = 0,
the locus in the b0 vs. Kp plane is a boundary of constant gain margin.
On the other hand, if A ¼ 1 and ϕ is assumed equal to constant value, Fig. 4. The block diagram of the linear reduced order ADRC.
then the locus is a boundary of constant phase margin. These
boundaries are called D-partition boundaries denoted as ∂.
According to conformal property, we can decide which side of the
4.2. D-partition and identification of stability region curve the left-half of s plane is mapped into by G1(s ). This side
usually indicates the stability region. Comparing Eqs. (47) and (49),
D-partition method can be employed to analyze and identify the we can find that when the x1 axis coincides with the b0 axis, they
stability region. The technique provides an effective approach for have the same split points in the x1(b0) axis. Thus, the stability
studying the closed loop stability of systems with adjustable intervals of b0 can be obtained. If controller has more parameters,
parameters. we can prespecify the values of other parameters and analyze the
The difference between the open loop transfer function with stability region in a two-dimensional plane.
and without the tester is ignored, then the open loop transfer
function is represented as G(s), and the closed loop characteristic 4.3. Stability margin calculation
equation is,
The block diagram of the control is shown in Fig. 4.
f (s ; K , T , τ , …, b0 , Kp, ω0) = 1 + G(s ) = 0 (41) After some block diagram algebraic manipulation of Fig. 4, the
open loop transfer function of the system is
where K , T and τ etc. are system parameters, b0 , Kp and ω0 etc.
are controller parameters. The boundaries of stability region are K ⎡⎣ s Kp + ω0 + Kpω0⎤⎦e−τs
( )
defined by the equations f (0; …) = 0, f (∞ ; …) = 0 and G(s ) =
b0s( Ts + 1) (50)
f ( ± jω; …) = 0 for ω ∈ (0, ∞) = 0, which are similar to the gain-
phase margin approach. The closed loop characteristic equation with gain-phase margin
The stability region has identified by the following procedure. tester is
For convenience, we assume that the closed loop characteristic
equation is linear with controller parameters. K [s(Kp + ω0) + Kpω0]e−τs
1 + Ae−jϕ =0
The controller has two parameters such as, b0 and Kp, and as- b0s(Ts + 1) (51)
sume that the closed loop characteristic equation is,
Eq. (51) is equivalent to:
G1(s ) − b0 ·Q (s ) − Kp·R(s ) = 0 (42)
b0s(Ts + 1) + Ae−jϕ K [s(Kp + ω0) + Kpω0]e−τs = 0 (52)
Suppose that we cut the D-partition boundaries with Kp = c ,
where c is a constant real number. Then, eq. (42) can be written as: From Eq. (52),

G1(s )/Q (s ) − b0 − Kp·R(s )/Q (s ) = 0 ∂0: Kp = 0 (53)

(43)

Let s = jω , then
G1( jω)/Q ( jω) = P1(ω) + jP2 (ω) ∂∞: b0 = 0 (54)
(44)
and ∂ω will be determined by the following equations

R( jω)/Q ( jω) = P3(ω) + jP4 (ω) ⎧

⎪ b0 = Δ1 / Δ
(45) ⎨
⎩ Kp = Δ2 /Δ
⎪

From Eqs. (43)–(45), (55)

P1(ω) + jP2 (ω) − b0 − cP3(ω) − jcP4 (ω) = 0 where,

(46)

Hence, Δ = Kω[(ω0 y − ωx + Tωω0x + Tω2y)sin(τω)

+ (Tωω0 y − Tω2x − ω0x − ωy)cos(τω)] (56)
⎧ b = P (ω) − cP (ω)
⎪

⎨ 0 1 3

⎩ P2(ω) − cP4(ω) = 0
⎪
(47) Δ1 = K2ωω02(x2 + y2 ) (57)
Eq. (42) is rewritten as
Δ2 = Kω2ω0[(Tωx + y)cos(τω) + (x − Tωy)sin(τw )] (58)
b0 = G1(s )/Q (s ) − KPR(s )/Q (s ) ≜ G¯ (s ) (48)

Consider G(s) as a map from the s plane to the two-dimensional x = A cos(ϕ), y = A sin(ϕ) (59)
plane x1 vs. y1. Let G¯ ( jω) = x1 + jy1 and Kp = c , then
⎧ x1 = P1(ω) − cP3(ω)
⎪
⎨
⎩ y1 = P2(ω) − cP4(ω)
⎪ 5. Application of linear ADRC to TITO process
(49)

when ω varies from −∞ to +∞, we obtain a directed curve. Let s be The concept of ADRC is to focus on disturbance rejection as the
equal to a very small negative real number s0 and compute G(s0). primary task, and the “active” part comes from mitigating the

