Received June 18, 2019, accepted June 28, 2019, date of publication July 10, 2019, date of current

version July 31, 2019.

Digital Object Identifier 10.1109/ACCESS.2019.2928050

A New Mechanical Resonance Suppression

Method for Large Optical Telescope by Using
Nonlinear Active Disturbance Rejection Control
XIN LI 1,2 , WENLIN ZHOU1 , JUN LUO1 , JUNZHANG QIAN1 , WENLI MA1 ,
PING JIANG1 , AND YONGKUN FAN3
1 Institute
of Optics and Electronics, Chinese Academy of Sciences, Chengdu 610209, China
2 Universityof Chinese Academy of Sciences, Beijing 100049, China
3 Sichuan Vocational and Technical College of Communications, Chengdu 611130, China

Corresponding author: Wenli Ma ([email protected])

This work was supported in part by the National Natural Science Foundation (NSFC) of China under Project 60978050, and in part by the
Graduate Student Innovational Foundation of the Institute of Optics and Electronics, University of Chinese Academy of Sciences, under
Project C12K011.

ABSTRACT Aiming at solving the problem of the multi-low-frequency mechanical resonances appearing
in the large optical telescope control system, this paper proposes a novel control method based on nonlinear
active disturbance rejection control (NADRC) and proportional–integral (PI) control. In the proposed control
framework, a nonlinear tracking differentiator (NTD)-based feedforward control is designed to improve
the tracking performance of the system. Then, the principle of suppression of mechanical resonance of
this method is analyzed. Compared with the most commonly used acceleration feedback control (AFC)
method, the theoretical analysis shows that the proposed method is more effective for suppressing the
low-frequency mechanical resonance. Finally, the proposed method is applied to a large optical telescope,
and the experimental results show that the proposed method is better than AFC.

INDEX TERMS Mechanical resonance suppression, active disturbance rejection control (ADRC), acceler-
ation feedback control (AFC), nonlinear tracking differentiator (NTD), optical telescope.

I. INTRODUCTION the mechanical resonance of large telescope has the following

The large optical telescope is a kind of high precision observ- characteristics: 1) The resonant frequency is very low, gener-
ing equipment, and a high precision motion control system is ally lower than 50 Hz, sometimes even only a few Hz [6].
essential. However, the tracking precision of the control sys- 2) The inertia ratio of the load and motor is very large [7].
tem is affected by various known and unknown disturbances, 3) The damping ratio is small which means that the magnitude
such as mechanical resonance [1], friction torque [2], of the Bode diagram of the plant increases to very high levels
wind disturbance torque [3], [4], dead zone [1], un-modeled at the nature frequency [3]. 4) There are usually multi low
dynamics, etc. Among these disadvantages, the mechanical frequency resonances in the system [9]. 5) The system con-
resonance which limits the bandwidth of speed has a serious tains a variety of nonlinearities: friction, backlash, rate, and
impact, and the problem of mechanical resonance in the large acceleration limits [7]. For the above reasons, the commonly
telescope control system is a long-standing problem. used methods for suppressing resonance in large telescopes
In [5], the low frequency mechanical resonance sup- include: limiting the speed loop bandwidth, AFC [6], Quan-
pression methods that commonly used in industrial are titative Feedback Theory (QFT) [1], and linear-quadratic-
explored, including low-pass filter method, notch filter Gaussian (LQG) [2], [3]. Among them, AFC is one of the
method, Bi-quad filter method, etc. Unfortunately, these sup- most widely used methods, and it is used in many large
pression methods are not suitable for large telescope. Because telescopes to suppress mechanical resonance [6], [8]–[13].
In recent years, ADRC is used to suppress the mechan-
The associate editor coordinating the review of this manuscript and ical resonance, and it takes the mechanical resonance as a
approving it for publication was Lei Wang. disturbance and compensates the disturbance by extended

94400 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/ VOLUME 7, 2019
X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

state observer (ESO). In [14], [15], the performance of

ADRC is better than AFC through extensive simulation
and experimental results, however, the theoretical analy-
sis was not conducted. In [16], [17], Model-based ADRC
and non-model-based ADRC is used to control a two-mass
benchmark problem, further demonstrating that ADRC has
a good suppression performance for mechanical resonance.
Reference [18] analyzes and proves that ADRC can com-
pensate for mismatched disturbances, and give the princi-
ple of mechanical resonance suppression by using ADRC
from the perspective of time domain. However, the phys-
ical meaning of how the ADRC suppresses mechanical
resonance is not clear enough in time domain. Mean-
while, the linear extended state observer (LESO) of the
linear active disturbance rejection control (LADRC) brings
phase loss or sacrifices the nonlinear performance when
the bandwidth of LESO is limited [19]–[21]. The nonlin-
ear extended state observer (NESO) of NADRC has high
observe efficiency [22], which reduces the phase loss, but is
not widely used.
Considering the above problems, the following researches
are carried out in this paper: 1) A composite control frame
which includes NADRC (speed loop), PI (position loop) and
NTD (speed and acceleration feedforward) is introduced to
solve the problem of multi low frequency mechanical res-
onances in large telescope. 2) The mechanism of ADRC
restraining mechanical resonance is studied from frequency
domain perspective. 3) By comparing with AFC method,
the advantages of suppressing low frequency mechanical res-
onance by using ADRC are explored. 4) The experimental
results carried on a large optical telescope are presented
to verify the correctness of the theoretical analysis and the
FIGURE 1. The 3-D sketch of a large optical telescope.
effectiveness of the proposed method.
The rest of this paper is organized as follows. In section II,
the model of the large telescope is analyzed, and a simpli-
fied model is deduced. In section III, the composite control and design [1], [23].The compliant couple of the motor and
method is proposed, the stability of the proposed method is the load in the telescope can be modeled as a two-inertia
proved, the resonance suppression mechanism of the method model as shown in Fig.3, and the feedback is a motor-side
is discussed, and then the advantages of this method are feedback [24].The inertia of the motor is JM , the inertia of
illustrated by comparing with AFC. In section IV, experimen- the load is JL , the motor is connected to the load by a
tal results are shown. In section V, concluding remarks are shaft, the stiffness coefficient is Ks , the damping coefficient
included. is bs , the electromagnetic torque of the motor is TM , the load
torque produced by the friction and wind disturbance is TL
II. MATHEMATICAL MODEL OF A LARGE TELESCOPE and in most cases TL is small, ωM and θM represent the speed
Figure 1 shows the three-dimensional (3-D) sketch of a and angle of the motor, ωL and θL represent the speed and
large optical telescope. Usually, the elastic deformation of angle of the load. The kinematics equations of the system
the transmission mechanism cannot be neglected due to the shown in Fig.3 can be expressed as
excessive load, and this deformation will lead to devia-
tion and lag in the transmission process, so the load can- JM ω̇M = TM − bs (ωM − ωL ) − Ks (θM − θL )


not be regarded as an ideal rigid body. Fig. 2 shows the

JL ω̇L = bs (ωM − ωL ) + Ks (θM − θL ) − TL


azimuth axis frequency response data (FRD) model of a (1)
large optical telescope, which has a low frequency res-


