actuators

Article
A Novel Control Method for Permanent Magnet Synchronous
Linear Motor Based on Model Predictive Control and Extended
State Observer
Zhuobo Dong, Zheng Sun * , Hao Sun, Wenjun Wang and Xuesong Mei

School of Mechanical Engineering, Xi’an Jiaotong University, Xi’an 710054, China;

dongzhuobo@stu.xjtu.edu.cn (Z.D.); sunhao000408@stu.xjtu.edu.cn (H.S.)
* Correspondence: zheng.sun@xjtu.edu.cn

Abstract: Permanent magnet synchronous linear motor (PMSLM) is widely used to meet the require-
ment of high dynamic accuracy positioning, such as in machine tools and devices of semiconductor
manufacturing. A new 2-DOF control structure is proposed in this paper to improve the dynamic per-
formance of the positioning servo system with PMSLM. Aiming at the position tracking performance,
a control algorithm based on the model predictive control (MPC) is developed with position and
speed as the feedback state variables. In addition, an extended state observer (ESO) is designed for
the rejection of various disturbances, which are not involved in the control model and are regarded
as the lumped disturbance to be estimated and compensated by the ESO. The experimental results
show that, compared with the commonly used PPI controller (proportional position controller and
proportional–integral speed controller), the proposed method enhances the position bandwidth and
servo stiffness effectively.

Keywords: PMSLM; MPC; ESO; dynamic performance; anti-disturbance performance

1. Introduction
Citation: Dong, Z.; Sun, Z.; Sun, H.;
Wang, W.; Mei, X. A Novel Control The precision servo feed system is a relevant prerequisite for precision machining.
Method for Permanent Magnet It is necessary to improve the closed-loop bandwidth of the positioning servo control
Synchronous Linear Motor Based on since both the tracking error and the servo stiffness are dominated by the bandwidth
Model Predictive Control and directly. A servo system with high bandwidth can improve the quality of machined
Extended State Observer. Actuators parts and reduce the machining time. However, for the commonly used feed drives with
2024, 13, 34. https://doi.org/ ball screws or rack pinions, the bandwidth of the position loop is limited by the first-
10.3390/act13010034 order resonance of these mechanical transmission systems included in the control loop [1].
Academic Editor: Guanghong Yang
Compared with the traditional feed drives with transmission systems, the “direct drive”
with the permanent magnet synchronous linear motor (PMSLM) eliminates the influence
Received: 19 December 2023 of mechanical resonance on the controller fundamentally and improves the servo accuracy
Revised: 9 January 2024 and dynamic response performance effectively.
Accepted: 12 January 2024 The feed drives are normally cascade controlled by a proportional (P) controller in the
Published: 14 January 2024 outer position loop and a proportional–integral (PI) controller in the speed loop (PPI). The
PPI controller is easy to tune and has high robustness, which meets the requirements of
most industrial applications. However, its bandwidth has a lower upper limit to avoid large
overshoots and oscillations. The PI-controlled speed loop can be regarded as a large inertia
Copyright: © 2024 by the authors.
delay block in series. The integral term improves the performance of disturbance rejection,
Licensee MDPI, Basel, Switzerland.
This article is an open access article
but the additionally introduced pole in the inner loop damages the stability reserve of the
distributed under the terms and
outer position control loop.
conditions of the Creative Commons As the substitution of cascade control, the current servo control methods mainly
Attribution (CC BY) license (https:// include sliding mode control, active disturbance rejection control, adaptive control based
creativecommons.org/licenses/by/ on servo parameter estimation, robust control, etc. Sliding Mode Control (SMC) is one
4.0/). of the most robust algorithms with a low-accuracy system model and is insensitive to

Actuators 2024, 13, 34. https://doi.org/10.3390/act13010034 https://www.mdpi.com/journal/actuators

Actuators 2024, 13, 34 2 of 13

internal and external disturbances. Extensive research has focused on the use of new
sliding mode control structures [2–6] and the composite control system [7–9] combining
SMC and other control methods. Most methods based on SMC have many parameters.
The coupling between parameters makes SMC difficult to adjust and limits the industrial
application. Active Disturbance Rejection Control (ADRC) can provide high servo stiffness
for linear motors due to its property of active disturbance rejection [10]. The current research
mainly focuses on improving the traditional ADRC with the compensation method [11].
Although it has a high disturbance rejection performance, parameter tuning is still a difficult
problem when it is applied. Servo parameter identification and adaptive control [12], robust
control [13], and other methods are also studied in positioning systems with linear motors.
However, the control method combining parameter identification and adaptive control
has a large amount of online calculations and requires high computational performance of
the controller. Moreover, the synthesized order of H ∞ controllers is normally too high to
realize in industry.
At present, the research and application of Model Predictive Control (MPC) in the
field of motor control have become more extensive. Kwon et al. present the principle of the
MPC and analyze the stability [14]. Based on the cascade structure some researchers adopt
the MPC to break through the performance limitations, such as speed/current integrated
MPC controllers [15–18]. The essential difference between electromagnetic and mechanical
characteristics differentiates the mechanical and electrical time constants greatly. It is not
suitable for the overall control. Therefore, some researchers study to control the speed
using the MPC separately. Li et al. proposed an improved predictive function for the speed
regulation of the servo system with PMSM, which effectively improved the disturbance
performance [19]. Wang et al. proposed an MPC controller combined with a Kalman filter,
which improved the tracking performance and disturbance rejection of system speed [20].
Shao et al. presented a generalized predictive controller with a high-order sliding-mode
observer [21]. Yao et al. proposed a new speed nonlinear direct predictive control method
for PMSM [22]. Consequently.
The controller design should meet the requirements of the fast response to track the
reference and strong robustness against the inner and outer disturbance. The traditional
controlled servo systems cannot meet these two demands at the same time, so the two-
degrees-of-freedom (2-DOF) control structure is applied with one controller located in
the forward channel for the tracking performance and another controller or observer
located in the feedback channel for the disturbance. Li et al. proposed a 2-DOF H ∞ robust
speed control method for the servo system [23]. The method has a good speed tracking
performance and a strong robustness against load disturbance and parameter perturbation,
but the design of the weight function is cumbersome. Chen et al. formed a 2-DOF controller
with a fractional order Proportional-Derivative (PD) controller and an Extended State
Observer (ESO) and applied it to the speed control of PMSM [24]. A 2-DOF controller
proposed by Yang et al. combines a PI controller and a Kalman filter and shows a good
balance between the disturbance rejection and tracking performance [25]. Until now, most
research focused on the speed control of PMSM. Although the bandwidth of the position
loop can also be improved with the enhancement of the speed performance, the direct
controller design for the position control is seldom seen.
This paper introduces the MPC into the positioning servo system with PMSLM and
proposes a high-performance positioning servo control algorithm based on the MPC and
ESO. Using the MPC, a forward controller of the servo system is designed to improve
the dynamic tracking performance of the system. Aiming at disturbances such as load
force variation, mass perturbation, and nonlinear thrust fluctuation, an ESO is designed,
which can observe and compensate in the form of lumped disturbance. The overall control
structure can be depicted in Figure 1.
PEER REVIEW 3 of
Actuators 2024, 13, 34 3 of 13