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
6 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

disturbance before it interrupts in systems output, which keeps ⎧

⎪ e11 = e12 − β1ω0e11 − β1ω0e21
̇
consistency in plant response. The higher order system without ⎨
⎪
⎩ e21 ̇ = h1(x, d) − h1(z, d) − β2ω02e11 − β2ω02e21 (64)
time delay is realized approximately from a system with time
delay in the input, which may lead error between the plant and its Now, let ηi =
e1i
, i ¼ 1,2 then Eq. (64) can be rewritten as,
model as a disturbance as well as unmodeled dynamics, para- ω0i − 1

metric uncertainty, external disturbance, these are considered as h1( x, d) − h1( z, d)

lumped disturbance and ADRC actively compensate those. In this
paper, the design of proposed control technique is based on de-
coupled systems, where the controller is tuned with the help of
FOPTD model of decoupled system. The decoupled systems are
used only for the purpose of tuning of proposed controller. Cor-
responding control structure for process with two-inputs two-
outputs is shown in Fig. 5.
5.1. Design of modified linear ESO to TITO process
The standard linear ESO is able to estimate the structured un-
certainty (variation due to system parameters), unstructured un-
certainty (unmodeled dynamics) and external disturbances. But
for the considered TITO coupled plant it is unable to provide full
estimate of the effect of coupling phenomenon on the controlled
outputs. Due to which the control law based on the standard linear
ESO can not be eliminate completely the interaction effect on the
corresponding controlled outputs. Thus, in order to nullify the
coupling effects of control inputs u1 to y2, and u2 to y1, the state
observers for loop-1 (ESO-1) and loop-2 (ESO-2) are modified, so
that, it can estimate the interactions within the loops as an ex-
tended states in the form
⎧ e11 = y1 − z11, e21 = y2 − z21
⎪ and regulated towards zero, i.e. z21 ≈ 0 and e21 ≈ y2.
Modified ESO−1⎨ z11
̇ (t ) = z12(t ) + l1e11 + l1e21 + b0 u1(t )
1
⎪
⎩ z12̇ (t ) = l2e11 + l2e21
⎧ e21 = y2 − z21, e11 = y1 − z11
⎪
Modified ESO−2⎨ z21
̇ (t ) = z22(t ) + l1e21 + l1e11 + b0 u2(t )
2
⎪ modified reduced order state for ESO-1 and ESO-2 are described as
⎩ z22
̇ (t ) = l2e21 + l2e11
where, l1 and l2 are observer gain parameters.
η ̇ = ω0Aη + B + ω0Ce21
ω0 (65)
where,
⎡ − β 1⎤ ⎡ −β ⎤
A=⎢
1 ⎥, B = ⎡⎣ 0 1⎤⎦T , C = ⎢ 1⎥
⎢⎣ −β2 0⎥⎦ ⎣⎢ −β2 ⎥⎦
Proposition 2. [46] If the observer gain l1 and l2 are chosen such
that matrix A is Hurwitz matrix as well as the estimation error e21 is
bounded, then the observer estimation error, e for the modified linear
ESO of loop-1 is bounded for any bounded h1(x, d ).
Proposition 3. [47] Assuming that the function h1(x, d ) with respect
to x is globally Lipschitz, that is, there exists a constants ω0 > 0 and m
such that, |h1(x, d ) − h1(z, d )| ≤ m ∥ x − z ∥ and lim Ae e(t ) = 0.
t →∞
Similarly, the convergence of modified linear ESO of loop-2 can be
derived as per the Section 5.1.1
5.1.2. Design of modified reduced order linear ESO
The better dynamic performance can be achieved with mod-
ified reduced order linear ESO under the following assumptions.
Assumption 1. The estimation error e21 with respect to ESO-1
satisfy the following conditions: (1) e21 is bounded; (2) y2 deviates
from equilibrium position and it is bounded; (3) z21 is bounded
(60) Assumption 2. The estimation error e11 with respect to ESO-2
satisfy the following conditions: (1) e11 is bounded; (2) y1 deviates
from equilibrium position and it is bounded; (3) z11 is bounded
and regulated towards zero, i.e. z11 ≈ 0 and e11 ≈ y1.
Under the Assumptions (1) and (2), and from Eqs. (60)–(61),
(61)
z12
̇ (t ) = −ω0z12(t )−ω0b0 u1(t ) + ω0y1̇ (t ) + ω0y2̇ (t )
1 (66)

