energies

Article
Dynamic Decoupling Method Based on Motor Dynamic
Compensation with Application for Precision
Mechatronic Systems
Kaixin Liu 1 , Yang Liu 2, *, Fazhi Song 2 and Jiubin Tan 2

1 Department of Control Science and Engineering, Harbin Institute of Technology, Harbin 150001, China;
19b904023@stu.hit.edu.cn
2 Center of Ultra-Precision Optoelectronic Instrument Engineering, Harbin Institute of Technology,
Harbin 150001, China; fazsong@hit.edu.cn (F.S.); jbtan@hit.edu.cn (J.T.)
* Correspondence: hitlg@hit.edu.cn

Abstract: Motors are widely employed in mechatronic systems, especially in precision multiple
degrees of freedom motion systems. In most applications, the dynamic equation between the motor
instruction and the actual driving force is simplified as a constant. Subsequently, the static decoupling
method can be utilized to design the feedback controller. However, in high-precision mechatronic
systems, motor dynamics cannot be neglected, and the static decoupling performance is compromised
due to discrepancies between motors. In this paper, a dynamic decoupling method is developed to
improve the decoupling performance of the multiple-input multiple-output systems. The effects of
transmission delays, motor dynamics, and discrepancies between different motors are taken into
consideration in the dynamic decoupling method. Furthermore, a data-driven optimization method
is developed to estimate the parameters of the dynamic decoupling controller. The effectiveness and
superiority of the proposed method are demonstrated through numerical simulations. The experi-
mental results show that the dynamic decoupling control method can achieve a 97.75% performance
improvement at least compared to the static decoupling control method.

Keywords: data-driven; dynamic decoupling control; motors; multiple-input multiple-output

Citation: Liu, K.; Liu, Y.; Song, F.; Tan,
(MIMO) systems
J. Dynamic Decoupling Method Based
on Motor Dynamic Compensation
with Application for Precision
Mechatronic Systems. Energies 2024,
17, 2038. https://doi.org/10.3390/
1. Introduction
en17092038 Multiple degrees of freedom (DoFs) mechatronic systems play a pivotal role in various
industrial applications, such as robot manipulators [1], vibration test systems [2], numerical
Academic Editor: Daniel
control machine tools [3], applications in the field of energy [4], etc. The parallel design
Morinigo-Sotelo
scheme, owing to its spatial and dynamic advantages, is widely adopted in these systems [5].
Received: 12 March 2024 However, the inherent simplicity of this design introduces challenges, particularly in control
Revised: 3 April 2024 aspects, which become more pronounced in applications to achieve high-precision motion,
Accepted: 10 April 2024 such as the wafer scanner utilized in lithography equipment [6].
Published: 25 April 2024 To obtain high-precision performance for each degree of freedom (DOF), numerous
control strategies have been developed [7], and the decoupling control method emerges
as the most widely employed approach. Based on the mechanical structure and operating
under the rigid body hypothesis, a static decoupling matrix can be established to effectively
Copyright: © 2024 by the authors.
decouple the motion of each DOF [8]. Nevertheless, three significant challenges arise in
Licensee MDPI, Basel, Switzerland.
This article is an open access article
high-precision motion scenarios. Firstly, the static decoupling matrix, which is employed
distributed under the terms and
for achieving rigid body decoupling, relies on mechanical structure parameters, including
conditions of the Creative Commons
the positions of the center of mass (CoM) and the stress point. Owing to the impact of
Attribution (CC BY) license (https:// machining and assembly errors, discrepancies may arise between the utilized parameters
creativecommons.org/licenses/by/ and their actual values, resulting in static decoupling errors [9]. Secondly, to achieve
4.0/). high-performance motion in mechatronic systems, high-performance motors, such as

Energies 2024, 17, 2038. https://doi.org/10.3390/en17092038 https://www.mdpi.com/journal/energies

