You are on page 1of 26

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/288542773

Active Disturbance Rejection Control for PMLM Servo System in CNC Machining

Article  in  Journal of Systems Science and Complexity · January 2015

CITATIONS READS
0 60

1 author:

Wenchao Xue
Chinese Academy of Sciences
47 PUBLICATIONS   730 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ADRC Analysis: time and frequency domain View project

ADRC: a new paradigm of the science of automatic control View project

All content following this page was uploaded by Wenchao Xue on 29 December 2015.

The user has requested enhancement of the downloaded file.


J Syst Sci Complex (2015) 28: 1–25

Active Disturbance Rejection Control for PMLM Servo


System in CNC Machining∗
GUO Jianxin · XUE Wenchao · HU Tao

DOI:
Received: 30 October 2013 / Revised: 4 May 2015
c
The Editorial Office of JSSC & Springer-Verlag Berlin Heidelberg 2015

Abstract Uncertain friction is a key factor that influences the accuracy of servo system in CNC ma-
chine. In this paper, based on the principle of Active Disturbance Rejection Control (ADRC), a control
method is proposed, where both the extended state observer (ESO) and the reduced order extended
state observer (RESO) are used to estimate and compensate for the disturbance. The authors prove
that both approaches ensure high accuracy in theory, and give the criterion for parameters selection.
The authors also prove that ADRC with RESO performs better than that with ESO both in distur-
bance estimation and tracking error. The simulation results on CNC machine show the effectiveness
and feasibility of our control approaches.
Keywords Active disturbance rejection control, CNC machine, nonlinear friction, parameter selec-
tion, servo system.

1 Introduction
With the development of the mechanical manufacturing technology, permanent magnet syn-
chronous motor (PMLM) gradually becomes one of the most competitive motion control prod-
ucts with high speed, high precision, high efficiency, which has been widely utilized in domain
such as NC milling machine, cutting machine tool, lathe or high-grade CNC machine tools[1] .
Due to the linear construction, PMLM motor has an incomparable advantage compared with
rotating motor. One of the most important points is that middle mechanical drive conversion
link is saved so as to realize the feeding system “zero transmission”, to improve the system
transmission stiffness and to reduce the mechanical wear substantially. Moreover, it possesses
higher speed and higher acceleration or deceleration ability. As is known, high speed cutting
GUO Jianxin · XUE Wenchao · HU Tao
KLMM, Academy of Mathematic and Systems Science, Chinese Academy of Sciences, Beijing 100190, China;
Key Laboratory of Systems and Control, Academy of Mathematics and Systems Science, Chinese Academy of
Sciences, Beijing 100190, China. Email: wenchaoxue@amss.ac.cn.
∗ This paper was partially supported by the National Key Basic Research Project of China under Grant No.

2011CB302400, the National Basic Research Program of China under Grant No. 2014CB845303 and the National
Center for Mathematics and Interdisciplinary Sciences, Chinese Academy of Sciences.
⋄ This paper was recommended for publication by Editor HONG Yiguang.
2 GUO JIANXIN · XUE WENCHAO · HU TAO

can significantly improve the metal processing precision, surface processing quality and produc-
tion efficiency as well as the service life of CNC machine tools, by the virtue of which PMLM
can meet the needs of high speed and high precision processing. So it is of great significance in
high-end CNC machine[2] .
Although the linear motor has many advantages compared with the rotating motor, the lin-
ear motor control system also claims higher requirements. The mathematical model of PMLM
is a multi-variable, nonlinear system[3] , which makes it difficult to utilize traditional controller.
With the development of research of the control theory and linear motor, many scholars put
forward various methods from classical control to modern control, such as PID control, adaptive
control, self-learning control[4–6] , and so forth. Nevertheless, these control methods have a vari-
ety of limitations, which either rely too much on accurate mathematical model or narrow scope
of robustness. Hence, a better quality, higher efficiency and more practical control algorithm
are the primary issues in need of solution.
The Active Disturbances Rejection Control (ADRC) is a new control method which does not
depend on the accurate mathematical model of the plant. Through improving the classic PID
inherent defects, the controller algorithm is simple, and the parameters of the scope is adaptable.
Since ADRC can utilize the extended state observer to estimate the total disturbance including
the internal and external disturbance of the plant, it can ensure the system to generate ideal
control effect, perform stronger robustness of maneuverability. These virtues attract much
attention both in theory (see [7–9]) and application (see [10–12]).
In this paper, in order to ensure the accuracy in manufacture process, we apply the ADRC
control method to PMLM mounted in CNC motion system with complicated nonlinear part.
We use series square wave to simulate the unspecific part aiming to approximate practical
situation to a great extend. Moreover, ESO used in ADRC is replaced via reduced order ESO
(RESO), the reason for which is that the measured output does not need to be estimated.
Comparisons are presented to verify the effectiveness of ESO and RESO as well. We also
obtain the relationship between the parameters of ESO (and RESO) and tracking performance.
In addition, the theoretical results help to offer the reference for us to select the parameters of
ESO (and RESO) reasonably in practice. An example from CNC is used to test the approach
in the end.
The paper is organized as follows. The model of the motor and the motion control problem
are introduced in Section 2. The general principle and structure of the ADRC are presented
in Section 3. Stability analysis and parameters tuning method are discussed in Section 4.
Simulation test is shown in Section 5. Finally, some concluding remarks are given in Section 6.

2 Modeling the PMLM Motor


In order to describe the proposed servo control scheme, the voltage controllable PMLM is
adopted, whose mechanical dynamics and circuit dynamics are expressed as the following forces
balance equation[13]
M ẍ1 + Dẋ1 + Fd = Kf i(t) (1)
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 3

and the voltage balance equation

Ke ẋ1 + Li̇ + Ri = u(t), (2)

where x1 is the motion position, ẋ1 = dxdt , and the parameters have the following meaning:
1

M is the moving thrust block mass,


D is the viscious coefficient;
Fd is the load force;
Kf is the amount of force produced by the motor;
i(t) is the armature current;
Ke is the back EMF voltage;
L is the electrical constant parameter;
R is the armature resistance;
u(t) is the armature voltage.
In this paper, the load force is assumed to be composed of three parts: The external load
force which is bounded, the ripple force arising from the magnetic structure of the PMLM
and other physical imperfections, and the frictional force[14, 15] . So the total load force can be
described as
Fd = Fload + Ffric + Fripple ,
where Fload , Ffric , and Fripple are the external load force, friction force, and ripple force respec-
tively. We assume
|Fload (t)| ≤ FlM , t > 0
and the friction is assumed to be
ẋ1 2
Ffric = (Fc + (Fs − Fc )e−( xs ) + Fv |ẋ1 |)sign(ẋ1 ),

where Fc is the Coulomb friction, Fs is the static friction force, Fv is the viscous friction
coefficient, and xs is the lubrication coefficient. The last part of the friction is a significant
disturbance from the uniform magnetic field of the permanent magnet and other physical im-
perfections in the process of linear driver. These factors could be represented as the wave power
to ripple, which often can be written in the form of trigonometric functions of the load position.
In reality, the ripple force is more complex due to variations in the magnet dimension, but it
has a comparable period and amplitude as described in [15]. So, we have

Fripple = Ar sin(ωx1 + φ),


where Ar , ω, φ are constant parameters. Additionally, all the parameters introduced above can
be obtained through experiments and can be regarded as constants.
Finally, since the electrical time constant parameter L is much smaller than the mechanical
ones, the delay of electrical response can be ignored and we obtain a second order system
from (1) and (2)  
1 Kf Ke Fd Kf u(t)
ẍ1 = − + D ẋ1 − + .
M R M MR
4 GUO JIANXIN · XUE WENCHAO · HU TAO

1 Kf Ke Kf
Regarding the current as the internal disturbance and defining a1 = M( R + D), b0 = MR
and fd (ẋ1 , x1 , w(t)) = FMd , the above equation can be written as

ẍ1 = −a1 ẋ1 − fd + b0 u(t).