5.1.1. Analysis of estimation error of modified linear ESO

z22
̇ (t ) = −ω0z22(t )−ω0b0 u2(t ) + ω0y1̇ (t ) + ω0y2̇ (t )
2 (67)
Consider decoupled subsystem of loop-1 is in the form of
⎧ ẋ = x + b u respectively.
⎪ 11 12 01 1
⎨
⎪
From Eq. (14), the corresponding control laws are obtained as
⎩ x12
̇ = h1(x, d) (62)
K p1(r1 − y1) − z12
u1 =
Let e1i = x1i − z1i , i ¼ 1,2 and the extended state observer is de- b01 (68)
signed as
⎧ e11 = y1 − z11, e21 = y2 − z21
⎪ K p2(r2 − y2 ) − z22
⎨ z11
̇ (t ) = z12(t ) + l1e11 + l1e21 + b0 u1(t )
1 u2 =
⎪ b02 (69)
⎩ z12̇ (t ) = l2e11 + l2e21 + h1(z, d) (63)
where, K p , K p , b01, b02 are the controller parameters, r1, r2, and u1,
Then the observer estimation error can be written as, 1 2
u2 be the reference inputs and outputs of controller 1 (ADRC-1)
and controller 2 (ADRC-2), respectively.

5.2. Case study

A multivariable two-input two-output system is illustrated to

demonstrate the performance of proposed control strategy as
compared with PID controllers suggested by [7,8]. In this paper,
the model of binary distillation column by Wood and Berry [3] has
Fig. 5. Control structure for process with two inputs and two outputs. considered, which is as shown below:

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

Fig. 6. (a) Boundary region concerning b01 and K p for ADRC-1, (b) Boundary region concerning b02 and K p for ADRC-2.
1 2

Fig. 7. (a) D-partition boundaries and stability boundary identification for loop 1, (b) D-partition boundaries and stability boundary identification for loop 2.

Fig. 8. (a) Various boundaries when ω0 = 5 for loop 1, (b) Various boundaries when ω0 = 5 for loop 2.

Table 1
⎡ 12.8e−s −18.9e−3 s ⎤
Controller parameters.
⎡ XD(s )⎤ ⎢ 16.7 s + 1 21 s + 1 ⎥ ⎡ FR(s )⎤
⎢ ⎥=⎢ ⎥⎢ ⎥
⎣ XB (s ) ⎦ ⎢ 6.6e−7 s −3 s ⎥ ⎣ F (s ) ⎦
ADRC-1 K p = 0.1943 b01 = 51.2124 ω0 = 5 K1 ¼ 1.5
1
⎢ − 19.4 e ⎥ S
ADRC-2 K p = 0.0795 b02 = − 47.8368 ω0 = 5 K2 ¼ 0.7
2 ⎣ 10.9 s + 1 14.4 s + 1 ⎦ (70)

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
8 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 9. (a) Output response, y1 for WB column under the effect of measurement noise, (b) Output response, y2 for WB column under the effect of measurement noise.

Fig. 10. (a) Control signal, u1 for WB column under the effect of measurement noise, (b) Control signal, u2 for WB column under the effect of measurement noise.

Fig. 11. (a) Output response, y1 for WB column, (b) Output response, y2 for WB column.

The distillate and bottom products (XD and XB) are the measure- The minimal FOPDT representation of the decoupled subsystems
ments, as outputs y1 and y2, respectively, the reflux flows FR and are,
the reboiler steam flow FS are the manipulated variables as inputs
u1 and u2, respectively. 6.37
g11FOPDT (s ) = e−1.06 s ,
The decoupler for the Wood Berry plant is obtained in [7] is, 5.41 s + 1
−9.65
⎡ 1.48(16 s + 1) −2 s ⎤ g22FOPDT (s ) = e−2.16 s
⎢ 1 e ⎥ 4.68 s + 1 (72)
21.0 s + 1
d(s ) = ⎢ ⎥
⎢ 0.34(14.4 s + 1) −4 s ⎥
⎢⎣ e 1 ⎥⎦ For those subsystems, PI controllers are designed by Tavakoli et
10.9 s + 1 (71) al.'s [7] and Maghade, Patre's [8] as

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

Fig. 12. (a) Control signal, u1 for WB column, (b) Control signal, u2 for WB column.