Energies 2024, 17, 2038 2 of 18

linear motors, voice coil motors, and piezoelectric ceramic motors, are often employed
as actuators [10]. In the static decoupling method, the characteristics of each motor are
considered to be consistent and unchanging during motion. However, owing to variations
between different motors and drivers, as well as limitations in magnetic field strength
and current loop bandwidth [11], the static decoupling method cannot achieve complete
decoupling in essence. Therefore, the decoupling performance of the static decoupling
method is compromised. Thirdly, the varying transmission delays across different channels
further exacerbate the decoupling effect [12]. Owing to the influences of these factors, the
decoupled system using a static decoupling matrix does not exhibit a diagonal form, leading
to the emergence of interactions among degrees of freedom, commonly known as cross-
talk [13,14]. With increasing accuracy requirements, cross-talk has become a significant
impediment to achieving the desired performance in precision mechatronic systems [15].
To eliminate the cross-talk and improve the decoupling performance, various methods
have been reported in the literature, which can be categorized into two main approaches.
The first approach treats cross-talk as a parasitic dynamic of each DOF, allowing the design
of a robust feedback controller to suppress its effects [16]. The most prevalent method
within this category is the equivalent transfer function (ETF) method [17]. In the ETF
method, the plant to be controlled in each loop is represented by a generalized model
when other loops are closed [18]. It is evident that the interactions are taken into account,
therefore, a better feedback controller can be designed for improved performance. However,
because the generalized plant of each DoF incorporates the influence of controllers from
other DoFs, it is complicated to refine the controller when the performance requirement
is not satisfied. In [19], an ETF method is proposed, in which the complicated interaction
modes are approximated as lower-order equivalent models to simplify the design of the
controller. In addition to the ETF method, other robust control methods can be utilized
such as the hybrid integrator–gain method, the H∞ method, the model predictive control
method, etc. [20]. However, since the interactions inevitably increase the complexity of
the plant and reduce the attainable bandwidth, only a limited reduction of cross-talk can
be obtained. Consequently, the second method, which directly improves the accuracy of
decoupling, proves to be the more effective approach in enhancing the performance of
precision mechatronic systems.
As highlighted earlier, interactions, stemming from errors in mechanical parameters,
variations between motors, and transmission delays, can compromise system performance.
Consequently, it becomes imperative to compensate for these factors to achieve full de-
coupling. Conventionally, mechanical parameters can be calibrated using professional
instruments to enhance the accuracy of the static decoupling matrix [21]. However, due to
limitations in workspace, certain instruments may no longer be suitable [22]. Furthermore,
this indirect optimization method may introduce additional calculation errors in the process.
Due to the influence of the other two factors, it can be noted that achieving full decoupling
of the system with a static decoupling matrix becomes impractical. Based on the compen-
sation for the inconsistency of motors and transmission channels, a dynamic decoupling
controller is proposed in this paper. The dynamic decoupling controller consists of two
components: a dynamic component parameterized as a rational transfer function, and a
static component. The dynamic component is employed to address inconsistencies between
motors and transmission channels, while the static component is utilized to achieve full
decoupling of the rigid body of the mechatronic system.
Additionally, to obtain the optimal estimate of the unknown parameter in the dynamic
decoupling controller, a data-driven optimization method is developed in this paper, which is
inspired by the virtual reference feedback tuning (VRFT) method [23]. The key contributions
of this work, in comparison to existing literature, can be succinctly outlined as follows:
(1) A dynamic decoupling control method is established in this paper, based on the
compensation for the motor dynamics and the transmission delays.
(2) A data-driven optimization method is developed in this paper to estimate the un-
known parameters of the dynamic decoupling controller, requiring only minimal
Energies 2024, 17, 2038 3 of 18

model information. Furthermore, the proposed method is non-iterative, necessitating

only a single experiment.
Notations: Let R be the set of real numbers. For n, a positive integer, Rn is the set of
real vectors of dimension n. Vector x ∈ Rn denotes [ x1 , ..., xn ] T . For a vector xc ∈ Rn ,
the i-th element can be denoted as xc,i . We define ∥ · ∥2 as the l2 norm of vectors, i.e.,
∥ x ∥2 = (∑in=1 xi2 )1/2 .