Figure 1. Structure of the1.proposed

Figure control
Structure of the method.
proposed control method.

The remainder of this paper is as follows. The MPC for PMSLM is designed and
The remainder of this
analyzed paper2. is
in Section as follows.
In Section The
3, the ESO for MPC for PMSLM
the disturbance is designed
is presented, and its pole and a
placement and noise sensitivity are analyzed. The experimental verification is shown in
alyzed in Section Section
2. In 4.Section 3, the ESO for the disturbance is presented, and its po
Section 5 concludes with comments.
placement and noise sensitivity are analyzed. The experimental verification is shown
Section 4. Section 2.
5 Model Predictive Control
concludes with comments.
2.1. Linear Motor State Equation
Neglecting the outer disturbance force, the linear motor can be abstracted as a one-
2. Model Predictive
massControl
system. Its dynamic equilibrium equation can be formulated as
.. .
2.1. Linear Motor State Equation m·x = f −d·x (1)

Neglecting the outer

where disturbance
x is the motor position,force,
m is thethe
totallinear
mass ofmotor can
the motor be and
mover abstracted as a on
load, d is the
viscose damper
mass system. Its dynamic from the guideway,
equilibrium equationand f is
canthe be
electromagnetic
formulated force,
aswhich can also be
regarded as the control output. It can be expressed in matrix form with the introduction of
state vector X.
𝑚 ⋅ 𝑥X. == 𝑓A −· X𝑑+⋅B𝑥 · f (
c c
" # " # " #
where x is the motor position, m with is the total
x mass 0of 1the motor 0 mover and load, (2) d is th
X= . Ac = Bc =
viscose damper from the guideway, and f xis the electromagnetic 0 −md 1
m force, which can also b

regarded as the controlSinceoutput.

the servoIt
system
can isbeoperated and controlled
expressed in matrixdiscretely,
formthewith
continuous state
the introductio
equation should be discretized through first-order approximation
of state vector X.
A = e Ac Ts ≈ I + Ac Ts
𝑋 = 𝐴R T⋅ 𝑋 + 𝐵 ⋅ 𝑓
B = 0 s e Ac t dt · Bc ≈ ITs · Bc
(3)

𝑥 0 1 0
servo𝑋control
with
where Ts is the = cycle.𝐴 = 0 𝐵 = (
𝑥
According to Equation (3), the discrete state space representation can be formulated as

X k +1 = A · X k + B · f k
Since the servo system is operated" and # controlled
" discretely,
# " # the continuous sta
xk 1 Ts 0 (4)
equation should be discretized through
with Xk = first-order
v
A= approximation
0 1− dT
B = Ts
k m s m

The subscript k means𝐴the

=k-th
𝑒 present𝐼moment.
+𝐴 𝑇
(
𝐵= 𝑒 𝑑𝑡 ⋅ 𝐵 𝐼𝑇 ⋅ 𝐵

where 𝑇 is the servo control cycle.

According to Equation (3), the discrete state space representation can be formulate
as
Actuators 2024, 13, 34 4 of 13

2.2. Design of the MPC

Equation (4) transforms the state at the moment k to the state of k + 1. Multiple using
Equation (4), the state of moment k + n can be predicted theoretically as follows.

Xk+2 = A( AXk + Buk ) + Buk+1

= A2 Xk + ABuk + Buk+1
Xk+3 = A3 Xk + A2 Buk + ABuk+1 + Buk+2 (5)
n
Xk+n = An Xk + ∑ Ai−1 Buk+n−i
i =1

Denoting the prediction step of the system as n p , the control step as nc , the state from
moment k + 1 to the moment k + n p and k + nc can be predicted. Normally, the control step
should not be beyond the prediction step nc ≤ n p , and the control output is unchanged
beyond the control step, f k+i = f k+nc with i = nc + 1, · · · , n p . We summarize all the
h iT
predicted states in a vector Z = Xk+1 Xk+2 · · · Xk + n p and all control outputs
 T
in another vector F = f k f k+1 · · · f k+nc −1 , the state prediction can be compactly
expressed as follows.

Z = M · Xk + Π · F
B 0 ··· 0
 

AB B .. ..
.
   
A  . 
.. ..
A2
 
   . . .. 0
 (6)
with M = 
 
Π =  n c −1
 . 
.. A n c −2 B
A B B
 
 .  ··· 
An p
 .. .. .. .. 
. .
 