Fig. 13. (a) Output response, y1 for WB column under parametric uncertainty, (b) Output response, y2 for WB column under parametric uncertainty.

Fig. 14. (a) Control signal, u1 for WB column under parametric uncertainty, (b) Control signal, u2 for WB column under parametric uncertainty.

⎡ 0.074 ⎤ respectively, and PID controller is derived by Maghade, Patre's as

⎢ 0.41 + s
0 ⎥
CT (s ) = ⎢
0.024 ⎥
⎢⎣ 0 −0.12 − s ⎥⎦ ⎡ 0.9733 + 0.0881 + 2.6887 s ⎤
⎢ s 0 ⎥
CMPID(s ) = ⎢ ⎥
5.5252 s + 1
and ⎢ −0.3134 −
0.0304
− 0.8070 s ⎥
⎢⎣ 0 s
5.1499 s + 1
⎥⎦
⎡ 0.0881 ⎤
⎢ 0.4867 + s
0 ⎥
CMPI (s ) = ⎢
0.0304 ⎥ Considering reduced FOPDT models of the decoupled subsystems
⎢⎣ 0 −0.1567 − s ⎥⎦
in Eq. (72), and after delay approximation the decoupled sub-
system becomes,

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
10 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Table 2
Summary of error performance indices without parametric uncertainty.

Controller Input(u) Error indices

Output(y) IAE ISE ITAE ITSE

Proposed u1-y1 18.07 9.806 641.3 82.35

u2-y2 28.71 15.5 4430 2474
Tavakoli et al. u1-y1 12.34 3.391 1539 211.2
u2-y2 28.84 15.44 5162 2558
Maghade, Patre PID u1-y1 18.96 5.856 2441 408.8
u2-y2 40.85 24.52 7599 4228
Maghade, Patre PI u1-y1 15.84 4.165 2176 357.9
u2-y2 35.92 19.75 6612 3362
Fig. 15. Two tanks multivariable level control system.

Table 3
Summary of error performance indices with parametric uncertainty.

Controller Input(u) Error indices

Output(y) ITAE ITAE

Proposed u1-y1 18.73 9.403 877.2 91.23

u2-y2 27.34 14.89 4061 2349
Tavakoli et al. u1-y1 21.3 5.066 3151 472.2
u2-y2 54.79 26.94 1.007 × 104 4848
Maghade, Patre PID u1-y1 45.21 14.4 7971 2241
u2-y2 118.2 94.91 2.38 × 104 1.973 × 104
Maghade, Patre PI u1-y1 39.71 11.27 7116 1821
u2-y2 104.2 75.32 2.105 × 104 1.561 × 104

IAE – Integral absolute error, ISE – Integral square error, ITAE – Integral time mul- Fig. 16. Simulink model of two tanks multivariable level control system for
tiplied absolute error, ITSE – Integral time multiplied square error. experimentation.

Table 4
Summary of time domain specifications without parametric uncertainty.

Controller Input(u) and Time domain specifications

Output (y) ts Mp(%)

Proposed u1-y1 212.59 10.32

u2-y2 222.61 0
Tavakoli et al. u1-y1 245.39 22.65
u2-y2 286.11 66.17
Maghade, Patre PID u1-y1 284.35 33.62
u2-y2 321.13 99.98
Maghade, Patre PI u1-y1 267.51 32.97
u2-y2 303.68 87.32

Fig. 17. Schematic diagram of two tanks multivariable level system.

Table 5 6.37 1
Summary of time domain specifications with parametric uncertainty. g11(s ) = ,
(5.411 s + 1) (1.065 s + 1)
Controller Input(u) and Time domain Specifications −9.655 1
g22(s ) =
(4.684 s + 1) (2.157 s + 1) (73)
Output (y) ts Mp(%)
The control laws u1 and u2 are simplified from Eqs. (66)–(69)
Proposed u1-y1 217.34 11.49 and designed based on by considering only dominant part of the
u2-y2 213.58 0
Tavakoli et al. u1-y1 349.95 30.10
decoupled system in Eq. (73) as follows:
u2-y2 413.43 88.11
Kp (s + ω0)(r1 − y1) ω0y1 ω0y2
Maghade, Patre PID u1-y1 965.64 47.11 u1 = 1
− − K1
u2-y2 1187.5 135.16 b01s b01 b01
Maghade, Patre PI u1-y1 926.43 44.93
u2-y2 1185.2 118.66 Kp (s + ω0)(r2 − y2 ) ω0y2 ω0y1
2
u2 = − − K2
b 0 2s b02 b02 (74)
ts – Settling time, Mp – Peak overshoot.