2. Problem Formulation
2.1. Control Configuration of the Multiple DoFs Mechatronic System
We consider a precision multiple DoFs mechatronic system, as depicted in Figure 1,
which is a prototype for the short stroke stage of the wafer stage. The prototype shown in
Figure 1 consists of a stator and a mover with sufficient stiffness. The mover is driven by
three voice coil motors labeled as m1 , m2 , and m3 , respectively, to accomplish nano-scale
motion. The voice coil motor can achieve precise and linear motion with high acceleration
and high force-to-weight ratio, which render it ideal for applications that require fast and
accurate positioning. Three linear encoders are mounted close to the motors; for simplicity,
the measurement data of the encoders are supposed to be equal to the displacement
distance of the voice coil motors. The target point, defined as the center point of the mover,
is controlled to track a predetermined trajectory. The reference trajectory is defined in a
Cartesian coordinate system, denoted as Cxyz , and the displacements of the three degrees
of freedom of the target point can be directly obtained as yz , yθx , and yθy , respectively.

Stator

m3 m2

Mover
C xyz x

m1

(a) Overhead view of the prototype.

Mover

A1

A2
m1
B3 m3
B1
m2
B2
Stator

A1 , A2 , A3 (hiden) : Magnetic steels of motors B1 , B2 , B3 : Coils of motors

(b) Detailed mechanical structure drawing of the prototype.

Figure 1. The diagram of the prototype of the short stroke stage.

Energies 2024, 17, 2038 4 of 18

To control the prototype, the decoupling control scheme based on a nominal static
decoupling matrix can be utilized, as Figure 2 shows. In this configuration, P(s) ∈ C3×3 is
the rigid body dynamics model of the mover describing the relationship between the motor
driving force and the displacement of the target point. Pm (s) ∈ C3×3 represents the transfer
function between motor instruction and motor driving force, which is in the form of a
diagonal. Pt (s) ∈ C3×3 is denoted as a transmission delays matrix of three transmission
channels from the digit controller to drivers. K ∈ R3×3 is the static decoupling matrix to
decouple the plant P(s). C (s) ∈ C3×3 is the diagonal feedback controller matrix comprised
of single-input single-output controllers designed with respect to the desired decoupled
system. The signals r (t) ∈ R3 and e(t) ∈ R3 are a high-order polynomial trajectory and
tracking error, respectively. u(t) ∈ R3 is the control signal calculated by C (s). f c (t) ∈ R3
and f (t) ∈ R3 represent the motor instruction and the actual driving force of the motor.
y(t) ∈ R3 is the measured displacement of the target point in coordinate Cxyz . In addition,
the displacements of three voice coil motors can be denoted as yv (t) ∈ R3 , which is not
represented in the block diagram. Based on the definitions above, the dynamic model of
the control system can be described as follows:


 y(t) = P(s) f (t)

 f (t) = Pm (s) Pt (s) f c (t)



f c (t) = Ku(t) . (1)

u(t) = C (s)e(t)





e(t) = r (t) − y(t)


r (t ) e(t ) u (t ) f c (t ) f (t ) y (t )
C (s) K Pt ( s ) Pm ( s ) P( s)
-

Figure 2. The configuration of the decoupling control.

2.2. Description of Nominal Static Decoupling Method

To simplify the controller design, the nominal static decoupling method can be pri-
marily used to stabilize the system. In the static decoupling method, the transmission
delays and the motor dynamics are neglected so that Pt (s) = I and Pm (s) = αI, where I is
the identity matrix. The simplified structure of the prototype is shown in Figure 3a, with
the nominal mechanical parameters known. From Figure 3a, it can be concluded that the
CoM coincides with the target point and the distances between motors and the CoM in
the coordinate Cxyz are (0, −l1 ), (l2 , l1 ), and (−l2 , l3 ), respectively. Under the rigid body
hypothesis, the dynamic equations of the motion of the target point can be established as

my¨z = f 1 + f 2 + f 3

Jx y¨θx = −l1 f 1 + l3 f 2 + l3 f 3 (2)

Jy y¨θy = −l2 f 2 + l2 f 3


where f 1 , f 2 , and f 3 are driving forces of three motors, respectively. m is the mass of the
mover. Jx and Jy are moment of inertia of θ x and θy at the CoM, respectively. From (2), P(s)
can be modeled as  
1 1 1
2 ms2 ms2
 msl1 l3 l3 
P(s) =  − Jx s2  . (3)
 Jx s2 Jx s2

0 − J l2s2 l2
Jy s2
y
 Energies 2024, 17, 2038 5 of 18

y y
m3 l2
m2

l3

x
x
l1
z m1 z

(a)

y m3 ( x2 , y2 )

( x3 , y3 ) m2

( x0 , y0 )