Define x2 = ẋ1 . Hence we obtain the state equation of the system




 ẋ1 = x2 ,

ẋ2 = x3 + b0 u, (3)



y = x1 ,

where
x3 = −a1 x2 − fd .
In the frame of ADRC, x3 is regard as the “total disturbance” which lumps the external
and the internal disturbance of the system.
1
Note that fd can be written as fd = fv x2 + f0 , fv = FMv , f0 = M [Fload + Fripple + (Fc +
x
−( x2s )2
(Fs − Fc )e )sign(x2 )] and f0 is bounded.

3 ADRC Designs with ESO and RESO


In this section, we give an ADRC based control for PMLM, which consists of an ESO and
a linear form in NLSEF part. Since the derivatives of the signals are known, TD is not used.
General form about the design of ADRC can be referred to [16].
For the PMLM system (5), the ESO for PMLM has the following form

 b2 − β1 (b
x1 − y),
x
 ḃ1 = x
x b3 − β2 (b
ḃ2 = x x1 − y) + b0 u, (4)



ḃ3 = −β3 (b
x x1 − y),

where β1 , β2 , β3 are determined by the ESO-bandwidth ω as follows β1 = 3ω, β2 = 3ω 2 , β3 = ω 3 .


In this paper, the following linear form is used in the place of NLSEF

u0 = kp (v − x
b1 ) + kd (v̇ − x
b2 ) + v̈, (5)

where v stands for the reference input signal, and kp , kd are determined by the controller
bandwidth ωc as follows: kp = ωc2 , kd = 2ωc .
There are two reasons for using the above linear form feedback. First, the controller struc-
ture is simple to be implemented in practice. Second, a large number of application studies
indicate that the linear feedback structure can also perform well to cope with the nonlinear
uncertainties[8, 9, 17] . Thus, the corresponding controller is
u0 − xb3
u= , (6)
b0
which will be used to implement the control in PMLM system (5).
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 5

Since the output of the system y = x1 can be directly measured, it is unnecessary to estimate
x1 by xb1 . According to the reduced order state observer (RESO) in [18], we can design the
following RESO for the PMLM system (5) as follows:
        

 v
ḃ −β 1 vb −β 1 1


  2  =  1   2  +  1  x1 +   b0 u,

 vḃ3 −β2 0 vb3 −β2 0 0
      (7)



 x
b vb β

  2  =  2  +  1  x1 .

x
b3 vb3 β2

Similar to the situation of ESO, β1 , β2 are determined by RESO bandwidth ω: β1 = 2ω, β2 =


ω 2 . Thus the following RESO controller for the PMLM system (5) is obtained:

 u = u0 − x b3
,
b0 (8)

u0 = kp (v − x1 ) + kd (v̇ − x
b2 ) + v̈,

where kp , kd are determined by the controller bandwidth ωc : kp = ωc2 , kd = 2ωc .

4 Stability and Boundary Analysis


In this section we aim to analyze the stability of the ADRC in PMLM. Based on the work
in [19], we further analyze performances of the closed-loop which adopt the stratagy of ESO
and RESO, respectively. Moreover, we prove that ADRC with RESO performs better than that
with ESO both in tracking error and disturbance estimation.
For the load Fload (t) and the reference signal v, we make the assumption as follows.
Assumption 1 |Fload (t)| ≤ FlM < ∞, t > 0, piecewise continuous, and only posses
discontinuity point of the first kind {ti }∞ i=1 , ti−1 < ti , and inf i {ti − ti−1 } > 0; Fload (t) is
derivable on (ti−1 , ti ), i = 1, 2, · · · , and the derivative is bounded as well.
Assumption 2 The reference position signal v and the velocity signal v̇ are continuously
differentiable. Besides, v and v̇ are bounded.
For simplication, we ignore the Coulomb friction and the static friction in the friction model.
So in the system (5), fd reduces to fd = f0 + fv x2 , where
1
f0 = (Fload + Fripple ). (9)
M
Define x∗ = (x∗1 , x∗2 )T ,

ẋ∗1 = x∗2 ,
ẋ∗2 = v̈ − kp (x∗1 − v) − kd (x∗2 − v̇). (10)

Here, the initial value of x∗ are x∗1 (t0 ) = x1 (t0 ), x∗2 (t0 ) = x2 (t0 ). We will show that the
states (x1 , x2 ) of the closed-loop system of ADRC can be closed to the (x∗1 , x∗2 ) for t ∈ [t0 , ∞).
Besides, for convenience, we denote x e = (e e2 )T = (x1 − x∗1 , x2 − x∗2 )T , e = (e1 , e2 , e3 )T =
x1 , x
x1 − x1 , x
(b b2 − x2 , xb3 − x3 )T .
6 GUO JIANXIN · XUE WENCHAO · HU TAO

Theorem 4.1 Based on Assumptions 1 and 2, (5) ensures the closed-loop possessing the
following properties: When ω is large enough,
 
1 ln ω
ke(t)k ≤ T 1 · , t ∈ ti−1 + 2c22 , ti , (11)
ω ω
ln ω
ke
xk ≤ T 2 · , t ∈ [t0 , ∞), (12)
ω
where c22 = 4.3372, T 1 and T 2 are positives depending on ωc , the initial values, uncertainty,
and the reference signals.
The corresponding proof is in the appendix B. From the analysis above, we can draw the
conclusions:
1) (11) ensures that x b1 , x
b2 and x
b3 can track x1 , x2 and x3 fast, so that the disturbance
estimation is smaller in the sections out of the removable points. Moreover, the disturbance
estimation

error can be adjusted via ω. In addition, from the proof in Appendix B, we get T 1 =
c22 c22
c21 s4 + c21 ρ3 , where s4 and ρ3 are positive constants related to the initial values, uncertainty

and the reference signal’s bound. Thus, we get that the larger bound of the uncertainty, the
larger ω required to achieve estimation.
2) (12) indicates that the smaller error will be between x(t) and x∗ (t). Viewed from the
δt
4c22 ρ3 r1 (ωc ) e c12
proof in Appendix B, we get T 2 = c11 · δt , where ρ3 is a positive constant related
e 2c12 −1
to the initial values, uncertainty and the reference signal’s bound and r1 (ωc ) is a decreasing
function with respect to ωc . Thus, we get that the larger ω or ωc , the better tracking effect will
be obtained.
The skill used in the proof of the ESO is similar to that of RESO, which is only different
from the other in the order and small change in parameters. Besides, define x e = (e e2 )T =
x1 , x
(x1 − x∗1 , x2 − x∗2 )T , eb = (b
e2 , eb3 )T = (b b3 − x3 )T . Hence, the closed-loop system of PMLM
x2 − x2 , x
using ADRC with RESO can be described as follows.
Theorem 4.2 Based on Assumptions 1 and 2, (8) ensures the following properties: When
ω is large enough,
 