 . . 
A n p −1
B A n p −2 B ··· A n p −nc B

The basic idea of the MPC of the servo system is to find a suitable F to minimize the
object function, which describes the difference (tracking error) between the predicted state
Z and the reference state Zre f . For the servo control of PMSLM, the reference position
and velocity are generated by the CNC system. Its current and future values can be easily
obtained from the buffer cache. In general, to prevent the oversized control output, the
electromagnetic force F should also be considered in an objective function by weight. So,
the object function can be constructed as follows.
 T  
J = Z − Zre f WZ Z − Zre f + F T WF F

F = argmin J
F
  (7)
wx 0 h i
with W Z = 
 wv 
 WF = w f
.. nc ×nc
0 .
2n p ×2n p

The weight wx and wv dominate the influence of position and speed tracking errors,
respectively, and w f is weight to constrain the electromagnetic force.
Take Equation (6) into Equation (7), the minimized value of J can be obtained through
the following equation.
∂J
∂F= 2Π T Wz ( MXk + ΠF − Z ) + 2WF F = 0

−1 T (8)
⇒ F = Π T WZ Π + WF Π WZ ( Z − MXk )
 Actuators 2024, 13, 34 5 of 13

The obtained F is a vector with the length nc , but only the first value of F is outputted
as the current command to the motor. For the implementation of the MPC in the hard-
ware, most coefficients can be calculated offline, such as in Equation (8), the term before
( Z − MXk ) is calculated offline since all the model parameters and weights are given.

2.3. Stability Analysis of the MPC

The control output can be reformed as the multiplication of the state error and the
coefficient K, which is defined as follows.
Actuators 2024, 13, x FOR PEER REVIEW 5 of 13
F = K ( Z − MXk )

−1 T (9)
· · · 0 1×n Π T Wz Π + WF Π WZ
 
with K = 1 0
c
Bringing it into Equation (4), we can obtain a close-loop recurrence relation.
Bringing it into Equation (4), we can obtain a close-loop recurrence relation.
𝑋𝑘+1 = 𝐴𝑋𝑘 + 𝐵𝐾(𝑍 − 𝑀𝑋𝑘 )
(10
Xk+1 = AX =k (𝐴 − 𝐵𝐾𝑀)
+ BK ( Z − MX∙ 𝑋k ) + 𝐵𝐾 ∙ 𝑍
(10)
( A − BKM re f
According to the stability=criterion, k + BK · Zsystem is asymptotically stable only
)· Xdiscrete
the
if allAccording
eigenvalues of stability
to the A−BKMcriterion,
are located in the unit
the discrete system circle.
is asymptotically stable only if
Assuming
all eigenvalues ofthat
A − the
BKMweights
are locatedof each
in thestep
unitare the same, there are three weights or pa
circle.
rameters
Assumingthat need
that tothebe tuned, of
weights namely,
each step𝑤 are𝑤 ,the 𝑤 correspond
andsame, to the
there are three position,
weights or speed
and motorthat
parameters force,
needrespectively. They describe
to be tuned, namely, wx wv , and different physical
w f correspond values
to the and speed,
position, cannot be di
and motor force, respectively. They describe different physical values
rectly compared. To find out the suitable order of magnitude of three weights, their unitsand cannot be directly
compared.
should be To find out
unified theAs
first. suitable order ofthe
a reference, magnitude
weight of of three weights, their units
the electromagnetic 𝑤 is set to
should
force
be unified first. As a reference, the weight of the electromagnetic force w f is set to be 1,
be 1, the speed is the integral of the acceleration, which is the ratio of the force and the
the speed is the integral of the acceleration, which is the ratio of the force and the mass.
mass. Therefore, the speed weight 𝑤 should be set based on the value of 𝑚⁄𝑇 . Similarly
Therefore, the speed weight wv should be set based on the value of m/Ts . Similarly, the
the position
position is theisintegral
the integral of velocity,
of velocity, so 𝑤 beshould
so wx should set based beonsetthe
based
valueon of the
m/Tvalue
2 of 𝑚⁄𝑇 .
s.
It is difficult to solve the eigenvalue of A − BKM analytically. Therefore, we calculatecalculate
It is difficult to solve the eigenvalue of A-BKM analytically. Therefore, we
the maximum
the maximumeigenvalueeigenvalue numerically
numerically by selecting
by selecting the weights
the weights withinwithin
a certaina range.
certainSet
range. Se
𝑛p =
n = 20,
20,nc𝑛==1, 1, andandwf 𝑤= 1,=the
1, the
norm norm
of theofmaximum
the maximum eigenvalue
eigenvalue is plottedis plotted
in Figure in2Figure 2
by w𝑤x from
by from 1 to to 200000
1 200, 000 × m/T× 𝑚/𝑇
2
s , and, w 𝑤 1 from
v from
and to 1001×to m/T100 × 𝑚/𝑇
s . The system parameters
. The are
system parameters
set
areassetthe
asvalues from the
the values testthe
from bench
testintroduced in Sectionin4.Section 4.
bench introduced

Figure2.2.Norm
Figure Normdistribution
distribution of maximum
of maximum eigenvalue
eigenvalue ofBKM.
of A − A-BKM.

It can be seen from Figure 2 that all the norms of eigenvalues are smaller than one
although for some parameters they are close to one. It means that the proposed MPC has
sufficient stability theoretically. However, in a real system, it is necessary to consider the
bandwidth limitation of the current control loop, the current saturation, the time delay
Actuators 2024, 13, 34 6 of 13

It can be seen from Figure 2 that all the norms of eigenvalues are smaller than one,
although for some parameters they are close to one. It means that the proposed MPC has
sufficient stability theoretically. However, in a real system, it is necessary to consider the
bandwidth limitation of the current control loop, the current saturation, the time delay, and
other factors, which damage the stability reasonably.