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 11

Fig. 18. (a) Level of tank 1, (y1) to input (u1), (b) Level of tank 1, (y1) to input (u2).

Fig. 19. (a) Level of tank 2, (y2) to input (u1), (b) Level of tank 2, (y2) to input (u2).

Table 6 parameters region and stability regions are plotted in parameter

Controller parameters. plane as shown in Figs. 6–8, respectively. These relating para-
ADRC-1 K1 ¼ 0.1
meters b01, K p , and b02 , K p are selected from stability region and
K p = 0.0306 b01 = 0.7982 ω0 = 5 1 2
1
ADRC-2 K p = 0.0266 b02 = 0.7969 ω0 = 5 K2 ¼ 0.1 those values are shown in Table 1. The feasible parameter regions
2
for ADRC-1 and ADRC-2 are obtained from Eq. (25), shown in Fig. 6
(a) and (b), respectively.
The laws mentioned above have high enough capability to reduce
the interaction between control loops by adjusting the gain
parameters K1 > 0 and K2 > 0. By keeping ω0 fixed, the feasible

Fig. 20. (a) Boundary region concerning b01 and K p for ADRC-1, (b) Boundary region concerning b02 and K p for ADRC-2.
1 2

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
12 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 21. (a) D-partition boundaries and stability boundary identification for loop 1, (b) D-partition boundaries and stability boundary identification for loop 2.

Fig. 22. (a) Various boundaries when ω0 = 5 for loop 1, (b) Various boundaries when ω0 = 5 for loop 2.

6. Simulation results
Considering the decoupled systems from Eq. (72) and the gain-
phase margin specifications are supposed to be gain margin,
GM > 2.5(8 dB) and phase margin, PM > 60°. When A ¼ 1, ϕ = 0,
ω0 = 5, the D-partition boundaries for loop 1 and loop 2, are
plotted as shown in Fig. 7(a) and (b) respectively, and also the
various boundaries with ω0 = 5 for loop 1 and loop 2, are shown in
Fig. 8(a)-(b). A negative real number close to the imaginary axis of
s-plane is mapped as a point labelled with * and the stability re-
gion should be the region containing the point labelled with *.
are tested. The process outputs y1 and y2 and, an applied control
signals u1 and u2 are shown in Figs. 11 and 12, respectively.
To assess the performance under parametric deviations, 20%
uncertainty is introduced into the system's gain, the time constant
and time delay, and corresponding output responses, as well as
control signals by proposed and other controllers, are shown in
Figs. 13 and 14, respectively.
From Tables 2–5, it can be demonstrated that the proposed
controller gives a superior performance without parametric un-
certainty and with parametric uncertainty respectively as com-
pared to other controllers.

6.1. Effect of measurement noise

The stochastic simulations are presented in Figs. 9 and 10,
2
where the normally distributed noise with variance σnoise = 0.01
was added to the process output. The controller parameter values
used for stochastic simulation are selected from Table 1. While the
controller based on observer are less susceptible to measurement
noise as compare to direct measurements, however the oscilla-
tions in controller output will still occur with some noise level.
From the simulations, it is demonstrate that the proposed con-
troller is able to compensate the effect of noise to the controller
output.
6.2. Effect of structural uncertainties
To verify the effectiveness of proposed controller, set point
changes in output y1 at time t ¼ 0 and output y2 at time t ¼ 150
7. Experimental setup
The laboratory setup of two tanks multivariable level process,
simulink model of the level control system and schematic diagram
of a level process with data acquisition are shown in Figs. 15–17. A
setup consists of two tanks, two-level transmitter, sump, two
pump, variable frequency drives (VFD), current to voltage con-
verter, the voltage to current converter, DAQ card, PCI 6014E and
BNC 2120 connector. DAQ card, PCI 6014E with BNC 2120 con-
nector is used to interface the level process with personnel com-
puter. The control algorithm is implemented in Mathworks
MATLABTM R2012a. A plant test is carried out to make a step
change in manipulated input by applying the step of 5 V to each
pump simultaneously and an empirical model of the level process
is achieved from an open loop step test data. The input voltage

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 13

Fig. 23. (a) Level of tank 1, (b) Level of tank 2.

