You are on page 1of 17

Circuits Syst Signal Process

DOI 10.1007/s00034-016-0302-y

On the Robust Stability of Active Disturbance Rejection


Control for SISO Systems

Xiaohui Qi1 · Jie Li1 · Yuanqing Xia2 ·


Zhiqiang Gao3,4

Received: 23 February 2015 / Revised: 15 March 2016 / Accepted: 16 March 2016


© Springer Science+Business Media New York 2016

Abstract Active disturbance rejection control (ADRC) is a new practical control


technique, which can integrally and effectively deal with various nonlinearities, uncer-
tainties and disturbances (collectively called the total disturbance throughout this
paper). This paper addresses the problem of the robust stability analysis and design
of linear and nonlinear ADRC for SISO systems. Firstly, a nonlinear ADRC-based
control system with total disturbances is transformed into a perturbed indirect Lurie
system. Then, the Popov–Lyapunov method is used to study its global or local stability
and derive the robust stability bound on allowable total disturbance. Furthermore, if
the total disturbance of the system is known, an estimated region of attraction can be
obtained. In addition, the paper illustrates how the modeled linear dynamics can be
easily integrated into ADRC to improve both the performance and stability charac-
teristics. The above approach can also be extended to a linear ADRC-based control

B Yuanqing Xia
xia_yuanqing@bit.edu.cn
Xiaohui Qi
qi-xh@163.com
Jie Li
lijienewlife1234@163.com
Zhiqiang Gao
z.gao@ieee.org

1 Department of Unmanned Aerial Vehicle Engineering, Mechanical Engineering College,


Shijiazhuang 050003, China
2 School of Automation, Beijing Institute of Technology, Beijing 100081, China
3 Center for Advanced Control Technologies, Cleveland State University, Cleveland, OH 44115,
USA
4 Tianjin University of Technology, Tianjin, China
Circuits Syst Signal Process

system. A numeral example is presented to verify the convenience and practicability


of the proposed method.

Keywords Active disturbance rejection control · Lurie system · Robust stability ·


Popov criterion · Region of attraction

1 Introduction

The active disturbance rejection control (ADRC) systematically proposed by Han in


[9,10] gradually shows exuberant vitality, owing to its unique philosophy and excellent
control quality. It is error-driven rather than model-based, which is similar to the most
widely used PID today. It makes use of an extended state observer (ESO) to estimate
and compensate the internal and external disturbances and uncertainties, and a tracking
differentiator (TD) to generate the arranged transition process and extract differentia
of each order of the input signal. All of the above lead to enough stability margins and
strong robustness, which have been confirmed by theory analyses and many practical
applications, see, e.g., [1,2,5,15,16,18,22] and the references therein.
Nonlinear ADRC was firstly proposed, which prefers to employ nonlinear functions
in the design of ESO and the control law. In return, it improves the system dynamics
and is potentially much more effective in tolerance to uncertainties and disturbances.
However, it may make the system produce some complex and colorful nonlinear
behaviors, such as multiple equilibrium points, limit cycles, bifurcations and chaos.
Meanwhile, it is difficult to perform stability and performance analyses, which inhibits
the application of ADRC. Nevertheless, the theoretical research of nonlinear ADRC
has been launched with some achievements. Convergence and estimate error analyses
of the nonlinear ESO were performed by the self-stable region theory proposed by
Huang and Han [11,12]. Convergence of tracking differentiator can be found in [3,4,
19]. Performance analysis of nonlinear ADRC-based control system was performed
by Wu [20].
There are some remarkable efforts to solve the problem of stability analysis of
a nonlinear ADRC-based control system. In [20,21], stability analysis of an ADRC-
based control system was performed via the describing function method. However, it is
only an approximate method, which requires the linear dynamics with property of low-
pass filter. Meanwhile, it is complicated to transform the ADRC-based control system
with several nonlinear terms into a Lurie-like system by graphics transformation. What
is more, the describing function method is confined to determine the existence and
stability of limit cycle in a nonlinear system. In [6,7], time domain convergences of
the NLADRC system were proved under the assumption that the derivative of the
total uncertainty is bounded. However, a suitable Lyapunov function still has not been
presented for the typical nonlinear ADRC-based control system, and there are many
limitations for the parameters, which make the method very difficult to apply.
To make it more close to practical application, linearized and parameterized ADRC
was proposed by Gao [8]. Compared with the nonlinear ADRC-based system, much
more work on performance and stability analysis has been done for the linear one
in recent years. Much progress about time domain performance analysis of the linear
Circuits Syst Signal Process