3. Extended State Observer

3.1. Construction of ESO
In the model of PMSLM, the disturbance forces, such as cutting force, are neglected.
In addition to that, many other factors, such as the changing of the load mass, static friction,
etc., also have a great impact on the control performance. The result of these factors can be
equivalented in the form of a lumped disturbance. Since the uncertainty of f d , it is difficult
to deal with in the MPC. In this section, an extended state observer is developed to estimate
the disturbance and improve the control performances.
Considering the lumped disturbance f d , the dynamic equation of PMSLM can be
reformulated as
..
mx = fe + fd (11)
In Equation (11), the viscose damper can further be neglected to simplify the derivation
procedure. As a new variable, the lumped disturbance force f d can be introduced into
the state equation, so the extended state equation disturbing force can be constructed
as follows.
. .
X e = Ae Xe + Be f + E· f d
       
x 0 1 0 0 0 (12)
.
with Xe =  x  Ae = 0 0 m1  Be = 1⁄m E = 0
     

fd 0 0 0 0 1

The lumped disturbance and its derivative cannot be measured directly, it can only
be estimated as a known quantity. For linear motors, the only directly measurable state is
the motor position. Therefore, the lumped disturbance can be estimated by observing the
actual position. The extended state observation equation is established as follows.
.̂
X e = Ae X̂e + Be f + GC ( Xe − X̂e )
   
x̂ g1 (13)
 .̂   
with X̂e =  x  G =  g2  C = 1 0 0
 

fˆd g3

In this paper, the values with hut are the estimated values. The vector G is the gain
vector including three gains for the observer. The vector C is the output vector denoting
that only the position can be directly measured and compared with the estimated value.
Similar to Equation (3), the disturbance observation is applied discretely. So, Equation (13)
is discretized through the Tyler expansion.

X̂e,k+1 = Aed X̂e,k + Bed f k + Gd (CX e,k − C X̂ e,k )

Ts2

T2
  2  
 
x̂k Ts
1 Ts 2ms g1 Ts + g2 2 
 .̂   2m  (14)
T2 
  
with X̂e,k =  x k  Aed =  Ts  B =  Ts  G =
0 1 ed d  g2 Ts + g3 2ms 

m  m
ˆf d,k 0 0 1 0 g3 Ts
Actuators 2024, 13, 34 7 of 13

3.2. Gains Tuning and Stability Analysis

The tuning and analysis in this section are still based on the continuous observation
process. Subtracting Equation (12) into Equation (13), the estimation error can be obtained.
. .
ξ = ( Ae − GC )ξ − E f d
 .         
ε1 0 1 0 g1  ε1 0 . (15)
.
0 m1  −  g2  1 0 0 ε 2  − 0 f d

⇒  ε2  =  0
.
ε3 0 0 0 g3 ε3 1

where ε 1 , ε 2 and ε 3 denote the estimation error of position, speed, and disturbance, respectively.
The observer gains were designed through the pole placement method. If the obser-
vation process is stable, the eigenvalues of Ae − GC must be located in the left half plane
of the s-plane. We can set that when this third-order matrix has a triple negative real pole
−ω0 , the characteristic polynomial of Ae − GC should on one hand satisfy the following
equation with the cut-off frequency ω0 .

(s + ω0 )3 = s3 + 3ω0 s2 + 3ω02 s + ω03 (16)

On the other hand, the eigenvalue is calculated through the determinant.

s + g1 −1 0
g3
det(sI − Ae + GC ) = g2 s − /m = s3 + g1 s2 + g2 s +
1 (17)
m
g3 0 s

Comparing the coefficients of Equation (16) and Equation (17), the gain vector can be
valued as follows.
g1 = 3ω0
g2 = 3ω02 (18)
g3 = mω03

3.3. Noise Sensitivity Analysis of ESO

The noise in the real system cannot be avoided. The observer with unsuitable param-
eters may enlarge the noise and bring the whole system into instable. Therefore, noise
sensitivity will be analyzed in this section. Introducing the position measuring noise η into
Equation (13), the observation equation can be extended as
 .̂ 
x
          
0 1 0 x̂ 0 g1 x+η x̂
 ..ˆ   1   .̂  +  1  f +  g  1
  .   .̂ 
 x  = 0
  0 m  x   m  e  2 0 0  x  −  x  (19)
.̂
fd 0 0 0 fˆd 0 g3 fd fˆd

Subtracting Equation (12) from Equation (19), the estimation error can be obtained
as follows
 .           
ε1 0 1 0 g1 ε1 0 g1
 .       .
ε2  = 0 0 m1  −  g2  1 0 0 ε 2  − 0 f d +  g2 η
  
(20)
 
.
ε3 0 0 0 g3 ε3 1 g3

With the error Equation (20) and the value of G determined by Equation (18), we can
obtain the transfer function from η to ε 3

ε 3 (s) mω03 s2
= (21)
η (s) ( s + ω0 ) 3
 Actuators 2024, 13, 34 8 of 13

The amplitude of the transfer function can be calculated as

ε 3 ( jω ) mω 2 ω03
= (22)
η ( jω ) ω 2 + ω02 )3/2

It is a monotonic increasing function about the cut-off frequency ω0 . Increasing ω0 ,

the amplitude from η to ε 3 is enlarged [26], which means that the jitter equivalent to the
estimated disturbance is also increased.

4. Experimental Verification
4.1. Setup of Test Bench
The test bench for the verification was built with a direct drive platform from CNBHC,
Actuators 2024, 13, x FOR PEER REVIEW 8 of 1
as shown in Figure 3. The applied linear motor type was TMLA0070-095-000, and its
specific parameters are listed in Table 1.

Figure 3. Test bench applied for the experiments.