 ωM = θ̇M
ωL = θ̇L

onance at 31Hz and 40Hz respectively. The FRD model
varies according to the azimuth and elevation angles, friction,
speed, and acceleration [1]. These nonlinearities and param- Then, the block diagram of the two-inertia model can be
eter uncertainties increase the difficulty of system analyzing obtained as Fig.4. By ignoring TL , the transfer function from

VOLUME 7, 2019 94401

 X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

ignored.
TM kT
Gacr (s) = =
2 (3)
Iref Tf s + 1
Substitute (3) into (2), the transfer function from Iref to ωM
can be induced
GPlant (s)
ωM
= = GModel · GASR
Iref
1 kT 1 JL s2 + bs s + Ks
=
2 · · · (4)
Tf s + 1 (JM + JL ) s JM JL
(JM +JL ) s
2 + bs s + Ks
Let
FIGURE 2. Measured telescope FRD model (from reference current Iref
to motor speed ωM ). 1 1
G1 (s) = · (5)
s Tf s + 1 2

kT JL s2 + bs s + Ks
G2 (s) = · (6)
(JM + JL ) JM JL
(JM +JL ) s
2 + bs s + Ks
G1 (s) is an ideal rigid body, whereas G2 (s) is a com-
pliance of the transmission component, which corrupts the
ideal plant. The anti-resonant frequency (ωA ) and resonant
FIGURE 3. Two-inertia model.
frequency (ωN ) of the system can be calculated as
 s
 Ks
 ωA = J



s L (7)
 Ks Ks
 ωN =


 +
JL JM
When in the low frequency band, s → 0, so G2 (s) ≈
kT (JM + JL ), (4) can be simplified to (8), in this case the


system is equivalent to a rigid body.

1 kT
G0Plant (s) = l1 (s) = · (8)
s (JL + JM )
While in the high frequency band, s → ∞, the dominator
is s2 , so G2 (s) ≈ kT JM (4) can be simplified to (9), which


is equivalent to the motor working without any load.

kT 1
G00Plant (s) = l2 (s) = · (9)
JM · s Tf s + 1 2

FIGURE 4. Block diagram of two-inertia model.

However, the specific values of JL and KS are very difficult
to get in the real practice. The usual method is to simplify
the FRD model and identify the parameters. Table 1 shows
TM to ωM can be described as the fitted parameters of the two-inertia model. It is important
to stress that these fitted parameters are not realistic and of
ωM 1 1 JL s2 + bs s + Ks
GModel (s) = = · · no physical significance, because of the simplification in the
TM (JM + J L) s
JM JL
(JM +JL ) s
2 + bs s + Ks process.
(2) It can be seen from Fig. 5 that the fitted transfer function
GPlant (s) perfectly fits the real tested FRD model. Due to the
existence of mechanical resonance, the amplitude-frequency
Usually, the motor current close loop can be regarded as characteristic of the system in the high frequency band is
(3) [24], kT represents the torque coefficient of the motor, Tf raised, thereby reducing the gain margin of the control sys-
is the time constant of the current close loop, Iref is current tem, and thus the bandwidth of the control system cannot be
reference, and in most cases, Tf is very small and can be improved.

94402 VOLUME 7, 2019

X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

TABLE 1. Identified parameters for two-inertia model.

FIGURE 6. Block diagram of proposed method for telescope control

system.

kT 1 JL s2 + bs s + Ks
= · · (10)
(JM + JL ) s2 JM JL 2
(JM +JL ) s + bs s + Ks
Considering the external disturbance w, (10) can be rewrit-
ten as
.... ...
JM JL θ M + bs (JM + JL ) θ M + Ks θ̈M


= kT JL Ïref + bs İref + Ks Iref + w (11)
FIGURE 5. Bode diagram of two-inertia model and tested model. Integrating (11) twice on both sides, the fourth-order sys-
tem with a relative degree of two becomes a second-order
system as below
III. ACTIVE DISTURBANCE REJECTION CONTROL FOR
θ̈M = b0 · Iref + f (12)
TWO-INERTIA MODEL
From the above discussion, we can see that if G2 (s) can be where
compensated, GPlant (s) becomes an ideal rigid plant which kT
is easy to control. ADRC is that kind of method that can b0 = (13)
JM Z
treat G2 (s) as disturbance, which can be estimated and com- kT bs kT Ks
ZZ
bs (JM + JL )
pensated in real time [25]. Based on this theory, this section f = Iref + Iref − θM
JM JL JZM JL JM JL
firstly presents the methodology of the framework and pro-
Ks (JM + JL )
ZZ
posed controller, and then the stability of the closed loop con- − θM + w (14)
trol system is proved. Secondly, based on frequency domain JM JL
analysis, the principles of suppressing mechanical resonance Here f , including both external disturbance and the res-
of LADRC and AFC are discussed. Finally, the comparison onance, represents the ‘‘ total disturbance’’ to be estimated
between ADRC and AFC is theoretically analyzed, and the and compensated. Based on (12), it is better to choose a
advantages and disadvantages of ADRC are discussed. third-order ESO [22]. With x1 and x2 denoting θM and ωM ,
and x3 denoting f , (12) can be rewritten as
A. PROPOSED METHOD FOR TELESCOPE
        
ẋ1 0 1 0 x1 0 0
The control method proposed in this paper is shown in Fig. 6.  ẋ2  =  0 0 1   x2  +  b0  u +  0  f˙
This scheme consists of three elements: the position loop con- ẋ3 0 0 0  x 3 0 1
troller, the speed loop controller, and the NTD feedforward. (15)
  x1
The position loop controller is traditional PI controller. The y = 1 0 0  x2 
speed loop controller is NADRC which consists of NESO and x3
a proportional feedback control law. The NTD feedforward
So the corresponding NADRC control law is constructed
controller consists of an acceleration feedforward controller
as
and a speed feedforward controller.
ε1 = z1 − θM



 ż1 = z2 − β01 · ε1 

1) SPEED LOOP CONTROLLER DESIGN 

 
The principle of the speed loop NADRC is shown in Fig. 6. 1
ESO : ż = z − β · fal ε , , δ + b̂0 Iref
Ignoring the Tf in (4), the transfer function from Iref to θM is 


2 3 02

1
2
 1
 ż3 = −β03 · fal ε1 , , δ


θM 1 4
Gvplant (s) = = Gplant (s) ·
Iref s (16)

VOLUME 7, 2019 94403

 X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

(ε1
|ε1 | 6 δ
fal (ε1 , a, δ) = δ (1−a) (17)
sign (ε1 ) |ε1 |a |ε1 | > δ
u0 = kp ωref − z2


(18)
z3
u = u0 − (19)
b̂0
According to [26], the anti-windup of NADRC is designed
as