( x1 , y1 )

x m1 z

x
y
(b)
x
Figure 3. The simplified structure diagrams of the prototype with (a) nominal mechanical parameters
and (b) actual mechanical parameters.
y
It is evident that P(s) is a multivariable system and is not in diagonal form. Directly
designing an appropriate feedback controller for such a system is challenging. To apply
the static decoupling method, a static decoupling matrix should be employed. Based on
the nominal model P(s), the nominal static decoupling matrix can be determined easily
as follows:  
l3 1
− 0
 l1 +l1
l3 l1 + l3
1 1 
K=  2(l1 +l3 ) 2(l1 +l3 ) − 2l2 . (4)

l1 1 1
2 ( l1 + l3 ) 2 ( l1 + l3 ) 2l2

Consequently, the decoupled system achieves full decoupling, which can be written as

D (s) = P(s)K
 1 
ms2
0 0
1 (5)
= 0 0 
.

Jx s2
1
0 0 Jy s2

Subsequently, three single-input single-output (SISO) feedback controllers, such as

PID controllers, can be easily designed with respect to 1/ms2 , 1/Jx s2 , and 1/Jy s2 . However,
in the actual system, the actual mechanical parameters may differ from the nominal values,
and Pt (s) and Pm (s) cannot be considered static under high precision requirements.
Energies 2024, 17, 2038 6 of 18

2.3. Formulation of Factors Causing Incomplete Decoupling

As Figure 3b shows, due to errors in manufacturing and assembly, the CoM of the
mover no longer coincides with the target point, and the actual installation positions of
the motors have also changed. In this situation, the actual rigid body dynamic would no
longer be equal to the nominal rigid body dynamic (3), which can be expressed as (6):
 
P11 (s) P12 (s) P13 (s)
P(s) =  P21 (s) P22 (s) P23 (s). (6)
P31 (s) P32 (s) P33 (s)

where the elements can be detailed as


 P (s) = Jx Jy + x0 ( x0 − x1 )mJx + y0 (y0 − y1 )mJy

11
mJx Jy s2







 Jx Jy + x0 ( x0 − x2 )mJx + y0 (y0 − y2 )mJy
P12 (s) =


mJx Jy s2







 Jx Jy + x0 ( x0 − x3 )mJx + y0 (y0 − y3 )mJy
P13 (s) =


mJx Jy s2





y0 − y1


 P21 (s) = − J s2





 x
y0 − y2 . (7)
P22 (s) = −


 Jx s2
y − y3


 P23 (s) = − 0


Jx s2




x0 − x1


P31 (s) =


Jy s2





x0 − x2


P32 (s) =


Jy s2





x0 − x3



 P33 (s) =


Jy s2

It is evident that the actual plant cannot be completely decoupled with the nominal
decoupling matrix, unless the following equations are satisfied:


 x0 =0

y0 =0





x1 =0





y = − l1

1
. (8)


 x2 = l2

y


 2 = l3

 x3 = − l2




y3 = l3


Although the actual system with a nominal decoupling matrix can maintain stability
in the case of small errors, it is difficult to obtain the desired performance. Therefore, the
decoupling matrix should be improved to enhance the level of decoupling. To decouple
the actual rigid body model, the optimal static decoupling matrix can be obtained based
on accurate mechanical parameters, such as xi and yi , for i = 0, 1, 2, 3. However, under
the requirements of high speed and high-precision accuracy, the motor dynamics and the
transmission delays cannot be neglected. As depicted in Figure 4, with the same coil current
input, the static output forces of different motors in different displacements are not constant.
Energies 2024, 17, 2038 7 of 18

Further considering the effect of the motor time constant, the dynamic model of the motor
can be expressed as follows:
g
1 ( yv,1 )

α1 s +1 0 0
 g2 (yv,2 ) 
Pm (s) = 
 0 α2 s +1 0 
 (9)
g3 (yv,3 )
0 0 α3 s +1

where yv,i is the i-th element of yv . gi (yv,i ) for i = 1, 2, 3, which indicates that the force
constant of the motor is related to the displacement of the motor.

11.6

11.4

11.2

11.0

10.8

10.6
-1.5 -1.0 -0.5 0 0.5 1.0 1.5

Figure 4. The measured force constants for different motors at different displacements.