ADRC-based system has been made [23,24], which lays a solid theoretical foundation
and provides a strict theoretical support for application. At the same time, much success
about frequency domain performance analysis has also achieved [5,25,26], which
is significantly important that the ADRC framework is understood using the almost
universal frequency domain analysis languages shared by practicing control engineers,
including both bandwidth and stability margins. However, although the linear ADRC
shows good control performance and is easier in theoretical analysis, it is only a special
case of ADRC.
The frequency domain stability analysis of the linear ADRC-based closed-loop
system is mostly done for a linear time-invariant plant [5,26]. Meanwhile, the time
domain stability analysis is always under the assumption that the derivative of the total
disturbances is bounded [27]: When the model of the plant is known, the linear ESO
and the linear ADRC-based closed-loop system are asymptotically stable; when there
exists a wide range of uncertainty of the plant, the estimate error of the linear ESO
and the tracking error of the linear ADRC-based closed-loop system are bounded and
decrease with the increasing bandwidth. However, the method is not direct to judge
the stability of the system.
Stability is a fundamental problem in the analysis and design of an automatic control
system, on which is this paper concerned. In general, frequency domain stability
analyses for both nonlinear and linear ADRC-based control system are based on a
nominal linear model of the plant; time domain stability analyses are performed under
assumptions that are not easy to apply to practical systems. Due to the presence of
nonlinearities, uncertainties and disturbances in practical applications, it is necessary
to perform quantitative robust stability analysis, which still has not been effectively
solved yet. To resolve this problem better, this paper performs the robust stability
analysis for ADRC-based SISO systems via the Popov–Lyapunov method, which is
based on a common assumption that the total disturbances meet linear growth bound.
The main contributions are as follows:
(1) The ADRC-based control system is transformed into a perturbed indirect Lurie
system, of which a Popov–Lyapunov method is proposed to analyze the robust
stability. In the design of the controller, the modeled linear dynamics can be
easily integrated into ADRC, which improves both the performance and stability
characteristics.
(2) The proposed method is applicable to both nominal system and robust system.
The robust stability bound on allowable total disturbances is derived. If the total
disturbances of the system are known, an estimated region of attraction can be
obtained.
(3) The proposed method is applicable to both the linear and nonlinear ADRC-based
control systems and any system that can be transformed into an indirect Lurie
system. Compared with the existing methods, there are no apriori assumptions
and it can effectively deal with robust stability.
The rest of this paper is organized as follows. Section 2 introduces the framework
and algorithm of ADRC and transforms the ADRC-based control system into an
indirect Lurie system. In Sect. 3, the Popov–Lyapunov method is proposed to analyze
the robust stability, which guarantees the global or local stability of the indirect Lurie
Circuits Syst Signal Process

system. In Sect. 4, a numeral example is presented to illustrate the application of the


proposed method. Finally, some concluding remarks are drawn in Sect. 5.

2 System Transformation

2.1 ADRC Algorithm and Framework

Consider the following typical SISO system described as



x (n) = f (x, ẋ, . . . , x (n−1) , w) + bu
(1)
y=x

where y is the regulated output, u is the input force and b is the gain coefficient. Its
state space model is described as


⎪ ẋ1 = x2


⎨ 2 = x3
⎪ ẋ
..
⎪ .



⎪ ẋ = f (x1 , x2 , . . . , xn , w) + bu
⎩ n
y = x1
(2)

Rewrite it as


⎪ ẋ1 = x2


⎨ 2 = x3
⎪ ẋ
..
⎪ .



⎪ ẋ = f 1 (x) + g(x, u) + b0 u
⎩ n
y = x1
(3)

where f 1 (x) = an x1 + an−1 x2 + · · · + a1 xn , g(x, u) = f 2 (x1 , x2 , ..., xn , w) + (b −


b0 )u. f 1 (x) represents the modeled linear dynamics; g(x, u) is treated as the total
disturbances that represent the combined effects of the internal nonlinear and un-
modeled dynamics and external disturbances on the derivate of xn . The core of ADRC
is to estimate and compensate g(x, u), which is very different from other control
strategies. It is a big breakthrough of ’internal model theory’ and ’absolute invariance
principle,’ which is one of the main reasons why ADRC has strong robustness.
ADRC generally consists of three components, which are tracking differentiator
(TD), ESO and state error feedback control law (SEF), which will be simply introduced
as follows.
TD generates the arranged transition process and extracts differentia of each order
of the input signal. Transition process is very helpful and necessary in some cases.
Referring to [2], a nonlinear TD is designed in the form
Circuits Syst Signal Process



⎪ v̇1 = v2



⎪ v̇2 = v3
⎨.
..



⎪ v̇n−1 = vn  


⎩ v̇n = λn ψ v1 − r, v2 , . . . , vn
λ λn−1
(4)

where r is the input force, vi (i = 1, 2, . . . , n) is the output and λ is the adjustable


speed factor.
ESO is used to estimate and compensate the total disturbances g(x, u), which is
the core and essence of the ADRC. Referring to [9], a nonlinear ESO is designed in
the form of ⎧

⎪ e = z1 − y



⎪ ż 1 = z 2 −β 1 · ϕ1 (e)

⎨ ż 2 = z 3 −β 2 · ϕ2 (e)
.
⎪ ..