Fig. 24. (a) Control signal of ADRC-1, (b) Control signal of ADRC-2.

Fig. 25. (a) Level of tank 1, (b) Level of tank 2.

signal is given into voltage to current converter (V/I converter) and help of capacitive type sensor with a transmitter having a range of
converted current signal is applied to the positive displacement 4–20 mA which corresponds to 0–100% level i.e., 0–100 cm and an
type pumps through VFDs. The water level is measured with the obtained voltage signal by using current-to-voltage converter is in

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
14 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 26. (a) Control signal of ADRC-1, (b) Control signal of ADRC-2.

the range of 0–5 V. The output data is the measured voltage signal. voltages are shown in Figs. 23 and 24, respectively.
The recorded input and output data with a sampling period of 1 s The effectiveness of the proposed controller under the para-
is applied for model identification is shown in Figs. 18 and 19. metric variations is evidenced by an experimental analysis with
changing set point of the level of tanks three times. With an in-
7.1. Plant step tests data for model identification fluence of parametric deviations, the original system retains un-
certain dynamics at each and every distinct steady state.
The FOPDT models of tank 1 and tank 2 (y1, y2) of level process In this experiment, a step change in the set point of tank 1 and
against the two pump voltages (u1,u2) are identified in transfer tank 2 from 0% to 30% is applied, respectively. Again, the level
function matrix as [9] change of 30–50% is enforced to both tanks at t ¼ 250 s and
⎡ 0.43 −5 s 0.172 −10 s ⎤
t ¼ 360 s, respectively. Now, the set point is reduced from 50% to
⎢ 29 s + 1 e 40 s + 1
e ⎥ 40% of the level of tanks at t ¼ 540 s and t ¼ 720 s, respectively.
Gp(s ) = ⎢ ⎥.
0.145 −10 s 0.37
e−5 s The effect of output responses and pump voltages (control inputs)
⎢⎣ 35 s + 1 e 27 s + 1
⎥⎦
are shown in Figs. 25 and 26, respectively.

8. Experimental results 9. Conclusion

The ideal decoupler is determined for two tanks interacting In this paper a modified reduced order extended state observer
level control system by the methodology used in [48], based control design through decoupled systems is proposed. In this
technique, the decoupled system is used for parameter tuning in
⎡ 2.55 −1 ⎤
⎢ e−5 s ⎥ control design whereas existing PID controllers exploit the decoupled
27 s + 1 40 s + 1
dp(s ) = ⎢ ⎥. system for controller parameter tuning as well as for decoupling the
⎢ −1 −5 s 2.5 ⎥ plant. The aim of proposed method is that it does not essentially
⎢⎣ 35 s + 1 e 29 s + 1 ⎥⎦ (75) require decoupled plant as compared to existing PID control tech-
Reduced FOPDT models of decoupled subsystems are, niques. The robust control system is obtained against the parameter
variations and external disturbances as well as tracking performance
0.9245 is guaranteed with proposed approach. Therefore, an introduced
gp (s ) = e−12.7 s ,
11FOPDT 113.2 s + 1 design turns out to be a good solution to compensate cross coupling
0.78 effect, parametric uncertainties and external disturbances in multi-
gp (s ) = e−12.7 s
22FOPDT 113.2 s + 1 (76) variable processes significantly. Simulation and experimental analysis
demonstrate that the proposed method improves the performance
respectively. The corresponding controller parameters for the two
and robustness of control system.
decoupled subsystems are as shown in Table 6 and feasible regions
for ADRC-1 and ADRC-2 are obtained from Eq. (25), shown in
Fig. 20(a) and (b), respectively.
References
Considering the decoupled systems from Eq. (76) and the gain-
phase margin specifications are supposed to be gain margin,
[1] Shinskey F. Process control systems: application, design, and tuning, chemical
GM > 2.5(8 dB) and phase margin, PM > 60°. When A ¼ 1, ϕ = 0,
engineering books. McGraw-Hill, New York 1996.
ω0 = 5, the D-partition boundaries for loop 1 and loop 2, are [2] Luyben WL, Vinante CD. Experimental studies of distillation decoupling. Kem
plotted as shown in Fig. 21(a) and (b) respectively, and also the Teoll 1972;29(8):499–514.
various boundaries with ω0 = 5 for loop 1 and loop 2, are shown in [3] Wood R, Berry M. Terminal composition control of a binary distillation col-
umn. Chem Eng Sci 1973;28(9):1707–17.
Fig. 22(a) and (b).
[4] Ogunnaike B, Ray W. Multivariable controller design for linear systems having
To verify the realistic effectiveness of the proposed algorithm, a multiple time delays. AIChE J 1979;25(6):1043–57.
step change is applied to tank 1 and tank 2 from 0% to 40% at time [5] Ziegler JG, Nichols NB. Optimum settings for automatic controllers. InTech
1995;42(6):94–100.
t ¼ 0, and 15% of set point change as a disturbance is imposed in
[6] Koivo H, Tanttu J. Tuning of PID controllers: survey of SISO and MIMO tech-
tank 1 at t ¼ 200 s and tank 2 at t ¼ 400 s, respectively. The niques. In: Proceedings of the IFAC intelligent tuning and adaptive control
corresponding output responses and control inputs as pump symposium; 1991. p. 75–80.

Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
 S.N. Pawar et al. / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎
[7] Tavakoli S, Griffin I, Fleming PJ. Tuning of decentralised PI (PID) controllers for
TITO processes. Control Eng Pract 2006;14(9):1069–80.
[8] Maghade DK, Patre BM. Decentralized PI/PID controllers based on gain and
phase margin specifications for TITO processes. ISA Trans 2012;51(4):550–8.
[9] Hajare VD, Patre BM. Decentralized PID controller for TITO systems using
characteristic ratio assignment with an experimental application. ISA Trans
2015;59(6):385–97.
[10] Hajare VD, Patre BM, Khandekar AA, Malwatkar GM. Decentralized PID con-
troller design for TITO processes with experimental validation. Int J Dyn
Control 2016:1–13 [URL 〈http://dx.doi.org/10.1007/s40435-016-0252-z〉].
[11] Han J. From PID to active disturbance rejection control. IEEE Trans Ind Electron
2009;56(3):900–6.
[12] Han J. Active disturbance rejection controller and its applications. Control
Decis 1998;13(1):19–23.
[13] Tian G, Gao Z. Frequency response analysis of active disturbance rejection
based control system. In: Proceedings of the IEEE international conference on
control applications; 2007. p. 1595–9.
[14] Wu D, Chen K. Frequency-domain analysis of nonlinear active disturbance
rejection control via the describing function method. IEEE Trans Ind Electron
2013;60(9):3906–14.
[15] Li J, Qi X, Xia Y, Pu F, Chang K. Frequency domain stability analysis of nonlinear
active disturbance rejection control system. ISA Trans 2015;56:188–95.
[16] Zheng Q, Gao L, Gao Z. On estimation of plant dynamics and disturbance from
input-output data in real time. In: Proceedings of the IEEE international
conference on control applications; 2007. p. 1167–72.
[17] Zhou W, Shao S, Gao Z. A stability study of the active disturbance rejection
control problem by a singular perturbation approach. Appl Math Sci 2009;3
(10):491–508.
[18] Csank J, Gao Z. Uncertainty reduction through active disturbance rejection. In:
Proceedings of the American control conference; 2008. p. 3689–94.
[19] Madoński R, Herman P. Survey on methods of increasing the efficiency of
extended state disturbance observers. ISA Trans 2015;56:18–27.
[20] Qing Z, Zhiqiang G. On practical applications of active disturbance rejection
control. In: Proceedings of the 29th Chinese control conference (CCC); 2010. p.
6095–100.
[21] Goforth F. On motion control design and tuning techniques. In: Proceedings of
the American control conference. Vol. 1; 2004. p. 716–21.
[22] Yang C, Ke Z. Study on the active disturbance rejection control of servo system.
In: Proceedings of the international conference on computer and communica-
tion technologies in agriculture engineering (CCTAE). Vol. 3; 2010. p. 364–7.
[23] Zheng Q, Gao Z. Motion control design optimization: problem and solutions.
Int J Intell Control Syst 2005;10(4):269–76.
[24] Tian G, Gao Z. Benchmark tests of active disturbance rejection control on an
industrial motion control platform. in: Proceedings of the American control
conference; 2009. p. 5552–7.
[25] Zhang Y, Cai L, Meng Q, Dong M. A novel active disturbances rejection fuzzy
servo system position controller. In: Proceedings of the 1st international
symposium on systems and control in aerospace and astronautics; 2006. p.
279–83.
[26] Zheng Q, Dong L, Gao Z. Control and rotation rate estimation of vibrational
mems gyroscopes. In: Proceedings of the IEEE international conference on
control applications; 2007. p. 118–23.
[27] Dong L, Zheng Q, Gao Z. On control system design for the conventional mode
of operation of vibrational gyroscopes. IEEE Sens J 2008;8(11):1871–8.