 Imin
 u < Imin
Iref = u Imin < u < Imax (20)
u > Imax

Imax


where θref is the reference position, ωref is the reference

speed, u0 is the virtual control signal, u is the compensated
control signal, Iref is the actual control signal, ε1 is the
FIGURE 7. Bode diagram of NTD and du/dt + LPF (h = 0.00001, h0 = 10h,
observer error, β01 , β02 and β03 are observer gains, b̂0 is the TD = 0.0005, r = 0.8772/TD2 ).
estimated value of b0 , δ is the parameter to be tuned, kp is
the controller gain. z1 and z2 are observer states tracking θM
and ωM respectively, here z3 is the extended state which is the In this paper, NTD is used to solve these problems [27],
estimation of f in real time. Assuming that the controller is as shown in (25)(26).
unsaturated, so
fh = fhan (x̃1 (k) − ν (k) , x̃2 (k) , r, h0 )

z3 
Iref = u = u0 − (21) 

b̂0 x̃1 (k + 1) = x̃1 (k) + hx2 (k) (26)

Substituting (21) into (12), 

x̃2 (k + 1) = x̃2 (k) + h · fh
z3
θ̈M = b0 · u0 − b0 · +f (22)

b̂0 

 d = rh0


With b̂0 ≈ b0 and z3 ≈ f , we have



 d0 = h0 d

y1 = (x̃1 (k) − ν (k)) + h0 x̃2 (k)


θ̈M ≈ b0 · u0
 


(23) 
ω̇M ≈ b0 · u0

 q
a = d 2 + 8r |y|

 0

It can be seen from (23) that the total disturbance is ‘‘can- 
(a0 − d) (27)
celled’’ and the plant of speed becomes an ideal rigid plant,

 
 x̃2 (k) + sign (y) , |y 1 | > d 0

 a1 = 2
so the ideal closed-loop control of speed can be achieved by
 y
(k) ,

x̃ + |y1 | 6 d0
  2

 
a simple proportional controller in (18). 

  h0
 rsign (a1 ) , |a1 | > d




 fhan = −

2) POSITIOIN LOOP CONTROLLER DESIGN
 r a1 ,


 |a1 | 6 d
The position loop controller in Fig. 6 is a traditional PI d
controller
where ν (k) is the reference input, x̃1 (k) is the state tracking
1 ν (k), x̃2 (k) is the differential of x̃1 (k), h is the sampling
GPI (s) = kp0 + ki0 · (24)
s interval, h0 , d, d0 , y1 , a0 and a1 are the intermediate variables,
kp0 is the proportional gain, ki0 is integral coefficient. r the speed factor which determines the tracking speed to be
tuned.According to [28], the relationship between r and the
3) FEEDFORWARD CONTROLLER DESIGN bandwidth ωTD is
In order to improve the accuracy of the control system, √
ωTD = 1.14 r (28)
the feedforward control is used. The differential du/dt cal-
culation can be not realized in reality. So the widely used By sine sweep measurement method, the Bode diagram
feedforward method is shown in (25), the differential result is of (24) and x̃2 in (27) is obtained. As shown in, the three
filtered by a first-order or second-order low-pass filter, is the methods have the same bandwidth of 318 Hz. It can be seen
time constant of the low-pass filter, however these methods that the phase angle of NTD basically maintains about 900
are either too noisy or have large phase lag. ahead inside of bandwidth, whereas outside of bandwidth the
s s phase angle is rapidly down to −900 , which shows that NTD
GD (s) = or (25)
TD s + 1 (TD s + 1)2 is an ideal differentiator.

94404 VOLUME 7, 2019

X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

Therefore, speed and acceleration feedforward controllers constants λ2i (i = 1, 2, 3, 4), β2 , and positive continuous
can be designed as NTD: differentiable function V2 , W2 :Rn+1 → R such that

λ21 kvk2 6 V2 (v) 6 λ22 kvk2 , λ23 kvk2 6 W2 (v) 6 λ24 kvk2


 θref _f (k + 1)
= θref _f (k) + hωf (k)

(31)




ω f (k + 1)
 n−1
∂V2 (v) ∂V2 (v)

 X
+ ϕ (v1 , v2 , . . . , vn ) 6 −W2 (v)

vi+1

 = ω (k) + hfhan θ
ref _f (k) − θref (k) , ωf (k) , r, h0


f ∂vi ∂vn
i=1


 ωref _f (k + 1) (32)
= ωref _f (k) + haf (k) ∂V2 (v)


6 β2 kvk , ∀v = (v1 , v2 , . . . , vn ) ∈ Rn

(33)

f (k + 1)

a ∂vn




= af (k) + hfhan ωref _f (k) − ωref (k) , af (k) , r, h0



.
Let Kω = 1, Ka = 1 b̂0 , (30) can be rewritten as
(29)
 Z 
θref (k) is the position reference signal, θref _f (k) is the 1
−b̂0 kp k 0 i θM − θref dt


Iref =
state tracking θref (k), ωf (k) is feedforward speed, ωref _f (k) b̂0
is the state tracking ωf (k), af (k) is feedforward acceleration,


+ −b̂0 kp k 0 p θM − θref
Kω and Ka are parameters to be tuned.  
+ −b̂0 kp z2 − ωf + af − z3


(34)
4) THE FINAL CONTROLLER
Based on the above design, the control algorithm proposed We can see from (30) that the controller of proposed
in this paper can be divided into three parts: 1) The first method is linear, and the corresponding ϕ (·) in Assumption 3
part is NESO, as shown in (16). 2) The second part is the is
feedforward part, which consists of two NTDs in series, ϕ (x1 , x2 , x3 ) = −b̂0 kp ki0 x1 − b̂0 kp kp0 x2 − b̂0 kp x3 (35)
as shown in (29). 3) The third part is speed loop controller
and position loop PI controller. Ignoring the saturation effect, Apparently ϕ (x1 , x2 , x3 ) is Lipschitz continuous.
(18)(19)(21) (24) can be written as Let
 
ADRC : Iref = kp k 0 p θref − θM

0 1 0
Z A = 0 0 1  (36)
+ kp k 0 i θref − θM dt + kp Kω ωf


−b̂0 kp k 0 i −b̂0 kp k 0 p −b̂0 kp
z3 With appropriate parameters, A can be Hurwitz.
+ Ka af − kp z2 − (30)
b̂0 Let

5) STABILITY ANALYSIS V (x) = hPA x, xi (37)

The stability proof of NADRC is very difficult, and keeps W (x) = kxk2 = x12 + x22 + x32 (38)
un-proved for decades. Up until recent years the problem
was solved, the rigorous convergence proofs of the NESO Here ∀x = (x1 , x2 , x3 )T ∈ Rn , PA is the positive definite
and the NTD were conducted in [29], [30], and the sta- matrix that is the unique solution of the Lyapunov equation
bility of traditional NADRC in [22] for SISO system was PA A + AT PA = −In , In is a 3-dimensional identity matrix.
proved in [31]. Theorem 5 in [31] indicates that the system Then
is stable if Assumptions A1-A4 are satisfied. Theorem 5 and
λ21 kxk2 6 V (x) 6 λ23 kxk2 (39)
Assumptions A1-A4 are shown in the Appendix A. This n−1
paper mainly focuses on the practical application, so only the X ∂V   ∂V
xi+1 − b̂0 kp k 0 i x1 + b̂0 kp k 0 p x2 + b̂0 kp x3
improved parts in this paper will be proved and some existing ∂xi ∂xn
i=1
conclusions in the literatures will be used. = − hx, xi = −W (x) (40)
Assumption 1 is made for the boundedness of the system
itself and the external disturbance. For general practical sys- And
tems, Assumption 1 holds. The convergence of ESO and TD ∂V
used in this paper are proved in [29] and [30], so Assumption 6 β2 (PA ) kxk (41)
∂xn
2 and A4 hold. Therefore, here we only need to prove that (30)
satisfies Assumption 3, and Assumption 3 in [31] is described Therefore, V (x) and W (x) satisfy the conditions of
as follows Assumption 3.
Assumption 3 ϕ (v) is continuously differentiable, So the Assumptions A1-A4 are all satisfied, and the pro-
ϕ (0) = 0, and
Lipschitz continuous with Lipschitz constant posed method is stable. A numerical analysis is presented in
L: ϕ (v) − ϕ v̂ 6 L v − v̂ for all v, v̂ ∈ Rn .There exist the Appendix B.