⎪ ż n = z n+1 −β n · ϕn (e) + f 1 (z) + b0 · u


ż n+1 = −β n+1 · ϕn+1 (e)
(5)

where z i (i = 1, 2, . . . , n + 1) are the outputs of the ESO, βi (i = 1, 2, . . . , n + 1) are


the observer gains, e is the observer error and b0 is a constant that is approximated to
b, f 1 (z) = a n z 1 + an−1 z 2 + · · · +a 1 z n . Additionally, the inputs of the ESO are the
system output y and the control signal u, and the output z n+1 of the ESO provides an
estimate of g(x, u). It is worth noting that modeled linear dynamics f 1 (x) is used in
the ESO, which lightens the burden of ESO. ϕi (e)(i = 1, 2, . . . , n + 1) is a nonlinear
function, particularly, defined as

e/δ 1−α |e| ≤ δ
ϕ(e) = ϕi (e)= fal(e, α, δ) = (6)
|e|α sgn(e) |e| > δ

The fal(e, α, δ) function was proposed by Han [9] and plays an important role in
the newly proposed ADRC framework, due to its characteristics of ‘small error, big
gain; big error, small gain.’ α and δ are two important parameters to be predetermined,
and fal(e, α, δ) is denoted as fal(e) throughout this paper.
SEF is used to restrain the residual error and achieve the desired control goal. The
control law is designed as
u 0 − z n+1 − f 1 (z)
u= (7)
b0

where u 0 may employ the simple linear proportional and derivative control law in the
form of


n
u0 = ki (vi − z i ) (8)
i=1
Circuits Syst Signal Process

Fig. 1 Nonlinear ADRC for a SISO plant

where ki (i = 1, 2, · · · , n) are the controller gains.


The TD, ESO and SEF together develop a type of ADRC algorithm for a SISO
plant, and its framework is shown in Fig. 1.

2.2 System Transformation

Assumption A1 The input r is zero and all the outputs vi (i = 1, 2, . . . , n) of TD are


zero.

Substituting Eqs. (7) and (8) into Eq. (3), and let X = [x1 , x2 , . . . , xn ]T , Z =
[z 1 , z 2 , . . . , z n ]T , we obtain


Ẋ = A11 X + A12 Z + A13 (z n+1 − g(x, u))
(9)
y = x1

where

⎡ ⎤
0 1 0 ··· 0
⎢ 0 0 1 ··· 0 ⎥
⎢ ⎥
⎢ .. .. ⎥
A11 =⎢ . . ⎥
⎢ ⎥
⎣ 0 0 ··· 0 1 ⎦
an an−1 · · · a2 a1
⎡ ⎤
0 ··· 0 0
⎢ .. ⎥
⎢ . ⎥
A12 =⎢ ⎥
⎣ 0 ··· 0 0 ⎦
−k1 − an −k2 − an−1 · · · −kn − a1
A13 = [0, 0, . . . , 0, −1]T ∈ R n
Circuits Syst Signal Process

Substituting Eqs. (7) and (8) into Eq. (5), we obtain



⎨ Ż = A21 Z + A22 u 
ż = βn+1 u 
⎩ n+1 
u = −ϕ(e)
(10)

where
⎡ ⎤
0 1 0 ··· 0
⎢ 0 0 1 ··· 0 ⎥
⎢ ⎥
⎢ .. .. ⎥
A21 =⎢ . . ⎥
⎢ ⎥
⎣ 0 0 ··· 0 1 ⎦
−k1 −k2 · · · −kn−1 −kn
A22 = [β1 , β2 , · · · βn ]T

Combining Eqs. (9) and (10), we obtain




⎪ Ẋ = A11 X + A12 Z + A13 (z n+1 − g(x, u))


⎨ Ż = A21 Z + A22 u 
ż n+1 = βn+1 u  (11)


⎪ σ = c1 X + c2 Z
⎪ T T
⎩ 
u = −ϕ(σ )

where c1 = [−1, 0, . . . , 0]T ∈ R n , c2 = [1, 0, . . . , 0]T ∈ R n .