1 ln ω
kb
e(t)k ≤ T1 · , t ∈ ti−1 + 2c22 , ti , (13)
ω ω
ln ω
ke
xk ≤ T2 · , t ∈ [t0 , ∞), (14)
ω
where c22 = 1.7071, T1 and T2 are positives depending on ωc , the initial values, uncertainty,
and the reference signals.
The corresponding proof is in Appendix A. Similar to the case in Theorem 4.1, we get that
1) (13) confirms that the higher estimation can be achieved through selecting the larger ω
especially when the bound of the uncertainty is considered.
2) Besides, from (14), smaller tracking error, i.e., the difference between x(t) and x∗ (t), will
be obtained via the monotonous selecting larger parameter ω or ωc .
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 7

Remark 4.3 In the light of the proof of Theorems 4.1 and 4.2, we can make a comparison
between ESO and RESO if we choose the same parameters ωc , ω and the same initial values:
1) Since c22 = 4.3372 > c22 = 1.7071, comparison between (11) and (13) shows

ti > t∗i , i = 0, 1, · · · , (15)

which revels that, in the discontinuous points compared with ESO, RESO can estimate the
disturbance faster;
2) According to the proof in Appendix C, we can see that T1 ≤ T1 and T2 ≤ T2 . Therefore,
compared with ESO, the estimation error kb ek is much smaller under RESO. In addition, the
tracking error ke
xk is smaller under RESO as well.
In the next section, simulation will be implemented to further investigate the control effect
between ESO and RESO.

5 CNC Machining Based on ADRC


The motion planning of computer numerical control (CNC) manufacturing systems can be
divided into three main courses: Path planning, trajectory generation and trajectory tracking.
The path planning consists of determining a geometric path according to the specified task.
The desired curve ~r(u) = (x(u), y(u)) (u is curve parameter) is generated from a solid modeler,
the process of which is also named as CAM (Computer Aided Manufacturing)[20] . In the
trajectory generation process, kinematic bounds information, such as velocity, acceleration,
and jerk (derivative of acceleration with respect to time), are determined, after which the
interpolator which calculates the intermediate coordinates using an appropriate interpolation
algorithm[21–24] . Due to the function of the interpolator relating the parameter u (the curve
parameter) and time t, the description of the trajectory can be used in the control system.
Such a relationship can be determined if the feedrate specifications are known. Feedrate will be
determined through optimization with considering the kinematic information determined at the
beginning, and it can be a constant or variable throughout the path. Given a feedrate profile
vf (t), it is possible to obtain the desired coordinates as a function of time
s
 2  2 q
dx(u) dy(u) du
vf (t) = + = x′ (u)2 + y ′ (u)2 ,
dt dt dt

d
where (·) = du (·).
Since the feedrate is assumed to be known, we have a differential equation between u and t
du vf (t)
=p ′ .
dt x (u)2 + y ′ (u)2

Using this differential equation, the command signal is now available as a function of time.
It is also possible to find the time derivative of the command for control purposes. Finally, it is
easy to implement trajectory tracking process using the closed-loop controllers to regulate the
actuators moving along the planned trajectory. The basic system consists of two motors, each
8 GUO JIANXIN · XUE WENCHAO · HU TAO

motor moving the tool along a single axis of motion. In the simulation below, we apply PMLM
motor to act as the axis drive motor. ESO and RESO are mounted on ADRC, respectively.
The overall system for ESO case is illustrated in Figure 1.

Table 1 Parameters used in the simulation


Meaning Parameter Unit Value
Mass of thrust block M Kg 5.4
Viscosity constant D Ns/m 10
Armature resistance R Ohms 16.8
Armature inductance L Henry 17.4 × 10−3
electrical-mechanical energy conversion constant Kf N/A 130
back electromotive force (EMF) voltage Ke v/m/s 123
Minimum level of coulomb friction fc Newton 10
Static friction fs Newton 20
Viscous friction parameter fv Newton 10
Ripple paremeter 1 Ar Newton 8.5
Ripple patemeter 2 ω Rad/m 314
Ripple patemeter 3 ϕ unit 0.05π
Bounded force flm Newton 100
Viscous parameter xs m/s 0.1

v&&
+
Interpolator v& +
/LQHDU
u0 (t ) 1 u (t ) y (t )
− :HLJKWHG6XP
3ODQW
+ − b0
v + −

Feedrate Profile x̂3 b0

x̂2
3UHSURFHVVLQJ ESO
x̂1

Figure 1 Overall system

Figure 2 is a butterfly curved segments[25] generated from the solid modeler of the CNC
system.
There is a lot of velocity planning methods. In this paper, a quintic splines trajectory
generation algorithm[26] is utilized to connect a series of reference knots that produces con-
tinuous position, feedrate optimization technique in [27] for minimizing the cycle time in ma-
chining spline tool paths with axis velocity, acceleration and jerk limits. The plane butter-
fly curve is discretized to more than 100 points which are close enough to be the reference
~max = (200, 200)mm/s, A
knots. The drive constraints are set to be V ~ max = (4000, 4000)mm/s2,
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 9

J~max = (50000, 50000)mm/s3. The initial and terminal velocities and accelerations are all zero,
and we set the initial position of the curve at origin points (0, 0); the sampling period is set to
be 1ms and the same physical parameters are assumed for both the x and y axes; the parame-
ters of PMLM are given in the Table 1. Especially, Fload is simulated as a piece constants, the
period of which is 3s with eighties percents of the periods being constant value 50 and the rest
being value 100. Simulation is carried out to make comparison between ESO and RESO which
are applied in ADRC by using Matlab/Simulink. The control bandwidth and the bandwidth
for the simulation is ωc =800 and ω =1000, which are adjusted under the guidance from the
theorem discussed in Section 4. The effects are demonstrated in the following figures.

Butterfly Curve
30

20

10

0
Y axis (mm)

−10

−20

−30

−40

−50

−60
−60 −40 −20 0 20 40 60
X axis (mm)

Figure 2 The reference knots and cubic spline trajectory

X−Axis (ESO based ADRC)


Tracking Errors(mm)

0.05

−0.05
0 1 2 3 4 5
Time (s)
Y−Axis (ESO based ADRC)
Tracking Errors(mm)

0.05

−0.05
0 1 2 3 4 5
Time (s)

Figure 3 Tracking error of ESO-ADRC


10 GUO JIANXIN · XUE WENCHAO · HU TAO

X−Axis (RESO based ADRC)

Tracking Errors(mm)
0.05

−0.05
0 1 2 3 4 5
Time (s)
Y−Axis (RESO based ADRC)

Tracking Errors(mm)
0.05

−0.05
0 1 2 3 4 5
Time (s)

Figure 4 Tracking error of RESO-ADRC

X−Axis (ESO based ADRC)


200
Voltage(V)

100

−100

−200
0 1 2 3 4 5
Time (s)
Y−Axis (ESO based ADRC)
200
Voltage(V)

100

−100

−200
0 1 2 3 4 5
Time (s)

Figure 5 Voltage of ESO-ADRC

X−Axis (RESO based ADRC)


200
Voltage(V)

100

−100

−200
0 1 2 3 4 5
Time (s)
Y−Axis (RESO based ADRC)
200
Voltage(V)

100

−100

−200
0 1 2 3 4 5
Time (s)

Figure 6 Voltage of RESO-ADRC

Figures 3 and 4 are the tracking error comparison. As can be seen from the picture, if we
PN
define mean error emean = ( i=1 |ei |)/N to demonstrate average accuracy in the total progress,
where N denotes the number of the interpolation points, then the corresponding axis average
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 11

accuracy for the two method are ~eESO


mean = (0.0060, 0.0079)mm and ~ eRESO
mean = (0.0032, 0.0038)mm,
respectively. Thus, the error generated by ESO is larger than that of RESO, at the cost of a
small vibration of the control voltage (near zeros in the horizontal axis) which is shown in
Figures 5 and 6.
Figures 7 and 8, 9 and 10 are the corresponding results for tracking error comparison and
voltage comparison when the input signals are combined with certain measurement noise, re-
spectively. We denote the input signals of X and Y axis as Xoriginal and Yoriginal , Xoriginal , Yoriginal ∈
RN . Considering the noise signals to X and Y axis as εX and εY , in which satisfy Gauss dis-
tribution:

εX ∼ N (0, 0.002% kXoriginal k∞ ), εY ∼ N (0, 0.002% kYoriginal k∞ ).