On the other hand, the transmission delays matrix can be expressed in detail as

e− β1 s
 
0 0
Pt (s) =  0 e− β2 s 0  (10)
0 0 e− β3 s

where β i for i = 1, 2, 3 represents the delays of different transmission channels.

It can be claimed that, if and only if the following conditions are satisfied, the ac-
tual plant can be decoupled by a static decoupling matrix. According to the conditions,
Theorem 1 can be established. 
 g1 , g2 , g3 = const

α1 = α2 = α3 . (11)

β1 = β2 = β3


Theorem 1. Suppose that controlled plant can be represented as G (s) = P(s) Pm (s) Pt (s), where
P(s), Pm (s) and Pt (s) are in the form of (6), (9), and (10), respectively. Then, if G (s) can be
completely decoupled by a static decoupling matrix, the condition (11) must be satisfied.

Proof. Using proof by contradiction, we suppose that there exists a constant matrix K such
that G (s)K is diagonal, and this can be expressed as
 
d1 ( s ) 0 0
D (s) = G (s)K =  0 d2 ( s ) 0  (12)
0 0 d3 ( s )

where the j-row elements of D (s) can be given as

Energies 2024, 17, 2038 8 of 18

D j (s) = Gj (s)K = [0, · · · , 0, d j (s), 0, · · · , 0] . (13)

| {z } | {z }
j −1 n− j

We denote k̂ j,i as the element in the j-th row and i-th column of the inverse matrix of
K. Then, Gj (s) can be written as

G j ( s ) = D j ( s ) K −1
= [0, · · · , 0, d j (s), 0, · · · , 0]K −1 . (14)
= d j (s)[k̂ j,1 , k̂ j,2 , · · · , k̂ j,n ]

a j (s)
We suppose that d j (s) can be expressed as d j (s) = b j (s)
, where a j (s) and b j (s) are
coprime polynomials and b j (s) is monic. From (6), the actual rigid body model can
be simplified:
 
p11 p12 p13
1
P(s) = 2  p21 p22 p23 . (15)
s
p31 p32 p33
Subsequently, the j-th row of G (s) can also be determined as

Gj (s) = Pj (s) Pm (s) Pt (s)

1 − β1 s p j1 g1 (yv,1 ) p j2 g2 (yv,2 ) ( β1 − β2 )s p j3 g3 (yv,3 ) ( β1 − β3 )s (16)
= e [ , e , e ]
s2 α1 s + 1 α2 s + 1 α3 s + 1

where Pj (s) is the row vector consisting of the elements of the j-th row of P(s).
Comparing (14) and (16), the conditions (11) can be directly obtained.

Therefore, to achieve full decoupling of the actual system with motor dynamics
and transmission delays, the static decoupling matrix should be substituted by a dynamic
decoupling controller. Based on the above analysis, the ideal dynamic decoupling controller
can be determined as the following form:

K (s) = Pt−1 (s) Pm−1 (s)Ks . (17)

In order to obtain a parameterized model that is easy to handle, two simplification

treatments are applied to Pm−1 (s) and Pt−1 (s). First, the variant force constants gi (yv,i ) are
approximated as polynomials ḡi (yv,i ) of order of m for i = 1, 2, 3. ḡi (yv,i ) can be written as

ḡi (yv,i (t)) = γ0i + γ1i yv,i (t) + · · · + γm

i m
yv,i (t) (18)

j
where yv,i (t) represents the j-th power of yv,i (t) and γij ∈ R for j = 0, ·, m are unknown
coefficients. Then the inverse of Pm (s) can be approximated as
α1 s +1
 
ḡ1 (yv,1 )
0 0
 α2 s +1 
P̂m (s) = 
 0 ḡ2 (yv,2 )
0 .
 (19)
α3 s +1
0 0 ḡ3 (yv,3 )

Second, to obtain the an invertible approximation of the pure delays term of Pt (s), the
Pade approximant [24] can be utilized, and the inverse of Pt (s) can be written approximately as
 
β1 s + 1 0 0
P̂t (s) =  0 β2 s + 1 0 . (20)
0 0 β3 s + 1
Energies 2024, 17, 2038 9 of 18

Finally, to ensure the causality of the dynamic decoupling controller, a Q-filter with a
relative order of 2 is required, and the desired dynamic decoupling controller based on the
compensation of motor dynamics and transmission delays can be parameterized as