Let Y = A11 X + A13 z n+1 , we obtain


⎪ Ẏ = A11 Y + A11 A12 Z + A13 βn+1 u  − A11 A13 g(x, u)

Ż = A21 Z + A22 u 
⎪ σ = c1T A−1

T −1
11 Y + c2 Z − c1 A11 A13 z n+1
T
⎩ 
u = −ϕ(σ )
(12)

Rewrite it as
⎧·


⎪ x̃ = A x̃ + bu  + E(x, u)

ξ̇ = u 

⎪ σ = cT x̃ + ρξ

⎩ 
u = −ϕ(σ )
(13)
 
 T  T A11 A11 A12
where x̃ = Y Z , E(x, u) = −A11 A13 g(x, u) 0 , A = ,b=
0 A21
 
A13 βn+1   βn+1
, cT = c1T A−1
11 c2T , ρ = −c1T A−1
11 A13 βn+1 = − an .
A22
The system (13) can be expressed by the block diagram as shown in Fig. 2, which
is called perturbed indirect Lurie system.
Circuits Syst Signal Process

Fig. 2 Block diagram of system (13)

Regardless of the disturbance term E(x, u), the transfer function of the linear
dynamics of system (13) is obtained as follows:

G(s) = cT (s I − A)−1 b + ρ/s (14)

Definition 2.1 ϕ ∈ F(μ1 , μ2 ), there exists

ϕ(0) = 0 (15)
μ1 y 2 < yϕ(y) < μ2 y 2 , ∀y ∈ R, y = 0 (16)

Theorem 2.1 fal ∈ F(μ1 , μ2 ), that is

μ1 y 2 < y · fal(y) < μ2 y 2 , ∀y ∈ R, y = 0 (17)

where μ1 = 0, μ2 = δ α−1 + τ, ∀τ > 0, α < 1.

Proof As fal(0) = 0, α < 1, we have

fal(y) − fal(0)
lim y α−1 ≤ ≤ δ α−1 (18)
y→∞ y−0

then we obtain

fal(y)
0< < δ α−1 + τ, ∀τ > 0, y = 0 (19)
y

and then

0 < fal(y)y < (δ α−1 + τ )y 2 (20)

which means μ1 y 2 < y · fal(y) < μ2 y 2 (μ1 = 0, μ2 = δ α−1 + τ, ∀τ > 0, y = 0).


Definition 2.2 System (14) is said to be absolutely stable: If ∀ϕ ∈ F(μ1 , μ2 ), the zero
solutions of the system are globally consistently asymptotically stable; System (14) is
Circuits Syst Signal Process

said to be absolutely stable with a finite domain: If ∀ϕ ∈ F(μ1 , μ2 ), the zero solutions
of the system are consistently locally asymptotically stable.

3 Robust Stability Analysis of the Perturbed Indirect Lurie System

The strong robustness of ADRC has been proved by many theoretical analyses and
practical applications, but quantitative characterization of its robust stability has not
been effectively solved yet. To quantitatively analyze the robust stability, we propose
the following theorem with the help of Popov criterion [17] and Lyapunov’s direct
method.
Theorem 3.1 If system (13) satisfies the following assumptions:
(A1) the pair (A, b) is controllable, and the pair (A, cT ) is observable;
(A2) the matrix A is Hurwitz;
(A3) ρ > 0 and ϕ ∈ F(μ1 , μ2 );
(A4) there exists a scalar r  > 0 such that

1
+ Re[(1 + jωr  )G( jω)] > 0 , ∀ω ∈ R (21)
μ2

(A5) Given a symmetric positive definite matrix W , there exists a scalar ε > 0,
γ ≥ 0, a vector q, symmetric positive definite matrices P and W0 , and a scalar
δ  > 0 satisfying

P A + AT P = −qq T − εW (22)

Pb − v = γ q (23)
εW = εW0 + δ  I (24)

(A6) the disturbance term E(x, u) satisfies

δ
E(x, u) 2 ≤ β x̃ 2 ≤ x̃ 2 (25)
2 P i2 + r  μ2 c 22

where P i2 denotes the spectral norm of the matrix P, i.e., [λmax (P ∗ P)]1/2 , · 2
represents 2-norm and β is positive real number.
Under the assumptions A1–A6, the point x̃ = 0 of the perturbed Lurie system is
uniformly asymptotically stable.
Remark 1 The first four assumptions A1–A4, which construct the Popov criterion,
guarantee the global asymptotical stability of the nominal system (i.e., without
consideration of the disturbance term E(x, u)); the assumption A5 is the famous
Kalman–Yakubovich lemma, which bridges the gap between the Lyapunov method
and frequency domain method; the assumption A6 and assumption A5 are used to
analyze the robust stability of the perturbed indirect Lurie system under the global
asymptotical stability of the nominal system.
Circuits Syst Signal Process

Proof Construct a positive definite Lyapunov function candidate, which possesses the
form of quadratic plus nonlinear integral, namely
 σ
V (x̃, σ ) = x̃ T P x̃ + α(σ − c T x̃)2 + r  ϕ(σ )dσ (26)
0

where P is a positive definite matrix to be chosen; α, r  > 0.