Figure 3. Test bench applied for the experiments.
Table 1. TMLA0070-095-000 linear motor parameters.
Table 1. TMLA0070-095-000 linear motor parameters.
Parameters Parameter Value Parameters Parameter Value
Parameters
R/Ω
Parameter
2.8
Value Parameters
Ld , Lq /mH
Parameter Value
6.8
ImaxR /A
/Ω 9.5 2.8 Ld , Lq /mH
IN /A 2.5 6.8
Udc /V 300 Km /V·m−1 ·s−1 21.4
I max
Ke /N ·A−/A
1 32 9.5 m/kg I N /A 6 2.5
U dc /V 300 K m /V·m−1·s−1 21.4
In the test bench, the linear motor was controlled by a PMAC controller with the
Ke /N·Afrom
type of CK3M
−1
OMRON. The 32 motor was driven bym/kg a Power Block from CDHD 6
through the Pulse Width Modulation (PWM) signals. The current controller and PWM
algorithm were
In the testimplemented
bench, the linearin CK3M
motorwith
wasa control frequency
controlled of 16 kHz.
by a PMAC The power
controller with the typ
block received only the switch signals for the inverter, whose frequency was the same as
of CK3M from OMRON. The motor was driven by a Power Block from CDHD throug
the current control frequency 16 kHz. The current controller applied in CK3M was a PI
the Pulse Width Modulation (PWM) signals. The current controller and PWM algorithm
controller (k i,p (1 + k i,i /s)). After the auto-tuning, the control gains were parameterized as
were implemented in CK3M with a control frequency of 16 kHz. The power block receive
k i,p = 35 V/A k i,i = 411 s−1 . The bandwidth of the closed current loop was over 1 kHz.
onlyThe thepresented
switch signals for thealgorithm
servo control inverter,was
whose frequency
implemented in was
CK3M thethrough
same as thethe current con
Power
trol frequency
PMAC IDE with 16 kHz. The
a control currentofcontroller
frequency appliedthe
8 kHz. Following in trial
CK3M andwas a PI the
tuning, weight (𝑘 , (1 +
controller
𝑘 , /𝑠) ). After
coefficients theproposed
for the auto-tuning,
MPC are thelisted
control gains
in Table 2. were parameterized
The prediction as20𝑘 and
step n p is , = 35 𝑉⁄
the = 411 step
𝑘 , control s . The
nc isbandwidth
1. With these parameters,
of the the linear
closed current loopmotor
was has
overa 1good
kHz.dynamic
The presented servo control algorithm was implemented in CK3M through th
Power PMAC IDE with a control frequency of 8 kHz. Following the trial and tuning, th
weight coefficients for the proposed MPC are listed in Table 2. The prediction step 𝑛 i
20 and the control step 𝑛 is 1. With these parameters, the linear motor has a good dy
Actuators 2024, 13, 34 9 of 13

performance, rare current saturation, and is robust against the parameter perturbation. For
comparison, a conventional PPI controller was also implemented in the CK3M. The control
gain can also be seen in Table 2.

Table 2. Control parameters of the MPC and PPI.

MPC PPI
Weight Coefficient Value Control Gain Value Unit
wx 35, 000 × m/T2s k x,p 300 rad/s
Actuators 2024, 13, x FOR PEER REVIEW wv 10 × m/Ts k v,p 240 As/m 9 of 13
wf 1 k v,i 200 rad/s

4.2.
4.2. Experimental
Experimental Verification
Verification of
of Tracking
TrackingPerformance
Performance
The
The setup
setup response
response isis applied
applied toto test
test the
the tracking
tracking performance
performance when
when thethe position
position
commends
commends stepwise changes from the initial position of 0 mm to 0.1 mm. For P-PI control,
stepwise changes from the initial position of 0 mm to 0.1 mm. For P-PI control,
the
the MPC
MPC and
and MPC
MPC ++ ESO
ESO with
with different
different poles,
poles, the
the experimental
experimental results
results are
are shown
shown and
and
compared in Figure
compared in Figure 4. 4.

Figure4.
Figure 4. Step
Step response
response of
ofPPI,
PPI,MPC,
MPC,MPC
MPC++ ESO.
ESO.

ItIt is
is obviours
obviours thatthat thethe tracking
tracking performance
performance of of the
the MPC
MPC is is better
better than
than thatthat of
of PPI
PPI
control. The settling time, in which the feedback position drives in the
control. The settling time, in which the feedback position drives in the 97% of step refer- 97% of step reference,
reduced
ence, reducedfrom 10.3fromms 10.3of PPI
ms ofover
PPI56%
overto56% 4.5 ms of the
to 4.5 ms ofMPC. Due toDue
the MPC. the to
existence of static
the existence of
friction in the guide,
static friction in the the MPCthe
guide, without ESO has ESO
MPC without a certain
has steady-state error, so anerror,
a certain steady-state additional
so an
anti-disturbance method is necessary.
additional anti-disturbance method The introducedThe
is necessary. ESOintroduced
eliminates the ESOsteady-state
eliminateserror the
in step response. But the overshoot can also be seen in Figure 4
steady-state error in step response. But the overshoot can also be seen in Figure 4 whenwhen the pole value of the
ESO increases.
the pole value The main
of the ESO reason is thatThe
increases. various
maindisturbances
reason is that andvarious
uncertainties in the real
disturbances and
system
uncertainties in the real system make the ESO estimate all these factors in theWith
make the ESO estimate all these factors in the form of lumped disturbance. formthe of
increase
lumped in the cut-off frequency
disturbance. (the placed
With the increase in thepole of ESO),
cut-off the high-frequency
frequency (the placed pole components
of ESO),
of
thethe disturbance are
high-frequency involved inofthe
components theestimated
disturbance results, which enlarge
are involved in the the instantaneous
estimated results,
change in compensation output, consequently. Although a slight
which enlarge the instantaneous change in compensation output, consequently. Although overshoot can be seen
here, the motor position converges to the steady-state value in a short
a slight overshoot can be seen here, the motor position converges to the steady-state value time, so it can still be
considered that the tracking performance is basically dominated by
in a short time, so it can still be considered that the tracking performance is basically dom-the MPC.
inated
4.3. by the MPC.
Experimental Verification of Anti-Disturbance Performance
The anti-disturbance
4.3. Experimental performance
Verification is tested by
of Anti-Disturbance applying a stepwise disturbance current
Performance
of 2.5 A in front of the current loop when the motor is in a steady state with the command
Theof
position anti-disturbance performance
0 mm. The experimental is tested
results by applying
of PPI a stepwise
control and the MPC disturbance cur-
with different
rent of 2.5 A in front of the current loop when the motor is in a steady state with
cut-off frequencies of ESO are shown in Figure 5a. The estimated disturbances as the output the com-
mand
of position
the ESO of 0 mm.
are shown The experimental
in Figure 5b. results of PPI control and the MPC with differ-
ent cut-off frequencies of ESO are shown in Figure 5a. The estimated disturbances as the
output of the ESO are shown in Figure 5b.
Actuators 2024, 13, 34 Actuators 2024, 13, x FOR PEER REVIEW 10 of 13 10 of 13