VOLUME 7, 2019 94405

 X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

B. THE PRINCIPLE OF ADRC AND AFC SUPPRESSING ωo is the observer bandwidth. By Laplace transformation,
MECHNIACL RESONANCE the disturbance observed is
1) THE PRINCIPLE OF ADRC SUPPRESSING MECHNIACL −b0 ωo2 ωo2 s
RESONANCE Z̃2 (s) = · Iref (s) + · M (s) (45)
(s + ωo ) 2
(s + ωo )2
The ESO is the core of ADRC, which improves the con-
trol performance through observing and compensating the If Tf = 0, combining (4) and (45), using the Masonąŕs
internal and external disturbances. This section focuses on gain formula (MGF), the transfer function of the enhanced
the principle of ESO suppressing mechanical resonance. plant can be expressed as
Fig.8 shows the disturbance observing and compensating part ωM kT
GLESO (s) = =
ωo2
 
of the ESO. Since the NESO cannot be analyzed by transfer u0 JM + 1 − · JL
function, a linearized ESO is used here for analysis. And in (s+ωo )2
order to simplify the discussion, it is assumed that there is JL · s2 + bs · s + Ks 1
· ·
a speed sensor in the system, so a second-order ESO will be  JM  · JL · s2 + bs · s + K s s
ωo2
used. All parts f0 that corrupt the ideal integral component are JM + 1−
(s+ωo ) 2 ·JL
regarded as ‘‘disturbance’’. If the estimated disturbance z̃2 of (46)
ESO rapidly converges to f0 ,then the disturbance encountered
by the ideal plant can be well compensated. Let
JM
fLESO (s) = (47)
ωo2
 
JM + 1 − · JL
(s+ωo )2
Within the observer bandwidth of ωo , we can get
ωo2 (s + ωo )2 = 1, so fLESO (s) = 1, Then equation (46) can


be simplified to
kT 1
GLESO (s) = · (48)
JM s
GLESO (s) is an ideal plant. Therefore, as long as the
observer bandwidth of the ESO satisfies ωo > ωN ,the
mechanical resonance of the system can be well suppressed,
and the system becomes an ideal plant. Fig. 9 shows the Bode
diagram of GLESO (s) under different observer bandwidth.
It indicates that the ADRC method can effectively suppress
mechanical resonance. The larger the observer bandwidth is,
FIGURE 8. Block diagram LESO. the closer the plant is to l2 (s), and the better the suppress
effect.
According to the above assumptions, and ignoring Tf and
saturation in Fig. 8, (4) can be written as follows

ω̇M = b0 · Iref + f0 (42)

where
Z ZZ
kT bs kT Ks
f0 = Iref + Iref
JM JL JM JL
bs (JM + JL ) Ks (JM + JL )
Z ZZ
− ωM − ωM + w
JM JL JM JL
(43)

Then the corresponding second-order LESO is

FIGURE 9. Bode diagram enhanced plant (from u0 to ωM ),

 ε̃1 = z̃1 − ωM

the parameters are shown in the Table 1 and the observer bandwidth are
˙˜z1 = ˙˜z2 − β̃01 · ε̃1 + b̂0 Iref (44) ωo , 10ωo , 100ωo (ωo = 200).

˜z2 = −β̃02 · ε̃1

˙

We can see from Fig. 9 that the principle of ADRC sup-
Here, z̃1 and z̃2 are observer states tracking ωM and f0 , pressing the mechanical resonance is reducing the load inertia
β̃01 and β̃02 are observer gains. Using the parameterization as close as possible to zero by electronic means, thus reducing
T T
technique proposed in [32], let β̃01 β̃02 = 2ωo ωo2 ,
 
the coupling between the motor and the load. In this scenario,

94406 VOLUME 7, 2019

X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

C. COMPARISON OF THE TWO METHODS

According to the above analysis, we can see that the two
methods both want to enhance the original plant show in (4)
to an ideal pure integral component, so that the mechanical
resonance is suppressed. Here is a comparison of the advan-
tages and disadvantages of the two methods.
Given the same observer bandwidth ro = ωo > ωN ,
the Bode diagrams of fLESO (s) and fAFC (s) are shown
in Fig. 11. When ωN < ωcross , ωcross is the intersection point
of fLESO (s) and fAFC (s), fLESO (s) is closer to 1 than fAFC (s),
so GLESO (s) is closer to the ideal plant than GAFC (s), and the
larger the feedback gain, the better the effect as can be seen
in Fig. 12.

FIGURE 10. Block diagram of ideal acceleration feedback.

the motor can work as equivalent to no load in the low fre-

quency band, and effectively suppressing the mechanical res-
onance. Theoretically, if we want to construct the enhanced
plant to a pure integral component through ESO, the observer
bandwidth ωo needs to be infinite.

2) THE PRINCIPLE OF AFC SUPPRESSING

MECHNIACL RESONANCE
In order to demonstrate the resonance suppressing perfor- FIGURE 11. Bode diagram of fLESO (s) and fAFC (s), the parameters are
mance of ADRC, in this section the AFC method is simply shown in the Table 1 and ro = ωo = 2000, a = 10, 2a = 20.

discussed.
As shown in Fig. 10, assuming that the transfer 
function of
the ideal acceleration observer is Gas (s) = ro 2 (s + ro )2 ,
ro is the observer bandwidth of the acceleration observer,
feedback gain is a, and Tf ≈ 0. The estimated acceleration is
feed back to the current loop. Using the MGF, we can get
ωM kT
GAFC (s) = = 
Iref a·ro 2
1+ · JM + JL
(s+ro )2
JL · s2 + bs · s + Ks 1
×  2
 ×
1+ a·ro 2 ·JM s
(s+ro )
 2
 · JL · s2 + bs · s + Ks
1+ a·ro 2 ·JM +JL
(s+ro )
FIGURE 12. Bode diagram of GLESO (s) and GAFC (s), the parameters are
(49) shown in the Table 1 and ro = ωo = 2000, a = 10, 2a = 20.