Suppose that for some θ > 0, the set

Ωc = {x̃ ∈ R 2n |V (x̃) ≤ θ } (27)

is a bounded set in M. Computing the derivative of (26) along the solution of (13), we
obtain
 
1  T T
V̇ (x̃, σ ) = x̃ (P A + A P)x̃ − 2 Pb − αρc − r A c x̃ϕ(σ )
T T
2
   
2αρ   T 1
− + r ρ + r c b) ϕ (σ ) − 2αρϕ(σ ) σ −
2
ϕ(σ )
μ2 μ2
+ 2 x̃ T P E(x, u) + r  cT E(x, u)ϕ(σ ) (28)

Let

2αρ
γ = + r  ρ + r  cT b (29)
μ2
1
v = αρc + r  AT c (30)
2

Evidently, 2αρϕ(σ )(σ − μ2 ϕ(σ ))


1
≥ 0. Suppose α, r  are chosen such that γ ≥ 0.
Let

V̇  (x̃, σ ) = x̃ T (P A + AT P)x̃ − 2(Pb − v)T x̃ϕ(σ )


−γ ϕ 2 (σ ) + 2 x̃ T P E(x, u) + r  cT E(x, u)ϕ(σ ) (31)

then

V̇ (x̃, σ ) ≤ V̇  (x̃, σ ) (32)

It is obvious that only if V̇  (x, σ ) is negative definite, V̇ (x, σ ) is negative definite.


According to the assumption A5, we have

V̇  (x̃, σ ) ≤ x̃ T (P A + AT P)x̃ − 2(Pb − v)T x̃ϕ(σ ) − γ ϕ 2 (σ )


+2 x̃ 2 P i2 E(x, u) 2 + r  μ2 c 22 x̃ 2 E(x, u) 2
≤ x̃ T (P A + AT P)x̃ − 2(Pb − v)T x̃ϕ(σ ) − γ ϕ 2 (σ )
Circuits Syst Signal Process

 
+x̃ T 2β P i2 + r  μ2 β c 22 x̃
≤ x̃ T (P A + AT P + δ  I )x̃ − 2(Pb − v)T x̃ϕ(σ ) − γ ϕ 2 (σ ) (33)

According to the assumption A5, it follows that


√ 2
V̇  (x̃, σ ) ≤ −x̃ T (εW − δ  I )x̃ − (x̃ T q + γ φ(σ ))

≤ −ε x̃ T W0 x̃ − (x̃ T q + γ φ(σ ))2 < 0 (34)

Therefore, combining inequality (32) with inequality (34), we have

V̇ (x̃, σ ) ≤ V̇  (x̃, σ ) < 0 (35)

Thus, the point x̃ = 0 is uniformly asymptotically stable in Ωc .


From system (13), if A, b, cT , ρ and μ2 are known, then the procedure of deriving
the robust stability bound on allowable E(x, u) is as follows:
Step 1: To determine the absolute stability of the nominal system by the Popov
criterion. If Popov criterion is satisfied, then find an appropriate r and α. α =
1/(2ρ) may be one solution.
Step 2: Evaluate v and γ from Eqs. (29) and (30).
Step 3: Select a symmetric definite matrix W and a positive real number ε, then P
can be obtained by solving the following algebraic Riccati equation:

P Ar + ArT P − P Rr P + Q r = 0 (36)

where Ar = A − γ1 bv T , Q r = εW + vvγ , Rr = − bbγ . If P is not positive definite


T T

solution, then select another W and ε until P is positive definite.


Step 4: Select a positive definite matrix W0 such that εW = εW 0 + δ  I , and then
obtain an appropriate δ  .
Step 5: According to inequality (25), the robustness measurement β can be
obtained.
Some remarks about the method are listed as follows:

Remark 2 For a stable Lurie system with a linear perturbation (i.e., g(x, u) is linear),
it is globally stable.

Remark 3 If the nominal system of system (13) is absolutely stable and the inequality
2 x̃ T P E(x, u) + r  cT E(x, u)ϕ(σ ) < 0 is definite, then the perturbed system (13) is
globally stable.

Corollary 1 The above method can be generalized to the robust stability analysis of
the linear ADRC-based control system.

Proof The linear ADRC adopts a linear ESO, i.e., φ(σ ) = σ . It is clear that σ ∈
F(0, 1) and the method is still effective.