(a) (b)

Figure 5. (a): disturbance

Figure 5.rejection of MPC rejection
(a): disturbance + ESO and P-PI; +(b):
of MPC ESOestimated
and P-PI;disturbance force
(b): estimated from ESO.
disturbance force from
ESO.
Generally, the proposed control method shows a better disturbance rejection than the PPI
control. The ESOs with differentthe
Generally, poles estimate
proposed disturbance
control currenta as
method shows the disturbance
better lumped disturbance
rejection than the
and converge valuePPI of 83 N. The
control. The higher
ESOs withthe cutoff
differentfrequency is set, the
poles estimate quicker the
disturbance ESO output
current as the lumped
converges. Consequently,
disturbance both
andthe maximal
converge position
value of 83 N.error
The and
highersettling timefrequency
the cutoff decrease iswith
set, the quicker
increase in the cut-off frequency.
the ESO output The results Consequently,
converges. of different control
both methods
the maximal are listed in Table
position error 3and
for settling
a better comparison. Compared with the ESO with the cutoff frequency of 300 rad/s, the ESO control
time decrease with the increase in the cut-off frequency. The results of different
methods are
with the cut-off frequency of listed in Table
700 rad/s 3 fora asignificant
shows better comparison.
enhancement Comparedof thewith the ESO with the
disturbance
cutoff frequency of 300 rad/s, the ESO with the cut-off
performance with the improvement of the maximal error and the settling time of 17.8 frequency of 700 rad/s
µmshows a
and 35.7 ms from PPI to 11.6 µm and 18.1 ms, respectively. For the further increase in the of the
significant enhancement of the disturbance performance with the improvement
cut-off frequency,maximal error and the settling
the anti-disturbance time of 17.8
performance isμm
stilland 35.7 ms from
improved PPI toa11.6
but with μm and 18.1
limited
decline from 11.6 ms,
µm respectively.
and 18.1 ms Fortothe
10.0further
µm and increase in the
12.8 ms, cut-off frequency, the anti-disturbance
respectively.
performance is still improved but with a limited decline from 11.6 μm and 18.1 ms to 10.0
μm and
Table 3. Position error and12.8 ms, respectively.
settling time of the MPC + ESO and PPI.

Table 3. Position error and settling

Controller time ofError/µm
Position the MPC + ESO and PPI.
Settling Time/ms
PPI Controller Position Error/μm
17.8 Settling Time/ms
35.7
MPC + ESO (ω0 = 300 rad/sPPI) 12.9 17.8 35.8 35.7
MPC + ESO (ωMPC
0 = 700
+ ESO (𝜔0 )= 300 rad⁄s)
rad/s 11.6 12.9 18.1 35.8
MPC + ESO (ωMPC
0 = 1100
+ ESOrad/s
(𝜔0 )= 700 rad⁄s) 10.0 11.6 12.8 18.1
MPC + ESO (𝜔0 = 1100 rad⁄s) 10.0 12.8
Considering the noise in the feedback signal, the cutoff frequency of the ESO cannot be
Considering
set with an excessively high value. theFigure
noise in
6 the feedback
shows signal,
the effect of the
thecutoff
noisefrequency of thepole
with different ESO cannot
be set with an excessively high value. Figure 6 shows the effect of the noise with different
places on the estimation output when the motor is in the steady state of 0 mm command
pole places on the estimation output when the motor is in the steady state of 0 mm com-
position. When the pole of the ESO is located at −700 rad/s, the jitter of the estimated
mand position. When the pole of the ESO is located at −700 rad/s, the jitter of the estimated
disturbance is about
Actuators 2024, 13, x FOR PEER REVIEW 0.05 N, but when it is increased to −3000 rad/s, the 11 jitter
of 13increases
disturbance is about 0.05 N, but when it is increased to −3000 rad/s, the jitter increases
significantly to about 1 N. The trend matches the theoretical analysis of Equation (22).
significantly to about 1 N. The trend matches the theoretical analysis of Equation (22).

Figure 6.
Figure 6. Jitter
Jitterofofthe estimated
the disturbance.
estimated disturbance.