K (s) = Q(s) P̂t (s) Pˆm (s)Ks

 (α1 s+1)( β1 s+1) 
0 0  
 ḡ1 (yv,1 ) L(s)
(α2 s+1)( β 2 s+1)  k11 k12 k13 (21)
=
 0 ḡ2 (yv,2 ) L(s)
0 k21
 k22 k23 
0 0
(α3 s+1)( β 3 s+1) k31 k32 k33
ḡ3 (yv,3 ) L(s)

where L(s) is a known second-order stable polynomial in the s domain. As mentioned

above, the unknown parameters of K (s) can be determined based on the mechanical pa-
rameters, motor dynamics, and transmission delays. However, it is difficult to measure and
identify these parameters, and additional errors will be introduced in the final calculation
process. To address these problems, an on-line optimization method is developed.

3. Data-Driven Optimization Method

In this section, a data-driven method is proposed to estimate the unknown parameters
of the dynamic decoupling controller K (s). The configuration of the proposed method is
depicted in Figure 5. As shown in Figure 5, a reference model is used which should be
diagonal, and the essential idea is to adjust the parameters of K (s) to make the decoupled
system as close to D̃ (s) as possible.

r (t ) e(t ) u (t ) f c (t ) f (t ) y (t )
C (s) K (s) Pt ( s ) Pm ( s ) P( s)
-

e (t ) u (t ) y (t ) -
C (s) D( s) e (t )
-

Figure 5. Block diagram of the proposed data-driven optimization method.

In other words, if the output of D̃ (s) is equal to the output of G (s)K (s) with the
same conditions of reference input and feedback controller, then D̃ (s) = G (s)K (s) can be
achieved. We suppose that the output of the reference model is equal to y(t); the output of
the feedback controller in reference model can be determined as

y(t) = D̃ (s)ũ(s). (22)

From (22), if the output of the feedback controller u(t) in the actual system can be equal
to ũ(t) by adjusting the parameters of K (s), the following equations can be established:

D̃ (s)ũ(s) = G (s)K (s)u(t)

. (23)
D̃ (s) = G (s)K (s)

It is obvious that the decoupled system is completely diagonal if (23) is satisfied.

Therefore, the following equivalent objective function can be established, and the optimal
estimate of the parameters of K (s) can be obtained by minimizing it:

K ∗ (s) = arg min J = arg min ∥ f c (t) − K (s)ũ(t)∥22 . (24)

K (s) K (s)

where ∥ x ∥2 denotes the l2 -norm of vector x.

Energies 2024, 17, 2038 10 of 18

Substituting (21) into (24), the objective function J can be derived as

J = ∥ f c (t) − Q(s) P̂t (s) P̂m (s)Ks ũ(t)∥22

3
= ∑ ∥ f c,i (t) − Q(s) P̂t,i (s) P̂m (s)Ks ũ(t)∥22 (25)
i =1
3
= ∑ Ji
i =1

where f c,i represents the i-th element of f c . P̂t,i (s) denotes a row vector consisting of the
i-th row elements of P̂t (s). Then, the optimization of J can be achieved by optimizing J1 , J2 ,
and J3 , respectively.
Substituting (21) into (25), Ji can be parameterized as follows:

Ji = ∥ f c,i (t) − Q(s) P̂t,i (s) P̂m (s)Ks ũ(t)∥22

3
(αi s + 1)( β i s + 1)
= ∥ f c,i (t) −
ḡi (yv,i ) L(s) ∑ kij ũ j (t)∥22
j =1 (26)
3
1 (α s + 1)( β i s + 1)
=∥ ( ḡ (y ) f (t) − i
ḡi (yv,i ) i v,i c,i L(s) ∑ kij ũ j (t))∥22
j =1

where P̂t,i (s) denotes the i-th row of P̂t (s). Since gi is not equal to 0 and fluctuates less
around the nominal value and the minimum of Ji depends on the error term in (26), the
minimization of Ji can be equivalent to the minimization of J̄i , which can be written as

3
(αi s + 1)( β i s + 1)
J̄i = ∥ ḡi (yv,i ) f c,i (t) −
L(s) ∑ kij ũ j (t)∥22 . (27)
j =1