Circuits Syst Signal Process

Fig. 3 Mass–damper–spring system

4 Simulation

Here we present a mass–damper–spring system as an example to show the application


of our proposed method. The schematic description of the controlled system is shown
in Fig. 3, in which the spring is a soft spring and satisfies k(x1 ) = 3 − x12 and m = 5,
c = 1. The dynamic equations of the nonlinear system are given as follows:

⎨ ẋ1 = x2
ẋ = −3x1 − 5x2 + x13 + u
⎩ 2
y = x1
(37)

where x1 and x2 represent the position and the velocity, respectively. x13 in Eqs. (37)
is treated as total disturbance here.
Our purpose is to design a stable ADRC-based control system, which can be divided
into three steps. Firstly, we guarantee stability of the control system for the nominal
linear plant. Secondly, we find the robustness bound on allowable g(x, u). Thirdly,
since the total disturbance is certain, we try to estimate the region of attraction.
Step one, design a stable ADRC-based control system for the nominal linear plant.
Since the plant is second order, we adopt a third-order ESO as follows:


⎪ e = z1 − y

ż 1 = z 2 −β1 · fal(e)

⎪ ż 2 = z 3 −β2 · fal(e) − 3z 1 − 5z 2 + u

ż 3 = −β3 · fal(e)
(38)

Some of the parameters of ADRC depend on two variables, which are ωo = 20 and
ωc = 10. Then, the parameters of ADRC are achieved as β1 = 3ω0 , β2 = 3ωo2 ,
β3 = ω03 , k p = ωc2 and kd = 2ωc . The two parameters of fal(e) are δ = 0.01,
α = 0.25.
Circuits Syst Signal Process

400

300

200

100
slope r’=0.4
wImG

−100

−200

−300

−400
−400 −300 −200 −100 0 100
ReG

Fig. 4 Popov plot of G( jω)

According to (13) and (38), we have


⎧⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤
⎪ ẏ 1 −k1 − a2
0 −k2 − a1 y1 0 1
⎪ ⎢ ẏ1 ⎥ ⎢ a
⎪ a1 −a1 (k1 + a2 ) −a1 (k2 + a1 ) ⎥ ⎢ y2 ⎥ ⎢ −β3 ⎥  ⎢
⎥ ⎢ ⎥ ⎢ ⎥ a1 ⎥

⎪ ⎢ 2 ⎥ ⎢ 2 ⎢ ⎥

⎪ ⎣ ż 1 ⎦ = ⎣ 0
0 0 1 ⎦ ⎣ z 1 ⎦ + ⎣ β1 ⎦ u + ⎣ 0 ⎦ g(x, u)



⎨ ż 2  0
0 −k1 −k2 z2 β2 0
ξ̇ = u ⎡ ⎤

⎪ 
y
 ⎢ y1 ⎥

⎪ −β

⎪ σ = a −a 1 0 ⎣ 2 ⎥
a1 1 ⎢
z1 ⎦
+ a 3ξ

⎪ 2 2 2


⎩  z2
u = −ϕ(σ )
(39)

Substituting the values of the corresponding parameters into (39), and according
to (14), the transfer function of the linear dynamics is obtained as follows:

60 × (s + 10)2 (s + 17.29)(s + 7.713)


G(s) = (40)
s(s + 10)2 (s + 4.303)(s + 0.6972)

The stability analysis for the nominal plant by the Popov criterion is shown in
Fig. 4. The vertical axis uses ωIm[G( jω)] instead of Im[G( jω)], which is different
from normal Nyquist plot. To distinguish from the Nyquist plot, Fig. 4 is called Popov
plot. As both of Re(G( jω)) and ωIm(G( jω)) are even functions of the independent
variable ω, we only consider ω ∈ [0, ∞). G( jω) has no eigenvalues on the imaginary
axis except for one eigenvalue at the origin. From the Popov plot of G( jω) shown in
Fig. 4, we find that the condition (21) is satisfied if the plot lies to the right of any line
with slope r  ≤ 0.4 and intercepting the point −1μ2 + j0(1/μ2 = 0.032).
Step two, find the robustness bound on allowable g(x, u).
Circuits Syst Signal Process

Since the nominal system is stable according to the Popov criterion, then we can
further obtain the robustness bound on allowable g(x, u) according to Step 1 ∼ 5 in
Sect. 3. Here, we select r  = 0.1, then v = [0.783, 0.167, 0.500, 0.050]T , γ = 6.03.
Let
⎡ ⎤
200 0 0 0
⎢ 0 200 0 0 ⎥
εW = ⎢ ⎣ 0 0 200 0 ⎦

0 0 0 200

Solving Eq. (36), we obtain


⎡ ⎤
43 −17 −504 −91
⎢ −17 14 383 74 ⎥
P=⎢
⎣ −504

383 10990 2001 ⎦
−91 74 2001 392

P is verified to be positive definite. Then, let δ  = 200. From inequality (25), we


obtain β = 0.0088.
Step three, estimate the region of attraction.
Based on the robustness bound on allowable g(x, u) and the Lyapunov function, it
is possible to estimate the domain of attraction. According to the robustness bound,
we have
    2  

M = [ x̃1 x̃2 x̃3 x̃4 ]T  1 + a12 x13 ≤ 0.0088 x̃12 + x̃22 + x̃32 + x̃42 (41)

where


⎪ x̃1 = x2

x̃2 = −3x1 − 5 x2 − z 3
(42)