Let
  As shown in Fig. 13, when ro = ωo = 2000 and
a·ro 2 load inertia JL increases from JL to 10JL , the magnitude of
1+ · JM
(s+ro )2 fAFC2 (s) decrease, thus the performance of AFC decreases.
fAFC (s) =   (50)
a·ro 2 Fig. 14 indicates that LESO still has very good effects. The
1+ · JM + JL
(s+ro )2
feedback gain a of AFC needs to be tuned larger to obtain an
The principle of AFC for suppressing the mechanical reso- effect similar to LESO. From the above discussion, we can
nance is raising the inertia of the motor by electronic means, conclude that for low-frequency mechanical resonance, ESO
effectively reduces the amplitude of the transfer function in is better; whereas for higher mechanical resonance, AFC is
the high frequency band, thereby suppressing the mechanical better.
resonance. Theoretically, if we want to change the plant to an As shown in Fig. 15, the ESO improves the open loop
ideal pure integral by AFC, the observer bandwidth ro needs gain at the low frequency and reduces the open loop gain
to be infinite and the value of a needs to be large enough. at the high frequency, thus a simple proportional controller

VOLUME 7, 2019 94407

 X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

FIGURE 15. Bode diagram of LESO and AFC (from Iref to ωM ), the
parameters are shown in the Table 1 ro = ωo = 2000, a = 10.
FIGURE 13. Bode diagram of fLESO (s) and fAFC (s), the parameters are
shown in the Table 1 and ro = ωo = 2000, a = 10.

FIGURE 16. Phase of GLESO at ωA when the bandwidth of the LESO

changing.
FIGURE 14. Bode diagram of GLESO (s) and GAFC (s), the parameters are
shown in the Table 1 and ro = ωo = 2000, a = 10.

and a large telescope. The measured FRD model of azimuth

axis of the telescope is shown in Fig. 2. The telescope con-
can satisfy the requirements. However, ESO also has its trol unit is an embedded computer with real-time operat-
disadvantage: ESO suppressing the resonant frequency at ing system. The servo controller can control and drive both
the cost of reducing the phase margin. Fig. 16 shows the azimuth and elevation axes at the same time. The MCU of
phase margin of the enhanced plant under different observer servo controller is TMS320F28335, and the IPM modules are
bandwidths. When ωo is less than 318 Hz, the phase angle of Mitsubishi PM200CLA120. The angular measuring system
the system becomes smaller as the ωo increases. When the ωo is consist of a Renishaw grating ruler and two read heads,
is greater than 318 Hz, the phase angle of the system increases and the grating ruler has 86400 pulses and the read head is
with the increase of ωo . In practical, the bandwidth of ESO 400 times subdivision. It can be calculated that one pulse rep-
should not be too low, because ESO with low bandwidth has resents 0.009375 arc-second (including a four-fold increase
limited effect; nor can it be too high, such that the noise will in resolution by the quadrature encoder pulse (QEP) mod-
be amplified. Therefore, ADRC is good at suppressing the ule). is the friction torque of the shaft. The current sampling
low frequency mechanical resonance. For higher frequency frequency is 8 kHz, and the encoder sampling frequency is
mechanical resonance, the bandwidth of ESO needs to be 2 kHz. Since the control systems of azimuth and elevation
very high which is not realizable in actual application. For- axes have the same architecture, the azimuth axis is used
tunately, NESO has much better observing efficiency than for comparison and verification. The acceleration observer
LESO which may relief the phase loss [28]. And in the appli- used here is Luenberger observer described in [5], and the
cation scenario of large telescopes, the mechanical resonant parameters are: a = 25, kap = 365, kai = 150, kad =
frequency is very low, and the noise of the system has already 179. The speed and position controller of AFC method are
been effectively reduced, so the bandwidth of the ESO can be PI controllers in series, the speed parameters are: ksp =
appropriately increased. 0.3, ksi = 0.07, the position parameters are: kpp = 11.7,
kpi = 2.29.
IV. EXPERIMENT The tuning of the proposed control strategy is as follows:
The experiment setup is shown in Fig. 17. It consists of an 1) The first step is to determine the parameters
operating console, a telescope control unit, a servo controller, of LADRC in the speed loop. According to Table 1,

94408 VOLUME 7, 2019

X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

FIGURE 17. Experimental setup.

TABLE 2. Main parameters used in the experiment. FIGURE 18. Enhanced plant comparison (from u0 to ωM ).

FIGURE 19. Speed close-loop Bode comparison (from ωref to ωM ).

b̂0 = kT JM ≈ 2011. The observer bandwidth ωo and the



controller gain kp can be tuned by according to the bandwidth

method [32].
2) The second step is to determine the parameters of
NADRC. According to the typical third-order NESO empir-
in [33]: δ = 0.05, β01 = 3ωo , β02 = 3ωo2 5,

ical formula
β03 = ωo3 9, so the parameters of NESO are determined.


And the controller gain of NADRC can be selected as kp in

step one.
3) The parameters kp0 , ki0 of PI in the position loop can be
tuned by using Ziegler-Nichols tuning rules.
4) In the feedforward NTD: h0 is determined by h0 = FIGURE 20. Speed step comparison.
10h = 0.005 [28](in this experiment h = 0.0005), can
be selected by the bandwidth method in (27). The feed-
forward
 gains Kω can be set to 1, and Ka can be set to for multi-inertia systems. What’s more, the phase angle of
1 b0 ≈ 0.0005 [34]. NADRC is bigger than LADRC, and much bigger than AFC,
From above discussion, It can be concluded that the pro- which may give better robustness. In Fig. 19, the bandwidth
posed controller only has 5 parameters to tune: ωo , kp , r, kp0 of the speed loop is higher than AFC by using NADRC. The
and ki0 . The parameters used in the experiment are listed in step response in Fig. 20 also demonstrates that NADRC is
the Table 2. faster than the other methods.
As shown in Fig. 18, both the AFC and the ADRC enhance The sensitivity Bode diagram of the position loop is shown
the plant to a relative ideal plant. Unlike above analyses in Fig. 21. The proposed composite control method still keeps
based on the simplified two-inertia model, the AFC method the same error sensitivity peak as the other methods, and
also has a large phase loss at the anti-resonant frequency the position error sensitivity attenuation is strengthened in

VOLUME 7, 2019 94409

 X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

FIGURE 21. Position error sensitivity comparison (from θref to θerr ). FIGURE 24. Position error of sinewave tracking.

TABLE 3. Position error comparison.

proposed method has the smallest overshoot and the fastest

convergence. Fig. 24 shows the tracking error curve of the
sine wave y = 360 sin (0.1667t).
FIGURE 22. Position close-loop bode comparison (from θref to θM ). Table 3 shows the different sinusoidal tracking errors. The
error is reduced by using the method proposed, and the pro-
posed method also has a good suppression effect on some
nonlinear and unknown disturbances such as friction, dead
zone and noise.