4.4. Experimental Verification in the Frequency Domain

The effectiveness of the presented control method can be seen not only in the time
domain but also in the frequency domain, as shown in Figure 7. The logarithmic frequency
sweep signal is generated as the position reference with an amplitude of 0.03 mm and a
 sweep signal is generated as the position reference with an amplitude of 0.03 mm and a
frequency range from 1 Hz to 300 Hz. The close-loop bandwidth is normally defined as
the frequency when the magnitude once attenuated to −3 dB. The bandwidth of the posi-
tioning system controlled by the PPI method is about 72 Hz. Using the presented control
Actuators 2024, 13, 34 11 of 13
method, the bandwidth can be increased by 94.4% to over 140 Hz.
As mentioned above, the frequency response for the position tracking is mainly dom-
inated by the MPC,
4.4. Experimental but the
Verification cut-off
in the frequency
Frequency Domainof the observer has a certain impact on it.
Increasing the cut-off frequency from 300 to 700, the property of magnitude in the middle-
The effectiveness of the presented control method can be seen not only in the time
frequency
domain range
but also in of
theabout 40 Hz
frequency will be
domain, better with
as shown less7.attenuation,
in Figure which
The logarithmic means less
frequency
transfer distortion in this frequency range. But when further increasing
sweep signal is generated as the position reference with an amplitude of 0.03 mm the cut-off
and fre-
aquency, a peak
frequency at about
range from 1135 HzHz willHz.
to 300 be obvious. This is bandwidth
The close-loop mainly caused by the amplify
is normally definedeffect
ofthe
as thefrequency
ESO on the measurement
when the magnitude noise, as attenuated
once to −3 dB.
shown in Figure 7 green, and will damage
The bandwidth of the the
positioning system controlled by the PPI method is about 72 Hz. Using the
stable reverse. Therefore, choosing a suitable cut-off frequency of the ESO is relevant for presented
control method,
industrial the bandwidth can be increased by 94.4% to over 140 Hz.
applications.

Figure7.7.Close-loop
Figure Close-loopbode
bodeplot of of
plot thethe
MPC + ESOF
MPC andand
+ ESOF PPI.PPI.

As mentioned above, the frequency response for the position tracking is mainly
dominated by the MPC, but the cut-off frequency of the observer has a certain impact
on it. Increasing the cut-off frequency from 300 to 700, the property of magnitude in the
middle-frequency range of about 40 Hz will be better with less attenuation, which means
less transfer distortion in this frequency range. But when further increasing the cut-off
frequency, a peak at about 135 Hz will be obvious. This is mainly caused by the amplify
effect of the ESO on the measurement noise, as shown in Figure 7 green, and will damage
the stable reverse. Therefore, choosing a suitable cut-off frequency of the ESO is relevant
for industrial applications.

5. Conclusions
In this paper, a control method for the positioning servo system with PMSLM is
proposed. It belongs to the 2-DOF control structure and consists of an extended state
observer, which estimates and compensates for the lumped disturbance, and a model
predictive controller dealing with the tracking performance through the feedback position
and speed. Compared with the existing servo controllers, which mainly use the current
state and reference value, the MPC predicts the future state, compares with the reference
sequence, and solves the optimized control output. The enhanced tracking performance
is expected since more state information at different times is involved. The two modules
combine the control ideas of prediction, estimation, compensation, and improve the overall
performance of the positioning servo system effectively.
The tuning of the MPC and ESO can be carried out separately. Although the derivation
of the MPC and ESO is complex, only three parameters need to be tuned (the weight of
position error wx , speed error wv , and the cut-off frequency of ESO ω0 ). The experimen-
tal results show that, compared with the mainstream PPI cascade control structure, the
Actuators 2024, 13, 34 12 of 13

designed control method has higher servo bandwidth and servo stiffness. The implementa-
tion of the proposed control method has a relatively low computation requirement for the
hardware. It is suitable for application as a substitution for the standard PPI controller.

Author Contributions: Conceptualization, Z.S.; methodology, Z.D.; software, Z.D.; validation,