⎪ x̃3 = z1

x̃4 = z2

then
 σ
V (x̃, σ ) = x̃ T P x̃ + α(σ − c T x̃)2 + r  ϕ(σ )dσ
0
 σ
ρ
= x̃ T P x̃ + ξ 2 + r  ϕ(σ )dσ (43)
2 0

Substituting Eqs. (42) and σ = z 1 − x1 , ξ = z 3 /β3 into Eq. (43), it follows that

V (X, Z , z 3 ) = 2(z 1 − x1 )1.25 25 + 46x1 x2 + 490x2
−34x1 z 2 + 6.2x1 z 3 − 1876z 1 − 335z 2

+64z 3 + 18.6x12 − 137x1 z 1 + z 32 48000 (44)
Circuits Syst Signal Process

Fig. 5 Estimation of the attraction of region

We estimate the region of attraction by


 
Ωc = x ∈ R 5 |V (X, Z , z 3 ) ≤ c (45)

where c ≤  min  V (X, Z , z 3 ) = 4.35 to ensure that Ωc is


(1 + a12 )(x13 )2 =0.0088(x̃12 + x̃22 + x̃32 + x̃42 )
 
contained in the set (1 + a12 )(x13 )2 ≤ 0.0088(x̃12 + x̃22 + x̃32 + x̃42 ) . Taking c = 4.35
and calculating by (44) and (45), we obtain the estimation of the region of attraction
shown in Fig. 5.
Remark 4 The region of attraction is not uniquely determined, because the Lyaponov
function is not unique. Furthermore, the region of attraction is different even by the
same Lyaponov function using different methods and is conservative more or less.
In addition, since there are five parameters in the Lyapunov function (44), it is not
convenient to draw the whole region in a multidimensional space.
Remark 5 Note that the performance of the controller that uses the modeled linear
dynamics is superior to the one that does not use in general, because the burden of
ESO is lightened.
Remark 6 The proposed method demands that the matrix A is Hurwitz, which may
not be guaranteed in some cases. Under the premise of good control performance, one
way to solve this problem is to appropriately adjust the parameters of the controller
or make use of another linear dynamics model instead of the real modeled linear
dynamics. For instance, if the model of the plant in this paper is modified as follows:

⎨ ẋ1 = x2
ẋ = −3x1 − x2 + x13 + u (46)
⎩ 2
y = x1
Circuits Syst Signal Process

Under the same parameters, the matrix A is not Hurwitz, which lead to the proposed
method is useless. However, if we still make use of the linear dynamics −3x1 − 5x 2
instead of the real linear dynamics −3x1 − x2 in the design, the nominal linear system
will be globally absolutely stable and the performance is still satisfactory.

Remark 7 This paper considers the robust stability of nonlinear ADRC-based control
system. It is worthy to investigate whether the results can be extended to the other
area, such as sliding mode control [13] and robust control [14].

5 Conclusions

This paper proposed a Popov–Lyapunov method to deal with the robustness stability
analysis and design of the ADRC-based control system. The method is applicable to
stability or robustness analysis for both the linear and nonlinear ADRC-based SISO
control systems, and any system that can be transformed into the perturbed indirect
Lurie system described by Eq. (13). In the design of controller, the modeled linear
dynamics can be easily integrated into ADRC, which improves both the performance
and stability characteristics. The Popov criterion is applied for the analysis of the global
stability of the nominal system, while the allowable bound of total disturbance can help
us to estimate the robust stability of an ADRC-based control system. Furthermore, if
the total disturbance of the system is known, the region of attraction can be estimated
by the proposed method. In a word, the proposed method further shortens the distance
between ADRC and practical applications.
It is very useful to transform the ADRC-based control system into an indirect Lurie
system, and many existing methods can be used. Future work will consider the ESO
that uses different nonlinear functions or parameters, which may make the ADRC
more flexible and efficient. Moreover, the total disturbance in this paper is assumed to
be vanishing, while the robust boundedness for nonvanishing total disturbance should
be further investigated.

Acknowledgments The authors would like to thank the referees for their valuable and helpful comments
which have improved the presentation. The work was supported by the open funding program of Joint
Laboratory of Flight Vehicle Ocean-Based Measurement and Control under Grant No. FOM2015OF011,
the National Basic Research Program of China (973 Program) under Grant No. 2012CB720000, the National
Natural Science Foundation of China under Grant No. 61225015 and Grant No. 61105092 and Foundation
for Innovative Research Groups of the National Natural Science Foundation of China under Grant No.
61321002.