V. CONCLUSION
This paper studies the use of the ADRC technique to sup-
press the multi low mechanical resonances appearing in the
large optical telescope. First, a simplified physical model
was established. Concerned about the control of the motor
position, a third-order NESO was designed to estimate and
compensate the original plant to an ideal rigid plant, which
decouples the compliance between the motor and the load.
Moreover, a PI controller was used to achieve a high precision
FIGURE 23. Position trapezoidal wave tracking comparison. position tracking of the telescope. Then based on the NTD,
a speed and acceleration feedforward is established such that
the tracking performance can be improved. The principle
low frequency range and boosted slightly in higher frequency of suppression mechanical resonance of ESO is analyzed in
range where the disturbance is small in telescope control frequency domain. Comparing with AFC, it can be concluded
system. By this way, the tracking accuracy is improved. that the ADRC is better than AFC at suppressing the low
Fig. 22 is closed-loop Bode diagram of position loop, it can frequency mechanical resonance, whereas for higher fre-
be seen that the resonance peak of the proposed method is the quency mechanical resonance, AFC is better. As confirmed
smallest. The presence of a resonant peak will cause position by the experimental results, the proposed control system
overshoot, whereas, for telescope, the reference signal is has good low-frequency mechanical resonances suppressing
often a low-frequency sinusoidal like signals. Fig. 23 shows effect, command-following performance and is robust against
the trapezoidal wave tracking curve of the telescope. The all kinds of disturbances.

94410 VOLUME 7, 2019

X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

where R is the tuning parameter and ψ ∈ C Rn+1 , R .

APPENDIX A
ASSUMPTIONS AND THEORME The control objective of ADRC is to make xi (t) converge to
This section describes the main conclusions of [31]. For an ziR (t) or υ (i−1) (t) in the way of reference system state xi∗ (t)
n-dimensional SISO nonlinear system which can be written converging to zero. The controller of ADRC is
as
1 h _∗ 
_
∗ i

˙ u (t) = ϕ x 1 (t) − zR (t) + z(n+1)R (t) − x n+1 (t)
_ _ b0

 x 1 (t) = x 2 (t)
 (55)
˙

_ _

x (t) = x 3 (t)


 2
  ∗ ∗ ∗  ∗ 
_ _ _ _
.. x 1 (t) , x 2 (t) , . . . , x n (t) , x n+1 (t) is the solu-

where
. (51)
tion of (52) and (z1R (t) , z2R (t) , . . . , znR (t)) , z(n+1)R (t) is



 ˙
_ _

_ _ _

_
x n (t) = f t, x 1 (t) , . . . , x n (t) , w (t) + b u (t)




 the solution of (54). The Assumption 1 below is made for the

_ _ original system in (51) itself and the external disturbance.
y (t) = x 1 (t)

_ ˙
_
Assumption 1: Both w and w are bounded on R, there exist
_ _
where y (t) is the output, u (t) is the input, w
_
∈ positive constants C̄1 , C̄2 and a non-negative function $ ∈
_ C (R) such that
C 1 ([0, ∞) , R) is the external disturbance, f C1 Rn+2 , R


∈  
_ _
represents the nonlinear function of the plant which is possi- ∂f t, x, w
_
bly unknown, and b > 0 is a constant control coefficient 6 C̄1 + $, i = 1, 2, 3, . . . n,
_
∂xi
which is not exactly known, but we have the nominal value  
_ _ _
b 0 that is sufficiently closed to b. f t, 0, w 6 C̄1 + $ (56)
The ESO for system (51) is _ _
   
_ _ _ _
∂ f t, x, w ∂ f t, x, w
 ∗
˙ ∗

_
 +



_ _
x 1 (t) = x 2 (t) + ξ n−1
(t) g1 θ (t) _
∂ wi ∂ti



 ∗   _
∗
6 C̄1 + C̄2 x + $
˙ _

 _
x (t) =
_
x (t) + ξ n−2
(t) g θ (t) (57)
 2

 3 2

..

The Assumption 2 is for ESO (52) and the unknown param-
. (52) _


 ˙
_
∗
_
∗

_

_ _
eter b 0 .
x n (t) = x n+1 (t) + gn θ (t) + b 0 u (t) Assumption 2: |gi (r)| 6 3i |r| for some positive constants




3i for i = 1, 2, . . . , n + 1. There exist constants λ1j ( j =


˙∗
  

_ 1 _
(t) θ (t) 1, 2, 3, 4), β1 , and positive definite continuous differentiable

x = g

 n+1
ξ (t)
 n+1
function V1 , W1 : Rn+1 → R such that
_

_ _
∗ .
_
∗ _
∗
where θ (t) = y (t) − x 1 (t) ξ n (t), y (t) = x 1 (t), gi ∈ λ11 kvk2 6 V1 (v) 6 λ12 kvk2
C (R, R),and ξ ∈ C 0, ∞) , R+ is the gain function to λ13 kvk2 6 W1 (v) 6 λ14 kvk2


be chosen to satisfy ξ (0) = 1, ξ̇ (t) = −aξ (t) , a > 0 if ∀v ∈ Rn+1 (58)

n
ξ (t) > ε, and ξ̇ (0) = 0. X ∂V1 (v) ∂V1 (v)
(vi+1 − gi (v1 )) −gn+1 (v1 ) 6 −W1 (v) ,
A reference system is given as ∂vi ∂vn+1
i=1
 ∗
ẋ1 (t) = x2∗ (t) ∀v ∈ Rn+1 (59)
∂V1 (v)


 ẋ2∗ (t) = x3∗ (t) 6 β1 kvk , ∀v = (v1 , v2 , . . . , vn+1 ) ∈ R n+1


.. (53) ∂vn+1


 . (60)
 ∗
ẋn (t) = ϕ x1∗ (t) , . . . , xn∗ (t) , ϕ (0, . . . , 0) = 0


b−b0
Moreover, the parameter b0 satisfies B , b0 3n+1 <
where ϕ ∈ C (Rn , R).
The TD can be rewritten as λ13
β1 .
 The Assumption 3 is for reference system (53).

 ż∗1R (t) = z2R (t) Assumption 3: ϕ (v) is continuously differentiable, ϕ (0) =
..


. 0, and Lipschitz continuous with Lipschitz constant L:



ϕ (v) − ϕ v̂ 6 L v − v̂ for all v, v̂ ∈ Rn . There exist



 ż∗ (t) = z(n+1)R (t)


nR
(54) constants λ2i ( i = 1, 2, 3, 4), β2 , and positive continuous
 ż(n+1)R (t) differentiable function V2 , W2 : Rn+1 → R such that
z2R (t) z(n+1)R (t)

 
= Rn ψ z1R (t) − υ (t) , ,...,


λ21 kvk2 6 V2 (v) 6 λ22 kvk2 , λ23 kvk2 6 W2 (v) 6 λ24 kvk2

Rn



 R
 ψ (0, 0, . . . , 0) = 0

(61)