Hence, the real input signals are:

Xreal = Xoriginal + εX , Yreal = Yoriginal + εY .

The simulation shows that the designed controllers have certain filtering effect for measurement
noise. Obviously, ESO is superior to the RESO in filtering effect for its higher order.
X−Axis (ESO based ADRC)
Tracking Errors(mm)

0.05

−0.05
0 1 2 3 4 5
Time (s)
Y−Axis (ESO based ADRC)
Tracking Errors(mm)

0.05

−0.05
0 1 2 3 4 5
Time (s)

Figure 7 Tracking error of ESO-ADRC with measurement noise

X−Axis (RESO based ADRC)


Tracking Errors(mm)

0.05

−0.05
0 1 2 3 4 5
Time (s)
Y−Axis (RESO based ADRC)
Tracking Errors(mm)

0.05

−0.05
0 1 2 3 4 5
Time (s)

Figure 8 Tracking error of RESO-ADRC with measurement noise


12 GUO JIANXIN · XUE WENCHAO · HU TAO

X−Axis (ESO based ADRC)


200

100

Voltage
0

−100

−200
0 1 2 3 4 5
Time (s)
Y−Axis (ESO based ADRC)
200

Voltage 100

−100

−200
0 1 2 3 4 5
Time (s)

Figure 9 Voltage of ESO-ADRC with measurement noise

X−Axis (RESO based ADRC)


200

100
Voltage

−100

−200
0 1 2 3 4 5
Time (s)
Y−Axis (RESO based ADRC)
200

100
Voltage

−100

−200
0 1 2 3 4 5
Time (s)

Figure 10 Voltage of RESO-ADRC with measurement noise

Remark 5.1 RESO performs better than the ESO both in disturbance estimation and
tracking error. The physical reason for this phenomenon should be addressed more vividly.
ω3 ω2
Actually, for ESO, x b3 = (s+ωe )3 x3 ; for RESO x
b3 = (s+ωe )2 x3 . Thus, RESO performs better
e e
than ESO both in estimation speed toward the disturbance and compensation speed toward
the disturbance, and hence the closed-loop property is better. On the other hand, the above
transfer functions mean that ESO is superior to the RESO in filtering the measurement noise.
Hence, we suggest to choose the ESO based ADRC and RESO based ADRC according to the
specific properties of the plant.

6 Conclusion
The permanent magnet linear motor (PMLM) has high speed, high precision, and is widely
used in CNC machining. However, because of its nonlinearity and uncertainty as well as easily
being subjected to the existing outside disturbance, the traditional control method such as
PID is hard to meet its performance requirements. To improve the performance of the system
PMLM, this paper utilizes the ADRC to design the controller in the system of PMLM due to
its good performance, robust, and simple for implementing.
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 13

To mimic the practical manufacture situation, the nonlinear load force of PMLM, including
friction and magnetic resistance as well as the nonlinear disturbances are considered. The
stability of closed-loop system is proved. The control accuracy and quality are analyzed. We
use a tangible manufacture example from CNC machining to perform the simulation of PMLM.
ADRCs using ESO and RESO are implemented to specify the effects of these two methods
respectively.
The result showed the effectiveness and feasibility of both ESO and RESO. Moreover, com-
pared with ESO, RESO exhibits better performance at the cost of a little vibration in control
signal.

References
[1] Brandenburg G, Bruckl S, Dormann J, et al., Comparative investigation of rotary and linear motor
feed drive systems for high precision machine tools, Advanced Motion Control, 2000, Proceedings
6th International Workshop on IEEE, 2000, 384–389.
[2] Seamus G and Michael T H, Development of a high-speed CNC cutting machine using linear
motors, Journal of Materials Processing Technology, 2005, 166(3): 321–329.
[3] Lin F J, Shyu K K, and Lin C H, Incremental motion control of linear synchronous motor, IEEE
Transactions on Aerospace and Electronic Systems, 2002, 38(3): 1011–1022.
[4] Chong G and Li Y, PID control system analysis, design, and technology, IEEE Transactions
Conontrol Systems Technology, 2005, 13(4): 559–576.
[5] Narendra and Kumpati S, Robust adaptive control, American Control Conference, 1984.
[6] Bristow D, Tharayil M, and Alleyne A G, A survey of iterative learning control, IEEE Transactions
Conontrol Systems Technology, 2006, 26(3): 96–114.
[7] Guo B and Zhao Z, On convergence of the nonlinear active disturbance rejection control for mimo
systems, SIAM J. Control and Optimization, 2013, 51(2): 1727–1757.
[8] Zheng Q, Chen Z Z, and Gao Z Q, A practical approach to disturbance decoupling control, Control
Engineering Practice, 2009, 17(9): 1016–1025.
[9] Xue W C and Huang Y, On performance analysis of ADRC for nonlinear uncertain systems
with unknown dynamics and discontinuous disturbances, 2013 Chinese Control Conference, 2013,
1102–1107.
[10] Dong J Y and Li D H, Active disturbance rejection control for complex dynamical systems, System
Science and Mathematic, 2013, 33(6): 639–652 (in Chinese).
[11] Jiang Z, Active disturbance rejection control for the yaw tracking for helicopter, System Science
and Mathematic, 2012, 32(6): 641—652 (in Chinese).
[12] Zheng Q, Gao L Q, and Gao Z, On validation of extended state observer through analysis and
experimentation, Journal of Dynamic Systems Measurement & Control, 2012, 134(2): 224–240.
[13] Xiao Y and Zhu K Y, Optimal synchronization control of highprecision motion systems, IEEE
Transactions on Industrial Electronics, 2006, 53(4): 1160–1169.
[14] Canudas-De-Wit C, Olsson H, Astrom K J, et al., A new model for control of systems with friction,
IEEE Transactions on Automatic Control, 1995, 40(3): 419–425.
[15] Tan K K, Huang S N, and Lee T H, Robust adaptive numerical compensation for friction and
14 GUO JIANXIN · XUE WENCHAO · HU TAO