Substituting (18) into (27), the objective function can be further derived as

3
(αi s + 1)( β i s + 1)
J̄i = ∥(γ0i + γ1i yv,i (t) + · · · + γm
i m
yv,i (t)) f c,i (t) −
L(s) ∑ kij ũ j (t)∥22 . (28)
j =1

For the simplicity of expression, we define the following symbols:

j



 hi,j (t) = yv,i (t) f c,i (t)

s2


 2
q j (t) = ũ (t)

L(s) j




s

q1j (t) =

ũ (t)

s) j



 L (

q0 (t) = 1 ũ (t)



j
L(s) j





γij

3 . (29)
 θ i,j = i


 γ 0

β i k ij


 2
α i
=

 θi,j
γ0i






θ 1 = (αi + β i )k ij



i,j
γ0i






k ij


 0
θi,j = i


γ0
Energies 2024, 17, 2038 11 of 18

Subsequently, (28) can be simplified as

m 3 3 3
1
J̄i =
(γ0i )2
∥ h i,0 ( t ) + ∑ θ 3
i,j h i,j ( t ) − ∑ θ 2 2
i,j q j ( t ) − ∑ θ 1 1
i,j q j ( t ) − ∑ 0 0
θi,j q j (t)∥22 . (30)
j =1 j =1 j =1 j =1

To solve the optimization problem in practice, (30) should be converted to a

discrete form

1 N 1 m 3 3 3
J̄i = ∑
N k=1 (γ0i )2
∥ h i,0 ( k ) + ∑ θ 3
i,j h i,j ( k ) − ∑ θ 2 2
i,j q j ( k ) − ∑ θ 1 1
i,j q j ( k ) − ∑ 0 0
θi,j q j (k)∥22 (31)
j =1 j =1 j =1 j =1

where hi,j (k ), q2j (k ), q1j (k ), and q0j (k) are the k-th sampling values of hi,j (t), q2j (t), q1j (t), and
q0j (t), respectively and N is the sampling number. For simplicity, we define
 i T N
 T = [hi,0 (1), hi,0 (2), · · · , hi,0 ( N )] ∈ R


ϕi (k) = [hi,1 (k), hi,2 (k), · · · , hi,m (k), q21 (k), q22 (k), q23 (k)






, q11 (k), q12 (k), q13 (k), q01 (k), q02 (k), q03 (k)] T ∈ Rm+9






Φi = [ϕi (1), ϕi (2), · · · , ϕi ( N )] T ∈ R N ×(m+9)






3 T
ηi3 = [θi,1
3 3
, θi,2 , · · · , θi,m ] ∈ Rm . (32)

2 2 2 2 T 3

ηi = [θi,1 , θi,2 , θi,3 ] ∈ R





 1 1 1 1 T
] ∈ R3



 ηi = [θi,1 , θi,2 , θi,3

0 T
ηi0 = [θi,1
0 0
] ∈ R3

, θi,2 , θi,3






 i
Ω = [−(ηi3 ) T , (ηi2 ) T , (ηi1 ) T , (ηi0 ) T ] T ∈ Rm+9

Substituting (32) into (31), the objective function can be rewritten as

1
J̄i = ∥ T i − Φi Ωi ∥22 . (33)
N (γ0i )2

Then, the estimate of Ωi can be obtained by using the least square method as follows:

Ω̂i = ((Φi ) T (Φi ))−1 (Φi ) T T i . (34)

3 , θ 2 , θ 1 , and θ 0 can easily obtained based on (29), which can be
The estimate of θi,j i,j i,j i,j
3 , θ̂ 2 , θ̂ 1 , and θ̂ 0 . In addition, the estimate of α β , α + β and g ( y ) /γi can
denoted as θ̂i,j i,j i,j i,j i i i i i v,i 0
be obtained as τ̂i2 , τ̂i1 , and ḡˆi (yv,i ), which can be written as

τ̂ 2 = [(η̂i0 ) T η̂i0 ]−1 (η̂i0 ) T η̂i2


 i


τ̂i1 = [(η̂i0 ) T η̂i0 ]−1 (η̂i0 ) T η̂i1 . (35)