VOLUME 7, 2019 94411

 X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

n−1
X ∂V2 (v) ∂V2 (v) TABLE 4. Nomenclatures.
vi+1 + ϕ (v1 , v2 , . . . , vn ) 6 −W2 (v)
∂vi ∂vn
i=1
(62)
∂V2 (v)
6 β2 kvk , ∀v = (v1 , v2 , . . . , vn ) ∈ Rn (63)
∂vn
The Assumption 4 is for TD in (54).
Assumption 4: Both υ (t) and υ̇ (t) are bounded in [0 , ∞),
and ψ (·) is locally Lipschitz continuous, and system (54)
with υ (t) ≡ 0, R ≡ 1 is globally asymptotically stable.
_
Theorem 5: Let x i (t) (1 6 i 6 n) and (1 6 i 6 n + 1) be
the solutions of closed-loop system combined of (51) (52).
_
∗
Let x n+1 (t) be the extended state defined in (52), and let
z1R (t) be the solution of (54). Under Assumptions A1-A4,
the following statements hold true for any given initial values
of (51) and the closed-loop system.
(1) For any σ > 0 and τ > 0, there exists a constant R0 > 0
such that |z1R (t) − ν (t)| < σ uniformly in t ∈ [τ , ∞) for
all R > R0 .
(2) For every R > R0 , there are an R-dependent constant
ε0 > 0 for any ε ∈ (0, ε0 ), and an ε− dependent constant
tε > 0 such that for all t > tε ,
_ _
∗
x i (t) − x i (t) 6 01 εn+2−i , i = 1, 2, . . . , n + 1

and
_
x i (t) − ziR (t) 6 02 ε, i = 1, 2, . . . , n

where 01 and 02 are R-dependent positive constants only.

APPENDIX B
NUMBERICAL ANALYSIS
If let b̂0 = 2011, kp = 0.014, kp0 = 12, ki0 = 2.3. Then the
Lipschitz continuous function ϕ (x1 , x2 , x3 ) is
ϕ (x1 , x2 , x3 ) = −64.8x1 − 337.8x2 − 28.1x3 (64)
And
 
0 1 0
A = 0 0 1
 

−64.8 −337.8 −28.1
 
2.9676 1.3827 0.0077
PA =  1.3827 6.2266 0.0056 
 

0.0077 0.0056 0.0180 V (x) = 2.9676x12 + 6.2266x22 + 0.0180x32

So, the Lyapunov function is + 2.7654x1 x2 + 0.0154x1 x3 + 0.0112x2 x3
V (x) = 2.9676x12 + 6.2266x22 + 0.0180x32 6 4.1580x12 + 7.6149x22 + 0.0313x32
 
+ 2.7654x1 x2 + 0.0154x1 x3 + 0.0112x2 x3 (65) 6 7.6149 x12 + x22 + x32 = λ23 kxk2 (67)
2
V (x) = 2.9676x12 + 6.2266x22 + 0.0180x32 X ∂V   ∂V
xi+1 − b̂0 kp k 0 i x1 + b̂0 kp k 0 p x2 + b̂0 kp x3
+ 2.7654x1 x2 + 0.0154x1 x3 + 0.0112x2 x3 ∂xi ∂x3
i=1

> 1.3772x12 + 4.8383x22 + 0.0047x32 = −x1 2 − x2 2 − x3 2 = − hx, xi = −W (x) (68)

 
> 0.0047 x12 + x22 + x32 = λ21 kxk2 (66) So Assumption 3 is satisfied.

94412 VOLUME 7, 2019

X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

APPENDIX C [23] W. P. Lounsbury and M. Garcia-Sanz, ‘‘High-performance quantitative