force ripple in permanent-magnet linear motors, IEEE Transactions on Magnetics, 2002, 38(1):
221–228.
[16] Han J Q, Auto-disturbances-rejection controller and its application, Control and Decision, 1998,
13(1): 19–23.
[17] Gao Z Q, Haung Y, and Han J Q, An alternative paradigm for control system design, IEEE
Conference on Decision and Control, 2001, 5: 4578–4585.
[18] Xue W C, On Theoretical analysis of active disturbance rejection control, Ph.D. Thesis, 2012.
[19] Hu T, Xue W C, and Huang Y, Active disturbance rejection control for permanent magnet linear
motor, Proceedings of the 31st Chinese Control Conference, 2012, 7: 296–301.
[20] Castelino K, D’Souza R, and Wright P K, Tool path optimization for minimizing airtime during
machining, Journal of Manufacturing Systems, 2003, 22(3): 173–180.
[21] Guo J X, Zhang Q, Gao X S, et al., Time optimal feedrate generation with confined tracking error
based on linear programming, Journal of Systems Science and Complexity, 2015, 28(1): 80–95.
[22] Zhang K, Gao X S, Li H, and Yuan C M, A greedy algorithm for feed-rate planning of CNC
machines along curved tool paths with confined jerk for each axis, Robotics and computer Integrated
Manufacturing, 2012, 28: 472–483.
[23] Fan W, Gao X S, Yan W, et al., Interpolation of parametric CNC machining path under confined
jounce, The International Journal of Advanced Manufacturing Technology, 2012, 62(5–8): 719–
739.
[24] Zhao C and Huang Y, ADRC bsed input disturbance rejection for minimum-phase plants with
unknown orders and/or uncertain relative degrees, Journal of Systems Science and Complexity,
2012, 25(4): 625–640.
[25] Yau H T, Lin M T, and Tsai M S, Real-time NURBS interpolation using FPGA for high speed
motion control, Computer-Aided Design, 2006, 38: 1123–1133.
[26] Erkorkmaz K and Altintas Y, High speed CNC system design Part I: Jerk limited trajectory gen-
eration and quintic spline interpolation, International Journal of Machine Tools and Manufacture,
2001, 41: 1323–1345.
[27] Altintas Y and Erkorkmaz K, Feedrate optimization for spline interpolation in high speed machine
tools, CIRP Annals—Manufacturing Technology, 2003, 52(1): 297–302.

Appendix
Before giving the corresponding proof, the closed-loop form of ESO and RESO are rendered
respectively.
From(1), (4), (5), (10), (73), we get the closed-loop system of ESO:
      
x
ė1 0 1 0 0 0 x
e1 0
      
x   2 2 2
−1   e   −f 
 ė2   −ωc −2ωc −ωc /ω −2ωc /ω x 2  0
      
 ζ̇1  =  0 0     
   0 −3ω ω   ζ1  +  0  , (16)
      
 ζ̇2   0 0 −3ω 0 ω     
     ζ2   ωf0 
ζ̇3 0 0 −ω 0 0 ζ3 η3
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 15

where
   
ζ1 ω 2 e1
   
ζ=  
 ζ2  =  ωe2  ,
 (17)
ζ3 e3 − f0

ω2
η3 = −(ωc2 x e2 + ωc2 ζ1 + 2 ωωc ζ2 + ζ3 + f0 )e
e1 + 2ωc x a1 + (ωc2 v + 2ωc v̇ + v̈ − ωc2 x∗1 − 2ωc x∗2 )e
a1 ,
a1 = a1 + f v .
e
Besides, from (5), (8), (10), (20), we can obtain the closed-loop system of RESO:
      
x
ė1 0 1 0 0 x
e1 0
      
x   2   e   −f 
 ė2   −ωc −2ωc −2ωc/ω −1   x 2  0
 =   +  , (18)
 ξ̇2   0 0 −2ω ω   ξ2   ωf0 
      
ξ̇3 0 0 −ω 0 ξ3 η1

where
   
ξ2 ωb
e2
ξ= = . (19)
ξ3 eb3 − f0

and η1 = −(ωc2 x e2 + 2ω
e1 + 2ωcx ω ξ2 + ξ3 + f0 )e
c
a1 + (ωc2 v + 2ωc v̇ + v̈ − ωc2 x∗1 − 2ωcx∗2 )e
a1 , e
a1 = a1 + f v .
The proof from ESO is similar to that from RESO. Nevertheless, the proof of RESO is more
presentative, which helps to understand the principle and process of the proof. Hence, we give
more detail to RESO initially.
Appendix A: The Proof of Theorem 4.2
We will prove the theorem in two steps.
Step 1 Prove that x1 , x2 , eb2 , eb3 are bounded in [t0 , ∞).
Make transformation
   
ξ2 ωb
e2
ξ= = . (20)
ξ3 eb3 − f0

Define
     
0 1 0 0 −2 1
Ac =  , Bc =  , A2 =  .
−ωc2 −2ωc −2ωc /ω −1 −1 0

 
Since Ac , A2 are Hurwitz matrixes, there exist positive matrixes P1 = mn and P2 =
n l
 
p q , satisfying
qr

AT
c P1 + P1 Ac = −I2 , c11 I2 ≤ P1 ≤ c12 I2 (21)
AT
2 P2 + P2 A2 = −I2 , c21 I2 ≤ P2 ≤ c22 I2 , (22)
16 GUO JIANXIN · XUE WENCHAO · HU TAO

where c11 , c12 are the minimum and maximum eigenvalues of P1 ; c21 , c22 are the minimum and
maximum eigenvalues of P2 . From (21), (22), we have
 2   
ωc +5 1 1 1
2 −
P1 =  4ωc ω2ω 2
c 
, P2 =  2 2
,
1 c +1 1 3
2ωc 2 4ωc 3 − 2 2
1 h 4 2
p i
c11 = 3
ωc + 6ωc + 1 − (ωc4 + 6ωc2 + 1)(ωc4 + 2ωc2 + 1) , (23)
8ωc
1 h p i
c12 = 3
ωc4 + 6ωc2 + 1 + (ωc4 + 6ωc2 + 1)(ωc4 + 2ωc2 + 1) ,
8ωc
√ √
2− 2 2+ 2
c21 = , c22 = .
2 2
k·k below is the 2-norm. Define Lyapunov function, along the trajectory of (18): V1 (e eT P1 x
x) = x e

V̇1 (e xT P1 x
x) = 2e xk2 + 2e
ė = −ke xT P1 (Bc ξ + ( 0 1 )T (−f0 ))

xk2 + 2ke
≤ −ke xk · kP1 ( 0 1 )T f0 k.
xk · kP1 Bc k · kξk + 2ke

In the formula above, when ω ≥ 2ωc , the second and the last parts of the equation are
!
√ √ −2nωc /ω −n
kP1 Bc k ≤ 2kP1 Bc k1 = 2


−2lωc /ω −l 1

= 2(n + l) max{2ωc /ω, 1}
 
√ 1 ωc2 + 1
= 2 + ,
2ωc2 4ωc3
s
p  2  2 2
T 1 ωc + 1
2 2
kP1 0 1 f0 k = |f0 | n + l = |f0 | + .
2ωc2 4ωc3
1
Due to the bound of the |f0 | (f0 = M (Fload+ Fripple )), suppose |f0 | ≤ κ1 and define
 
√ 1 ωc2 + 1
r1 (ωc ) = 2 2 + , (24)
2ωc2 4ωc3
s 2  2 2
1 ωc + 1
r2 (ωc ) = 2 + . (25)
2ωc2 4ωc3
Then

V̇1 (e xk2 + r1 (ωc )ke


x) ≤ −ke xk · kξk + κ1 r2 (ωc )ke
xk. (26)

Construct Lyapunov function: V2 (ξ) = ξ T P2 ξ and define K1 = ( ωc2 2ωc ), K2 = ( 2ωc /ω 1 ),


R = ( ωf0 η1 )T .
Along the trajectory of (18):

V̇2 (ξ) = 2ξ T P2 (ωA2 ξ + R)