Z.D. and H.S.; formal analysis, Z.S.; investigation, Z.D.; writing—original draft preparation, Z.D.;
writing—review and editing, Z.S. and H.S.; visualization, Z.D.; supervision, X.M.; project administra-
tion, W.W.; funding acquisition, W.W. All authors have read and agreed to the published version of
the manuscript.
Funding: This research was funded by the National Key Research and Development Program of
China under grant number 2021YFF0500203 and funded by the National Natural Science Foundation
of China under grant number 51975461.
Data Availability Statement: The topic of this paper is not data research of data analyse. The CNBHC
dose not want to publish the experiment raw data.
Acknowledgments: The authors would like to thank CNBHC Co., Ltd. Chengdu, China for providing
the high-precision linear motor and Mo Sang from OMRON Industrial Automation (China) Co., Ltd.
Beijing branch for the technical support.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Pritschow, G. A comparison of linear and conventional electromechanical dives. CIRP Ann. Manuf. Technol. 1998, 47, 541–548.
[CrossRef]
2. Du, H.B.; Chen, X.P.; Wen, G.H.; Yu, X.H.; Lü, J.H. Discrete-Time Fast Terminal Sliding Mode Control for Permanent Magnet
Linear Motor. IEEE Trans. Ind. Electron. 2018, 65, 9916–9927. [CrossRef]
3. Jin, H.Y.; Zhao, X.M. Approach Angle-Based Saturation Function of Modified Complementary Sliding Mode Control for PMLSM.
IEEE Access 2019, 7, 126014–126024. [CrossRef]
4. Jiang, J.; Wang, M.; Tang, Y.; Zhou, C.; Li, L. Super Twisting-Based Position Control of PMLSM with Load Force Observer. In
Proceedings of the 22nd International Conference on Electrical Machines and Systems, Harbin, China, 11–14 August 2019; pp. 1–4.
[CrossRef]
5. Jin, H.Y.; Zhao, X.M.; Wang, T.H. Modified complementary sliding mode control with disturbance compensation for permanent
magnet linear synchronous motor servo system. IET Electr. Power App. 2020, 14, 2128–2135. [CrossRef]
6. Fu, D.X.; Zhao, X.M.; Zhu, J.G. A Novel Robust Super-Twisting Nonsingular Terminal Sliding Mode Controller for Permanent
Magnet Linear Synchronous Motors. IEEE Trans. Power Electr. 2022, 37, 2936–2945. [CrossRef]
7. Huang, Y.S.; Sung, C.C. Function-Based Controller for Linear Motor Control Systems. IEEE Trans. Ind. Electron. 2010, 57,
1096–1105. [CrossRef]
8. Chen, S.Y.; Chiang, H.H.; Liu, T.S.; Chang, C.H. Precision Motion Control of Permanent Magnet Linear Synchronous Motors
Using Adaptive Fuzzy Fractional-Order Sliding-Mode Control. IEEE-Asme Trans. Mechatron. 2019, 24, 741–752. [CrossRef]
9. Li, J.; Du, H.B.; Cheng, Y.Y.; Wen, G.H.; Chen, X.P.; Jiang, C.H. Position tracking control for permanent magnet linear motor via
fast nonsingular terminal sliding mode control. Nonlinear Dyn. 2019, 97, 2595–2605. [CrossRef]
10. Li, M.; Zhao, J.; Hu, Y.; Wang, Z. Active disturbance rejection position servo control of PMSLM based on reduced-order extended
state observer. Chin. J. Electr. Eng. 2020, 6, 30–41. [CrossRef]
11. Gao, W.; Li, S.; Wang, Y.F.; Wu, Z.S.; Ran, Z.Q.; Zhou, M.L. Phase-lead Linear Active Disturbance Rejection Controller of
Permanent Magnet Synchronous Linear Motor Platform. In Proceedings of the Chinese Automation Congress, Shanghai, China,
6–8 November 2020; pp. 392–397. [CrossRef]
12. Liu, R.; Zhang, C.; Shen, L.Y.; Zhao, F.; Zhang, Y.L.; Zhang, S.C. Parameters Identification and Adaptive Feedforward Control
of Permanent Magnent Linear Synchronous Motor. In Proceedings of the IEEE 11th Conference on Industrial Electronics and
Applications, Hefei, China, 5–7 June 2016; pp. 1898–1903. [CrossRef]
13. Huang, W.L.; Wang, Y.H.; Kuo, F.C.; Yen, J.Y.; Kuo, F.C.; Chou, S.C.; Chung, T.T.; Wang, F.C. Integrating time-optimal motion
profiles with position control for a high-speed permanent magnet linear synchronous motor planar motion stage. Precis. Eng.
2021, 68, 106–123. [CrossRef]
14. Kwon, W.H.; Han, S.H. Receding Horizon Control: Model Predictive Control for State Models; Springer: London, UK, 2005. [CrossRef]
15. Errouissi, R.; Ouhrouche, M. Nonlinear predictive controller for a permanent magnet synchronous motor drive. Math. Comput.
Simul. 2010, 81, 394–406. [CrossRef]
16. Errouissi, R.; Ouhrouche, M.; Chen, W.H.; Trzynadlowski, A.M. Robust Nonlinear Predictive Controller for Permanent-Magnet
Synchronous Motors With an Optimized Cost Function. IEEE Trans. Ind. Electron. 2012, 59, 2849–2858. [CrossRef]
Actuators 2024, 13, 34 13 of 13

17. Chai, S.; Wang, L.P.; Rogers, E. Model predictive control of a permanent magnet synchronous motor with experimental validation.
Control. Eng. Pract. 2013, 21, 1584–1593. [CrossRef]
18. Zhao, Y.; Liu, X.D.; Zhang, Q. Predictive Speed-Control Algorithm Based on a Novel Extended-State Observer for PMSM Drives.
Appl. Sci. 2019, 9, 2575. [CrossRef]
19. Li, S.H.; Liu, H.X.; Fu, W.S. Predictive functional control of PMSM based on a composite prediction model. In Proceedings of the
IEEE International Symposium on Sensorless Control for Electrical Drives and Predictive Control of Electrical Drives and Power
Electronics, Munich, Germany, 17–19 October 2013. [CrossRef]
20. Wang, S.; Zhu, W.J.; Shi, J.; Ji, H.; Huang, S.R. A High Performance Permanent Magnet Synchronous Motor Servo System Using
Predictive Functional Control and Kalman Filter. J. Power Electron. 2015, 15, 1547–1558. [CrossRef]
21. Shao, M.; Deng, Y.T.; Li, H.W.; Liu, J.; Fei, Q. Robust Speed Control for Permanent Magnet Synchronous Motors Using a
Generalized Predictive Controller With a High-Order Terminal Sliding-Mode Observer. IEEE Access 2019, 7, 121540–121551.
[CrossRef]
22. Wei, Y.; Wei, Y.J.; Sun, Y.N.; Qi, H.H.; Guo, X.Q. Prediction Horizons Optimized Nonlinear Predictive Control for Permanent
Magnet Synchronous Motor Position System. IEEE Trans. Ind. Electron 2020, 67, 9153–9163. [CrossRef]
23. Li, L.; Pei, G.; Liu, J.; Du, P.; Pei, L.; Zhng, C. 2-DOF Robust H Speed Control for Permanent Magnet Synchronous Motor with
Disturbance Observer. IEEE Trans. Power Electron. 2020, 36, 3462–3472. [CrossRef]
24. Chen, P.C.; Luo, Y. A Two-Degree-of-Freedom Controller Design Satisfying Separation Principle With Fractional-Order PD and
Generalized ESO. IEEE-ASME Trans. Mechatron. 2022, 27, 137–148. [CrossRef]
25. Xia, C.; Ji, B.; Shi, T.; Yan, Y. Two-degree-of-freedom proportional integral speed control of electrical drives with Kalman-filter-
based speed estimation. IET Electric. Power Appl. 2016, 10, 18–24. [CrossRef]
26. Peng, H.; Yang, Z.J.; Xue, W.C.; Huang, R.R.; Huang, Y. The Design and Control of a Rigid-Flexible Coupling Positioning Stage for
Enhanced Settling Performance. J. Dyn. Syst. Meas. Control. 2021, 143, 111003. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.