NOMENCLATURE robust switching control for optical telescopes,’’ Proc. SPIE, Jul. 2014,
Art. no. 91521F.
See Table 4. [24] Y. Chen, M. Yang, J. Long, K. Hu, D. Xu, and F. Blaabjerg, ‘‘Analysis
of oscillation frequency deviation in elastic coupling digital drive system
and robust notch filter strategy,’’ IEEE Trans. Ind. Electron., vol. 66, no. 1,
REFERENCES
pp. 90–101, Jan. 2019.
[1] M. Garcia-Sanz, T. Ranka, and B. C. Joshi, ‘‘High-performance switching [25] Y. Xia, M. Fu, C. Li, F. Pu, and Y. Xu, ‘‘Active disturbance rejection control
QFT control for large radio telescopes with saturation constraints,’’ in for active suspension system of tracked vehicles with gun,’’ IEEE Trans.
Proc. IEEE Nat. Aerosp. Electron. Conf. (NAECON), Dayton, OH, USA, Ind. Electron., vol. 65, no. 5, pp. 4051–4060, May 2018.
Jul. 2012, pp. 84–91. [26] L. Sun, D. Li, K. Hu, K. Y. Lee, and F. Pan, ‘‘On tuning and practical
[2] W. Gawronski, B. Bienkiewicz, and R. E. Hill, ‘‘Wind-induced dynamics implementation of active disturbance rejection controller: A case study
of a deep space network antenna,’’ J. Sound Vib., vol. 178, no. 1, pp. 67–77, from a regenerative heater in a 1000 MW power plant,’’ Ind. Eng. Chem.
1994. Res., vol. 55, no. 23, pp. 6686–6695, 2016.
[3] W. Gawronski and K. Souccar, ‘‘Control systems of the large millimeter [27] J. Q. Han and W. Wang, ‘‘Nonlinear tracking-differentiator,’’ (in Chinese),
telescope,’’ IEEE Antennas Propag. Mag., vol. 47, no. 4, pp. 41–49, J. Syst. Sci. Math. Sci., vol. 2, pp. 177–183, Jul. 1994.
Aug. 2005. [28] J. Q. Han, Active Disturbance Rejection Control Technique-the Technique
[4] D. Qiu, M. Sun, Z. Wang, Y. Wang, and Z. Chen, ‘‘Practical wind- for Estimating and Compensating the Uncertainties. Washington, DC,
disturbance rejection for large deep space observatory antenna,’’ IEEE USA: National Defense Univ. Press, 2008, pp. 73–75.
Trans. Control Syst. Technol., vol. 22, no. 5, pp. 1983–1990, Sep. 2014. [29] Z.-L. Zhao and B.-Z. Guo, ‘‘A nonlinear extended state observer based
[5] G. Ellis and Z. Gao, ‘‘Cures for low-frequency mechanical resonance in on fractional power functions,’’ Automatica, vol. 81, pp. 286–296,
industrial servo systems,’’ in Proc. IEEE Ind. Appl. Conf. 36th IAS Annu. Jul. 2017.
Meeting, Chicago, IL, USA, Sep./Oct. 2001, pp. 252–258. [30] B.-Z. Guo and Z.-L. Zhao, ‘‘On convergence of tracking differentia-
[6] B. Sedghi, B. Bauvir, and M. Dimmler, ‘‘Acceleration feedback control on tor and application to frequency estimation of sinusoidal signals,’’ in
an AT,’’ Proc. SPIE, Jul. 2008, Art. no. 70121Q. Proc. 8th Asian Control Conf. (ASCC), Kaohsiung, Taiwan, May 2011,
[7] W. Gawronski, ‘‘Control and pointing challenges of large antennas pp. 1470–1475.
and telescopes,’’ IEEE Trans. Control Syst. Technol., vol. 15, no. 2, [31] Z.-L. Zhao and B.-Z. Guo, ‘‘On convergence of nonlinear active distur-
pp. 276–289, Mar. 2007. bance rejection control for SISO nonlinear systems,’’ J. Dyn. Control Syst.,
[8] S. J. Dyke, B. F. Spencer, P. Quast, M. K. Sain, D. C. Kaspari, and vol. 22, no. 2, pp. 385–412, 2016.
T. T. Soong, ‘‘Acceleration feedback control of MDOF structures,’’ J. Eng. [32] Z. Gao, ‘‘Scaling and bandwidth-parameterization based controller
Mech., vol. 122, no. 9, pp. 907–918, 1996. tuning,’’ in Proc. Amer. control Conf., Denver, CO, USA, 2003,
pp. 4989–4996.
[9] E. Dumetz, J. Y. Dieulot, P. J. Barre, F. Colas, and T. Delplace, ‘‘Control
[33] J. Li, Y. Xia, X. Qi, and Z. Gao, ‘‘On the necessity, scheme, and basis of the
of an industrial robot using acceleration feedback,’’ J. Intell. Robotic Syst.,
linear–nonlinear switching in active disturbance rejection control,’’ IEEE
vol. 46, no. 2, pp. 111–128, Jul. 2006.
Trans. Ind. Electron., vol. 64, no. 2, pp. 1425–1435, Feb. 2017.
[10] A. M. Higginson, S. Sanders, and C. Wallett, ‘‘Estimated acceleration
[34] C. Liu, G. Luo, Z. Chen, W. Tu, and C. Qiu, ‘‘A linear ADRC-based robust
feedback applied to a telescope servo system,’’ Mechatronics, vol. 1, no. 4,
high-dynamic double-loop servo system for aircraft electro-mechanical
pp. 509–523, Jan. 1991.
actuators,’’ Chin. J. Aeronaut., to be published.
[11] M. Suárez, J. Rosich, J. Ortega, and A. Pazos, ‘‘The GTC main axes servos
and control system,’’ Proc. SPIE, Jul. 2008, Art. no. 70190J.
[12] Q. Wang, H.-X. Cai, Y.-M. Huang, L. Ge, T. Tang, Y.-R. Su, X. Liu,
J.-Y. Li, D. He, S.-P. Du, ‘‘Acceleration feedback control (AFC) enhanced
by disturbance observation and compensation (DOC) for high precision
tracking in telescope systems,’’ Res. Astron. Astrophys., vol. 16, no. 8,
p. 1254, 2016.
[13] T. Tang, T. Zhang, J.-F. Du, G. Ren, and J. Tian, ‘‘Acceleration feedback of XIN LI received the B.S. and M.S. degrees from the
a current-following synchronized control algorithm for telescope elevation School of Automation, Northwestern Polytechni-
axis,’’ Res. Astron. Astrophys., vol. 16, no. 11, p. 165, 2016.
cal University, China, in 2011 and 2014, respec-
[14] S. Zhao and Z. Gao, ‘‘An active disturbance rejection based approach
tively. He is currently pursuing the Ph.D. degree
to vibration suppression in two-inertia systems,’’ in Proc. Amer. Control
with the Institute of Optics and Electronics (IOE),
Conf., Baltimore, MD, USA, Jun./Jul. 2010, pp. 1520–1525.
Chinese Academy of Sciences (CAS), China. His
[15] S. Zhao and Z. Gao, ‘‘An active disturbance rejection based approach to
vibration suppression in two-inertia systems,’’ Asian J. Control, vol. 15,
research interests include robust control, ADRC,
no. 2, pp. 350–362, 2013. and high-precision servo control systems in large
[16] Q. Zheng and Z. Gao, ‘‘On observer-based active vibration control of two- telescopes.
inertia systems,’’ in Proc. Amer. Control Conf., Washington, DC, USA,
Jun. 2013, pp. 6619–6624.
[17] H. Zhang, S. Zhao, and Z. Gao, ‘‘An active disturbance rejection control
solution for the two-mass-spring benchmark problem,’’ in Proc. Amer.
Control Conf. (ACC), Jul. 2016, pp. 1566–1571.
[18] S. Chen, W. Bai, Y. Hu, Y. Huang, and Z. Gao,‘‘On the conceptualization
of ‘total disturbance’ and its profound implications,’’ Sci. China Inf. Sci.,
vol. 54, pp. 54–146, Apr. 2018.
[19] Y. Chen, B. M. Vinagre, and I. Podlubny, ‘‘Fractional order disturbance
observer for robust vibration suppression,’’ Nonlinear Dyn., vol. 38, WENLIN ZHOU received the B.S. and M.S.
nos. 1–4, pp. 355–367, 2004. degrees from the School of Automation,
[20] Y. Luo, T. Zhang, B. Lee, C. Kang, and Y. Chen, ‘‘Disturbance observer Northwestern Polytechnical University, China,
design with Bode’s ideal cut-off filter in hard-disc-drive servo system,’’ in 2012 and 2015, respectively. She is currently
Mechatronics, vol. 23, no. 7, pp. 856–862, 2013. a Research Associate with the Institute of Optics
[21] J. Cui, S. Zeng, Y. Ren, X. Chen, and Z. Gao, ‘‘On the robustness and and Electronics (IOE), Chinese Academy of Sci-
reliability in the pose deformation system of mobile robots,’’ IEEE Access, ences (CAS), China. Her research interests include
vol. 6, pp. 29747–29756, 2018. robust control, ADRC, PMSM, and motion control
[22] J. Han, ‘‘From PID to active disturbance rejection control,’’ IEEE Trans. systems.
Ind. Electron., vol. 56, no. 3, pp. 900–906, Mar. 2009.

VOLUME 7, 2019 94413

 X. Li et al.: New Mechanical Resonance Suppression Method for Large Optical Telescope by Using NADRC

JUN LUO received the B.S. degree from Northeast PING JIANG graduated from the Sichuan
Normal University and the Ph.D. degree from the University, China. He is currently a Research
Changchun Institute of Optics, Fine Mechanics Fellow with the Institute of Optics and Electron-
and Physics (CIOMP), Chinese Academy of Sci- ics (IOE), Chinese Academy of Sciences (CAS),
ences (CAS), where he is currently a Research China. His research interests include photoelectric
Associate with the Institute of Optics and Elec- detection, precision machinery, and large photo-
tronics (IOE). His research interests include high electric telescope.
precision sensor and motion control systems.

JUNZHANG QIAN received the B.S. degree from

the School of Automation, Northwestern Poly-
technical University, China, and the Ph.D. degree
from the Institute of Optics and Electronics (IOE),
Chinese Academy of Sciences (CAS), China,
where he is currently a Research Associate at Insti-
tute of Optics and Electronics. His research inter-
ests include PMSM, sliding mode control (SMC),
and robust control.

WENLI MA graduated from the University of YONGKUN FAN graduated from the Chang’an
Electronic Science and Technology of China. He is University, China. He is currently an Associate
currently a Research Fellow and a Doctoral Super- Professor with the Sichuan Vocational and Tech-
visor with the Institute of Optics and Electron- nical College of Communications. His research
ics (IOE), Chinese Academy of Sciences (CAS), interests include PMSM, new energy electric vehi-
China. His research interests include photoelectric cles, and robust control.
detection, precision machinery, and large photo-
electric telescope.

94414 VOLUME 7, 2019