= −ωkξk2 + 2ωξ T P2 ( 1 0 )T f0 + 2ξ T P2 ( 0 1 )T (−K1 x
e − K 2 ξ − f0
+ωc2 v + 2ωc v̇ + v̈ − ωc2 x∗1 − 2ωc x∗2 )e
a1 .
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 17

Suppose |f0 − (ωc2 v + 2ωc v̇ + v̈ − ωc2 x∗1 − 2ωc x∗2 )| ≤ κ2 , and define
p
s1 (ωc ) = |ea1 | 10((2ωc )2 + ωc4 ), (27)

s2 = 2 5|e a1 |, (28)

s3 = κ2 10|e a1 |. (29)

Hence, when ω ≥ 2ωc , in consideration of


1 !
T 2 − 21  T 
kP2 0 1 K1 k ≤ 01 · k ω 2 2ωc k = s1 (ωc ) ,
1 3 c a1 |
2|e
−2 2
1 !
T 2 − 21  T 
kP2 0 1 K2 k ≤ 0 1 · k 2ωc /ω 1 k ≤ s2 ,
1 3 2|ea1 |
−2 2
T s3
kP2 0 1 (f0 − (ωc2 v + 2ωc v̇ + v̈ − ωc2 x∗1 − 2ωc x∗2 ))k ≤ ,
a1 |
2|e
we have

V̇2 (ξ) ≤ −(ω − s2 )kξk2 + ( 2ωκ1 + s3 )kξk + s1 (ωc )ke
xk · kξk. (30)

Define
  
p √ ρ2
ρ1 = max V1 (e
x(t0 )), 2 c12 √ r1 (ωc ) + κ1 r2 (ωc ) , (31)
c21
p √ √
ρ2 = max{ V2 (ξ(t0 )), ( 2 + 1) c22 κ1 }. (32)
p
We will prove that when ω > κ11 ( √ρc222 s1 (ωc ) √ρc111 + s2 + s3 ), then Ω = {(e x) ≤
x(t0 ), ξ)| V1 (e
p
ρ1 , V2 (ξ) ≤ ρ2 } is a positive invariant set of (18). We develop the proof in two situations:
p p
x) = ρ1 , V2 (ξ) ≤ ρ2 , then from (26), along the trajectory of (18), we have
i) if V1 (e

x) ≤ −ke
V̇1 (e xk · [ke
xk − (r1 (ωc )kξk + κ1 r2 (ωc ))]
  
ρ1 ρ2
≤ −kexk · √ − r1 (ωc ) √ + κ1 r2 (ωc )
c12 c21
 
ρ2
≤ −kexk · r1 (ωc ) √ + κ1 r2 (ωc )
c21
≤ 0; (33)
p p
x) ≤ ρ1 , V2 (ξ) = ρ2 , then from (30), along the trajectory of (18), we have
ii) if V1 (e

V̇2 (ξ) ≤ −kξk · [(ω − s2 )kξk − (( 2ωκ1 + s3 ) + s1 (ωc )ke xk)]
  
ρ2 √ ρ1
≤ −kξk · (ω − s2 ) √ − ( 2ωκ1 + s3 ) + s1 (ωc ) √
c22 c11
   
ρ2 √ ρ2 ρ1
≤ −kξk · √ − 2κ1 ω − √ s2 + s3 + s1 (ωc ) √
c22 c22 c11
  
ρ2 ρ1
≤ −kξk · κ1 ω − √ s2 + s3 + s1 (ωc ) √
c22 c11
< 0. (34)
18 GUO JIANXIN · XUE WENCHAO · HU TAO

From (33) and (34), we can come to the comclusion: If ω > κ11 ( √ρc111 s1 (ωc ) + √ρc222 s2 + s3 ), Ω is
a positive invariant set of (18).
From the analysis above, we can draw the conclusion that when the bandwidth of RESO ω
and the bandwidth of control ωc satisfy
  
1 ρ1 ρ2
ω > max 2ωc , √ s1 (ωc ) + √ s2 + s3 , (35)
κ1 c11 c22

then ke
xk, kξk are bounded:
ρ1 ρ2
ke
xk ≤ √ , kξk ≤ √ . (36)
c11 c21
 
Since x
e= x1 − x∗ , and x∗ , ẋ∗ are bounded, x and x are bounded. At the same time, from
1 2
x2 − ẋ ∗

the transformation (20) we can get eb2 , eb3 are bounded.


Step 2 Due to Assumption 1, the load force |Fload (t)| only contains the first discontinuous
1
points {ti }∞
i=1 , ti−1 < ti , and inf i {ti − ti−1 } > 0. Since f0 = M (Fload + Fripple ), f0 onle
contains the discontinuous points of the first kind {ti }i=1 , ti−1 < ti , and inf i {ti − ti−1 } > 0.

Define δt = inf i {ti − ti−1 }. The derivative of f0 in a general form can be defined as

 df0 , t 6= ti ,
f˙0 = dt (37)
 H δ(t − t ), t = t ,
i i i

where |f˙0 | ≤ D < ∞, |Hi | ≤ H < ∞, δ(·) have the properties:



Z  0, t < t
i
δ(t − ti )dt =
 1, t ≥ ti .

Make a transformation:
   
ξ2 ωb
e2
ξ= = . (38)
ξ3 eb3 .

From (5), (8) and (10), we obtain the closed-loop


      
x
ė1 0 1 0 0 x
e1 0
      
x   2 e   0 
 ė2   −ωc −2ωc −2ωc/ω −1   x2  
˙ =   +  , (39)
ξ   0 0 −2ω ω   ξ 2  
   
 2  0
ξ˙ 3 0 0 −ω 0 ξ3 η2

where

η2 = −(ωc2 x
e1 + 2ωc x
e2 + 2ωc eb2 + eb3 )e a1 + f˙0 .
a1 + (ωc2 v + 2ωc v̇ + v̈ − ωc2 x∗1 − 2ωc x∗2 )e (40)
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 19

Comparing (20) and (38), we can get


q q
2 2
kξk = ξ 2 + ξ 3 = ξ22 + (ξ3 + f0 )2
√ q
≤ 2 · ξ22 + ξ32 + f02

≤ 2(kξk + |f0 |).

Since ξ and f0 are bounded, and kξk ≤ √ρ2 , |f0 | ≤ κ1 ,


c21
 
√ ρ2
kξk ≤ ρ3 , ρ3 = 2 √ + κ1 . (41)
c21
When t0 ≤ t < t1 , along the trajectory (39),
T T 
V̇2 (ξ) = 2ξ P2 ωA2 ξ + 0 1 η2 .