Please cite this article as: Pawar SN, et al. Modified reduced order observer based linear active disturbance rejection control for TITO
systems. ISA Transactions (2017), http://dx.doi.org/10.1016/j.isatra.2017.07.026i
15
[28] Dong L, Zheng Q, Avanesov D. The design and implementation of driving mode
control for vibrational gyroscopes. In: Proceedings of the American control
conference; 2008. p. 4419–24.
[29] Zheng Q, Dong L, Lee DH, Gao Z. Active disturbance rejection control for mems
gyroscopes. IEEE Trans Control Syst Technol 2009;17(6):1432–8.
[30] Qing Z, Zhiqiang G. Disturbance rejection in mems gyroscope: problems and solu-
tions. In: Proceedings of the 30th Chinese control conference (CCC); 2011. p. 6334–9.
[31] Xingling S, Honglun W. Back-stepping active disturbance rejection control
design for integrated missile guidance and control system via reduced-order
ESO. ISA Trans 2015;57:10–22.
[32] Hou Y, Gao Z, Jiang F, Boulter BT. Active disturbance rejection control for web
tension regulation. In: Proceedings of the 40th IEEE conference on decision
and control. Vol. 5; 2001. p. 4974–9.
[33] Zhou W, Gao Z. An active disturbance rejection approach to tension and ve-
locity regulations in web processing lines. in: Proceedings of the IEEE inter-
national conference on control applications; 2007. p. 842–8.
[34] Yang X, Cui J, Lao D, Li D, Chen J. Input shaping enhanced active disturbance
rejection control for a twin rotor multi-input multi-output system (TRMS). ISA
Trans 2016;62:287–98.
[35] Zheng Q, Chen Z, Gao Z. A dynamic decoupling control approach and its ap-
plications to chemical processes. in: Proceedings of the American control
conference; 2007. p. 5176–81.
[36] Zheng Q, Chen Z, Gao Z. A practical approach to disturbance decoupling
control. Control Eng Pract 2009;17(9):1016–25.
[37] Guannan L, Wen T, Qin Ling Z. Linear active disturbance rejection control for
the coordinated system of drum boiler-turbine units. Proc CSEE 2011;31
(23):94–100.
[38] Huang CE, Li D, Xue Y. Active disturbance rejection control for the alstom
gasifier benchmark problem. Control Eng Pract 2013;21(4):556–64.
[39] Han J. Extended state observer for a class of uncertain plants. Control Decis
1995;10(1):85–8.
[40] Xia Y, Shi P, Liu G, Rees D, Han J. Active disturbance rejection control for
uncertain multivariable systems with time-delay. IET Control Theory Appl
2007;1(1):75–81.
[41] Zheng Q, Gao Z. Predictive active disturbance rejection control for processes
with time delay. ISA Trans 2014;53(4):873–81.
[42] Zhao S, Gao Z. Modified active disturbance rejection control for time-delay
systems. ISA Trans 2014;53(4):882–8.
[43] Han KW, Chang CH. Gain margins and phase margins for control systems with
adjustable parameters. J Guid Control Dyn 1990;13(3):404–8.
[44] Hwang C, Hwang LF, Hwang JH. Robust d-partition. J Chin Inst Eng 2010;33
(6):811–21.
[45] Yang R, Sun M, Chen Z. Active disturbance rejection control on first-order
plant. J Syst Eng Electron 2011;22(1):95–102.
[46] Li S, Yang J, Chen WH, Chen X. Generalized extended state observer based
control for systems with mismatched uncertainties. IEEE Trans Ind Electron
2012;59(12):4792–802.
[47] Zheng Q, Gao L, Gao Z. On stability analysis of active disturbance rejection
control for nonlinear time-varying plants with unknown dynamics. In: Pro-
ceedings of the 46th IEEE conference on decision and control; 2007. p. 3501–6.
[48] Khandekar AA, Patre BM. Design and application of discrete sliding mode
controller for TITO process control systems. In: Azar AT, Zhu Q, editors. Ad-
vances and applications in sliding mode control systems, Vol. 576 of Studies in
computational intelligence. Springer International Publishing; 2015. p. 255–77.