References
1. K. Erenturk, Fractional-order and active disturbance rejection control of nonlinear two-mass drive
system. IEEE Trans. Ind. Electron. 60(9), 3806–3813 (2013)
2. Z.Q. Gao, S.H. Hu, F.J. Jiang, A novel motion control design approach based on active disturbance
rejection, in Proceedings of the 40th IEEE Conference on Decision and Control (2001), pp. 4877–4882
3. B. Guo, Z. Zhao, On convergence of tracking differentiator. Int. J. Control. 84(4), 693–701 (2011)
4. B. Guo, Z. Zhao, On finite-time stable tracking differentiator without Lipschitz continuity of Lyapunov
function, in 30th Chinese Control Conference (2011), pp. 354–358
Circuits Syst Signal Process

5. T. Gang, Z.Q. Gao, Frequency response analysis of active disturbance rejection based control system.
in IEEE International Conference on Control Applications (2007), pp. 1595–1599
6. B. Guo, Z.L. Zhao, On convergence of nonlinear active disturbance rejection control for SISO systems,
in 24th Chinese Control and Decision Conference (2012), pp. 3507–3512
7. B.Z. Guo, Z.L. Zhao, On convergence of the nonlinear active disturbance rejection control for MIMO
systems. SIAM J. Control Optim. 52(2), 1727–1757 (2013)
8. Z.Q. Gao, Scaling and bandwidth-parameterization based controller tuning, in Proceedings of the
American Control Conference (2006), pp. 4989–4996
9. J.Q. Han, Active disturbance rejection control technique-the technique for estimating and compensating
the uncertainties (National Defense Industry Press, Beijing, 2008)
10. J.Q. Han, From PID to active disturbance rejection control. IEEE Trans. Ind. Electron. 56(3), 900–906
(2009)
11. Y. Huang, A new synthesis method for uncertain systems the self-stable region approach. Int. J. Syst.
Sci. 30(1), 33–38 (1999)
12. Y. Huang, J.Q. Han, Analysis and design for the second order nonlinear continuous extended states
observer. Chin. Sci Bull. 45(21), 1938–1944 (2000)
13. F. Li, L. Wu, P. Shi, State estimation and sliding mode control for semi-Markovian jump systems with
mismatched uncertainties. Automatica. 51, 385–393 (2015)
14. P. Shi, Y. Yin, F. Liu, Robust control on saturated Markov jump systems with missing information. Inf.
Sci. 265, 123–138 (2014)
15. M.W. Sun, Z.Q. Chen, Z.Z. Yuan, A practical solution to some problems in flight control, in Proceedings
of the 48th IEEE Conference on Decision and Control and 28th Chinese Control Conference (2009),
pp. 1482–1487
16. H. Sira-Ramirez, J. Linares-Flores, C. Garcia-Rodriguez et al., On the control of the permanent mag-
net synchronous motor: an active disturbance rejection control approach. IEEE Trans. Control Syst.
Technol. 99, 2056–2063 (2014)
17. M. Vidyasagar, Nonlinear systems analysis (Siam, Philadelphia, 2002)
18. D. Wu, K. Chen, Design and analysis of precision active disturbance rejection control for noncircular
turning process. IEEE Trans. Ind. Electron. 56(7), 2746–2753 (2009)
19. X. Wang, Z. Chen, G. Yang, Finite-time-convergent differentiator based on singular perturbation tech-
nique. IEEE Trans. Autom. Control. 52(9), 1731–1737 (2007)
20. D. Wu, K. Chen, Frequency-domain analysis of nonlinear active disturbance rejection control via the
describing function method. IEEE Trans. Ind. Electron. 60(9), 3906–3914 (2013)
21. D. Wu, K. Chen, Limit cycle analysis of active disturbance rejection control system with two nonlin-
earities. ISA Trans. 53(4), 947–954 (2014)
22. H.L. Xing, X.Z. Zhong, J. Li, Linear extended state observer based back-stepping control for uncertain
SISO nonlinear systems. Int. J. Innov. Comput. Inf. Control. 11(4), 1411–1419 (2015)
23. W.C. Xue, Y. Huang, Comparison of the DOB based control, a special kind of PID control and ADRC,
in American Control Conference (2011), pp. 4373–4379
24. W.C. Xue, Y. Huang, The active disturbance rejection control for a class of MIMO block lower-
triangular system, in 30th Chinese Control Conference (2011), pp. 6362–6367
25. W.C. Xue, Y. Huang, On frequency-domain analysis of ADRC for uncertain system, in American
Control Conference (2013), pp. 6637–6642
26. D. Yuan, X.J. Ma, Q. Zeng, Research on frequency-band characteristics and parameters configuration
of linear active disturbance rejection control for second-order systems. Control Theo. Appl. 30(12),
1630–1640 (2013)
27. Q. Zheng, L.Q. Gao, Z.Q. Gao, On stability analysis of active disturbance rejection control for nonlinear
time-varying plants with unknown dynamics, in Proceedings of the 46th IEEE Conference on Decision
and Control (2007), pp. 12–14

You might also like