Define kη2 (t)k ≤ κ3 , t 6= ti , i = 1, 2, · · · , here κ3 = κ3 (ωc , ρ1 , ρ2 , v, v̇, v̈, Ar , D). Since η2 is


bounded in the continuous points of f0 , then
√ √
T 10 10 s4
kP2 ( 0 1 ) k · |η2 | = |η2 | ≤ κ3 = , (42)
2 2 2

where s4 = 10κ3 .
Hence V̇2 (ξ) ≤ −ωkξk2 + s4 kξk, and
q
d V˙2 (ξ) kξk2 s4 kξk
V2 (ξ) = q ≤ −ω q + q
dt
2 V2 (ξ) 2 V2 (ξ) 2 V2 (ξ)
q
V2 (ξ) s4
≤ −ω + √ . (43)
2c22 2 c21
From Gronwall-Bellman inequation, we can get, when t0 ≤ t < t1 ,
q q
1 c22 s4 ω
V2 (ξ) ≤ √ + V2 (ξ(t0 ))e− 2c22 (t−t0 )
ω c21
1 c22 s4 √ ω
≤ √ + c22 kξ(t0 )ke− 2c22 (t−t0 )
ω c21
1 c22 s4 √ − ω (t−t0 )
≤ √ + ρ3 c22 e 2c22 . (44)
ω c21
2c22
If δt = inf i {ti − ti−1 } ≤ then {ω|δt = 2c22ωln ω } 6= ∅. Define
e ,
  
 2c ln ω 2c22
 max ω|δt = 22 , δt ≤ ,
ω1 = ω e (45)

 0, 2c22
δt > .
e
Then when ω > ω1 , we have
2c22 ln ω
< δt . (46)
ω
20 GUO JIANXIN · XUE WENCHAO · HU TAO

Define
2c22 ln ω
t∗0 = t0 + . (47)
ω
From (45), t∗0 < t1 , then
ω 1
e− 2c22 (t−t0 ) ≤ , t ≥ t∗0 . (48)
ω
From (44) and (48), we get
q  
1 c22 s4 √
V2 (ξ) ≤ √ + ρ3 c22 , t ∈ [t∗0 , t1 ). (49)
ω c21

Hence,
 √ 
1 c22 c22
kξk ≤ s4 + √ ρ3 , t ∈ [t∗0 , t1 ). (50)
ω c21 c21

From the transformation (38), we get


 √ 
1 c22 c22
kb
ek ≤ kξk ≤ s4 + √ ρ3 , t ∈ [t∗0 , t1 ). (51)
ω c21 c21

Similar to (44), when t1 ≤ t < t2 ,


q q
1 c22 s4 − 2cω (t−t1 )
V2 (ξ) ≤ √ + V2 (ξ(t+1 ))e
22
ω c21
1 c22 s4 √ − 2cω (t−t1 )
≤ √ + c22 kξ(t+1 )ke
22
ω c21
1 c22 s4 √ − ω (t−t1 )
≤ √ + ρ3 c22 e 2c22 . (52)
ω c21

Define
2c22 ln ω
t∗1 = t1 + . (53)
ω
Then
− 2cω (t−t1 ) 1
e 22 ≤ , t ≥ t∗1 . (54)
ω
From (52) and (54), we obtain
q  
1 c22 s4 √
V2 (ξ) ≤ √ + ρ3 c22 , t ∈ [t∗1 , t2 ). (55)
ω c21

Hence,
 √ 
1 c22 c22
kξk ≤ s4 + √ ρ3 , t ∈ [t∗1 , t2 ). (56)
ω c21 c21
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 21

Similarly, we can prove for i ≥ 0,


 √ 
1 c22 c22
kξk ≤ s4 + √ ρ3 , t ∈ [t∗i , ti+1 ), (57)
ω c21 c21
where
2c22 ln ω
t∗i = ti + . (58)
ω
From the transformation (38), for i ≥ 1,
 √ 
1 c22 c22
kbek ≤ kξk ≤ s4 + √ ρ3 , t ∈ [t∗i , ti+1 ). (59)
ω c21 c21

Next, we will prove that, when ω is large enough (ω ≫ 1), kx(t) − x∗ (t)k ≤ O( lnωω ), t ∈
[t0 , ∞).
Along the trajectory of (39) we have

V̇1 (e xT P1 x
x) = 2e xk2 + r1 (ωc )kξ|k · ke
ė ≤ −ke xk.

So
dp V˙1 (e
x) kexk2 xk · kξk
r1 (ωc )ke
x) = p
V1 (e ≤− p + p
dt 2 V1 (e x) 2 V1 (ex) 2 V1 (ex)
p
V1 (e
x) r1 (ωc )
≤− + √ kξk.
2c12 2 c11
Since x
e(t0 ) = 0, from Gronwall-Bellman inequation, we get
p Z
r1 (ωc ) t − 2c
t−τ
x) ≤ √
V1 (e e 12 kξ(τ )kdτ. (60)
2 c11 t0
When tk−1 ≤ t < tk ,
X
k Z t∗ X Z ti
k−1
p r1 (ωc ) i−1 t−τ
− 2c t−τ
x) ≤ √
V1 (e e 12 kξ(τ )kdτ + e− 2c12 kξ(τ )kdτ
2 c11 i=1 ti−1 i=1 t∗
i−1
Z t 
t−τ
+ e− 2c12 kξ(τ )kdτ . (61)
t∗
k−1

From (41),
X Z ti
k−1 Z t  √ 
t−τ t−τ c22 c22 1
e− 2c12 kξ(τ )kdτ + e− 2c12 kξ(τ )kdτ < s4 + √ H , (62)
i=1 t∗
i−1 t∗
k−1
c21 c21 ω
k Z
X t∗ k t δ
i−1 t−τ
− 2c ln ω X − t−ti e 2c12 ln ω
e 12 kξ(τ )kdτ ≤ 4c ρ e 2c12
≤ 4c ρ . (63)
22 3 22 3 δt
i=1 ti−1 ω i=1 1 − e− 2c12 ω
Then, from (61), (62) and (63), we get
 √  δt 
p r1 (ωc ) c22 c22 1 e 2c12 ln ω
x) ≤ √
V1 (e s4 + √ ρ3 + 4c22 ρ3 δt . (64)
2 c11 c21 c21 ω 1 − e− 2c12 ω
22 GUO JIANXIN · XUE WENCHAO · HU TAO

Hence,
 √  δt 
r1 (ωc ) c22 c22 1 e 2c12 ln ω
ke
xk ≤ s4 + √ ρ3 + 4c22 ρ3 δt . (65)
2c11 c21 c21 ω 1 − e− 2c12 ω
√ δt
c e 2c12
When ( cc22 s4 + √ 22 H) 1 ≤ 4c22 ρ3 ln ω
21 c21 ω −
δt ω , namely,
1−e 2c12


 c22 s + c
√ 22 H
δt 
c21 4 c21 e 2c12 − 1
ω ≥ ω2 = exp · δt , (66)
4c22 ρ3 e c12
δt
4c22 ρ3 r1 (ωc ) e c12 ln ω
ke
xk ≤ · δt · . (67)
c11 e 2c12 − 1 ω
To sum up, from (45), (51), (59), (66) and (67), when ω ≥ ω ∗ we have
   
1 ρ1 ρ2
ω ∗ = max 2ωc , √ s 1 (ω c ) + √ s 2 + s 3 , ω 1 , ω 2 , (68)
κ1 c11 c22
 
1 ln ω
kb
e(t)k ≤ T1 · , t ∈ ti−1 + 2c22 , ti , (69)
ω ω
ln ω
ke
xk ≤ T2 · , t ∈ [t0 , ∞), (70)
ω
where

c22 c22
T1 = s4 + √ ρ3 , (71)
c21 c21
δt
4c22 ρ3 r1 (ωc ) e c12
T2 = · δt , (72)
c11 e 2c12 − 1
are constants depend on ωc , the initial values, the uncertainty and the reference signals.
Appendix B: The Proof of Theorem 4.1
The proof is similar to that of Theorem 4.2. We only render the important variables form
so that can make comparison with RESO.
Make a transformation
   
ζ1 ω 2 e1
   
ζ=  
 ζ2  =  ωe2  .
 (73)
ζ3 e3 − f0

Define  
  −3 1 0
0 1  
Ac =  , Aζ =  
 −3 0 1 .
−ωc2 −2ωc
−1 0 0

For Ac , Aζ are Hurwitz matrixes, so there exist positive matrix P1 and P 2 , satisfying

AT
c P1 + P1 Ac = −I2 , c11 I2 ≤ P1 ≤ c12 I2 (74)
AT
ζ P 2 + P 2 Aζ = −I3 , c21 I2 ≤ P 2 ≤ c22 I2 , (75)
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 23

where c11 , c12 are the minimum and maximum eigenvalue of P1 , c21 , c22 are the minimum and
maximum eigenvalue of P 2 . Corresponding to (23), we have
 
 2  1 − 1
−1
ωc +5 1 2
2  
P1 =  4ωc ω2ω c 
, P 2 =  − 1 1 − 1 ,
1
2
c +1
 2 2 
2ωc2 4ωc3 −1 − 12 4
1 h 4 2
p i
4 + 6ω 2 + 1)(ω 4 + 2ω 2 + 1) ,
c11 = ω + 6ω + 1 − (ω (76)
8ωc3 c c c c c c

1 h p i
4 2 4 + 6ω 2 + 1)(ω 4 + 2ω 2 + 1) ,
c12 = ω c + 6ω c + 1 + (ω c c c c
8ωc3
c21 = 0.1966, c22 = 4.3372.

Define lyapunov function V1 (e x) = xeT P1 x


e, V 2 (ζ) = ζ T P 2 ζ, and
 
√ 1 ωc2 + 1
r1 (ωc ) = 2 2 + , (77)
2ωc2 4ωc3
s 2  2 2
1 ωc + 1
r2 (ωc ) = 2 + , (78)
2ωc2 4ωc3
p
a1 | (2ωc )2 + ωc4 ,
s1 (ωc ) = 10|e (79)

s2 = 10 3|e a1 |, (80)
a1 | · κ 2 ,
s3 = 10|e (81)
  
p √ ρ
ρ1 = max V1 (ex(t0 )), 2 c12 √ 2 r1 (ωc ) + κ1 r2 (ωc ) , (82)
c21
q √ √
ρ2 = max{ V 2 (ζ(t0 )), ( 2 + 1) c22 κ1 }. (83)

Likewise, if the parameters ω and ωc satisfy


  
1 ρ1 ρ2
ω > max 2ωc , √ s1 (ωc ) + √ s2 + s3 , (84)
κ1 c11 c22

then ke
xk, kζk are bounded:
ρ ρ
xk ≤ √ 1 ,
ke kζk ≤ √ 2 . (85)
c11 c21
 
For x
e= x1 − x∗ and transformation (73), so x, eb are bounded.
x2 − ẋ∗
Make a transformation:
   
ζ1 ω 2 e1
   
ζ=  
 ζ 2  =  ωe2  .
 (86)
ζ3 e3
24 GUO JIANXIN · XUE WENCHAO · HU TAO

From (16), (86), we obtain


      
x
ė1 0 1 0 0 0 x
e1 0
      
x   2 2 2
−1  e   0 
 ė2   −ωc −2ωc −ωc /ω −2ωc /ω x 2  
˙      
ζ  =  0 0   ζ1  
   + 
 1  0 −3ω ω  0 , (87)
˙      
ζ   0 0 −3ω 0 ω     
 2    ζ2   0 
˙ζ 0 0 −ω 0 0 ζ3 η4
3

where
 
2 ωc2 2ωc
η4 = − ωc x e1 + 2ωc xe2 + 2 ζ 1 + ζ + ζ3 e a1
ω ω 2
c c 1 a1 + f˙0 .
+(ω 2 v + 2ωc v̇ + v̈ − ω 2 x∗ − 2ωc x∗ )e 2 (88)

Comparing (73) and (86), we get


q q
2 2 2
kζk = ζ 1 + ζ 2 + ζ 3 = ζ12 + ζ22 + (ζ3 + f0 )2
√ q
≤ 2 · ζ12 + ζ22 + ζ32 + f02

≤ 2(kζk + |f0 |).
ρ
For ζ and f0 are bounded, and kζk ≤ √2 ,
c21
|f0 | ≤ κ1 , so ζ is bounded. Define
 
√ ρ
kζk ≤ ρ3 , ρ3 = 2 √ 2 + κ1 . (89)
c21
From (88), η4 is bounded in the continuous points in f0 . Define kη4 (t)k ≤ κ3 , t 6= ti , i =
1, 2, · · · , here κ3 = κ3 (ωc , ρ1 , ρ2 , v, v̇, v̈, Ar , D), namely, κ3 is a constant, which depends on ωc ,
the initial values, uncertainty, and the reference signal.

Define s4 = 67κ3 , so similar to the proof of RESO in Step 2, we can prove, when
  
 2c22 ln ω 2c22

 max ω δt = , δt ≤ ,
ω1 = ω e (90)

 2c22
 0, δt > ,
e
and then whenω > ω1 , we have
2c22 ln ω
< δt . (91)
ω
For i ≥ 0, we have
 √ 
1 c22 c22 ∗
kek ≤ s4 + √ ρ3 , t ∈ [ti , ti+1 ), (92)
ω c21 c21
where
∗ 2c22 ln ω
ti = ti + < ti+1 , (93)
ω
ACTIVE DISTURBANCE REJECTION CONTROL FOR PMLM SERVO SYSTEM 25

and
 √  δt 
r1 (ωc ) c22 c22 1 e 2c12 ln ω
ke
xk ≤ s4 + √ H + 4c22 ρ3 δt · , t ∈ [t0 , ∞). (94)
2c11 c21 c21 ω 1 − e− 2c12 ω

√ δt

√c22 H) 1 e 2c12 ln ω
So when ( cc22
21
s4 + c21 ω ≤ 4c22 ρ3 −
δt · ω , i.e.,
1−e 2c12


 c22 s + √c22 H
δt 
c21 4 c21 e 2c12 − 1
ω ≥ ω 2 = exp · δt , (95)
4c22 ρ3 e c12
δt
4c22 ρ3 r1 (ωc ) e c12 ln ω
ke
xk ≤ · δt · . (96)
c11 e 2c12 − 1 ω
Summing up the above (90), (92), (93), (95), (96), we have when
   
1 ρ ρ
ω ∗ = max 2ωc , √ 1 s1 (ωc ) + √ 2 s2 + s3 , ω 1 , ω 2 , (97)
κ1 c11 c22

then when ω ≥ ω ∗ ,
 
1 ln ω
kb
e(t)k ≤ T 1 · , t ∈ ti−1 + 2c22 , ti , (98)
ω ω
ln ω
ke
xk ≤ T 2 · , t ∈ [t0 , ∞), (99)
ω
where

c22 c22
T1 = s4 + √ ρ3 , (100)
c21 c21
δt
4c22 ρ3 r1 (ωc ) e c12
T2 = · δt (101)
c11 e 2c12 − 1
are constants which depend on ωc , initial values, uncertainty, and the reference signals.
Appendix C: The Proof of T1 ≤ T1 and T2 ≤ T2
Comparing (41) with (89), (40) with (83), we have ρ3 ≥ ρ3 . From (40), (88) and kη4 (t)k ≤
κ3 , kη2 (t)k ≤ κ3 , t 6= ti , i = 1, 2, · · · , we have κ3 > κ3 . Hence, s4 > s4 . Because c22 > c22 and
c22 c22
c21 = 22.0610 > c21 = 5.8284, comparing (71) and (100), we have T1 ≤ T1 .
Besides, for ρ3 ≥ ρ3 and c22 > c22 , comparing (72) with (101), we have T2 ≤ T2 .

View publication stats