 ḡˆ (y ) = 1 + θ̂ 3 y (t) + · · · + θ̂ 3 ym (t)

i v,i i,1 v,i i,m v,i

Finally, the estimate of K (s) can be expressed as

 τ̂2 s2 +τ̂1 s+1 
1 1
ḡˆ (yv,1 ) L(s)
0 0  0 0 0 
 θ̂11 θ̂12 θ̂13
τ̂22 s2 +τ̂21 s+1

 0 0 0 
K̂ (s) =  0 0 . (36)

ḡˆ (yv,2 ) L(s)
 θ̂21 θ̂22 θ̂23
 2 2 1
 0 0 0
τ̂3 s +τ̂3 s+1 θ̂31 θ̂32 θ̂33
0 0 ḡˆ (yv,3 ) L(s)
Energies 2024, 17, 2038 12 of 18

4. Simulations Results
In this section, the effectiveness of the proposed method is validated by comparing it
with the nominal decoupling method and the ideal static decoupling method, which can
decouple the rigid body model of the system in the absence of the motor dynamics and
transmission delays. These methods can be represented as follows:
(1) M0 : the nominal static decoupling control method.
(2) M1 : the ideal static decoupling control method.
(3) M2 : the proposed dynamic decoupling control method.
The static decoupling matrices obtained by M1 and M1 are denoted as K0 and K1 ,
respectively. The dynamic decoupling controller determined by M2 is represented as K2 (s).

4.1. Simulation Setup

A simulation experiment was conducted to demonstrate the effectiveness of M2 and
the simulation block diagram was established, as Figure 6 shows. A fixed step size of 0.0002
was adopted in the numerical calculations of differential equations, and the algorithm of
the solver was set to ode4. The total simulation time was set to 1 s, which meant that the
sampling number was 5000. The actual values of mechanical parameters in the coordinate
Cxyz are provided in Table 1. We supposed that the mass of the SS stage was m = 20 kg and
the moment of inertia of θ x and θy at the CoM were Jx = 0.33 kg · m and Jy = 0.45 kg · m,
respectively. By the rigid body hypothesis, the transfer function model between the driving
forces of motors and the three DoFs’ displacement of the target point can be given as
 
0.05367 0.04523 0.04987
s2 s2 s2
0.3333 0.2879 0.2576 
P(s) =  . (37)

s2 s2 s2 
0.03333
s2
− −0.1889
s2
0.2444
s2

rz uz f c ,1 g1 ( yu ,1 ) f1 yz
Cz ( s) e - b1s
- a1 s + 1

rq x uq x f c ,2 g 2 ( yu ,2 ) f 2 yq x
Cq x ( s ) K e- b2 s P( s)
- a2s +1

rq y uq y f c ,3 f3 yq y
g3 ( yu ,3 )
Cq y ( s ) e - b3 s

- a3s + 1

Figure 6. The diagram of the simulation experiment.

Table 1. Actual mechanical parameters used in the simulation.

Points Values (m)

( x0 , y0 ) (0.01, 0.01)
( x1 , y1 ) (−0.005, −0.1)
( x2 , y2 ) (0.095, 0.105)
( x3 , y3 ) (−0.1, 0.095)

The nominal mechanical parameters presented in Figure 3a are l1 = 0.1m, l2 = 0.1m,

and l3 = 0.1m. Based on the nominal mechanical parameters, the nominal static decoupling
matrix K0 can be determined as
 
0.5 −5.0 0
K0 = 0.25 2.5 −5.0. (38)
0.25 2.5 5.0
Energies 2024, 17, 2038 13 of 18

We specify the desired decoupled model as follows:

 
0.05
2 0 0
 s 3.0303
D (s) =  0 0 . (39)

s2
2.2222
0 0 s2

Based on the actual rigid body model of the plant and the desired decoupled model,
the ideal static decoupling matrix K1 can be determined as
 
0.4535 −4.7284 −0.4581
K1 = 0.3432 2.6454 −5.1557. (40)
0.2033 2.6890 5.1694

To investigate the effect of motor dynamics and transmission delays, the dynamic
equations in the form of (9) and (10) were added into the simulation. The parameters used
are listed in Tables 2 and 3. In the simulation, the variant force constants for different
motors were approximated as a polynomial of order 5, according to the measurement data
shown in Figure 4. Because these parameters were close to the nominal values, the diagonal
dominance of the system could still be ensured by using the nominal static decoupling
method when these errors were ignored. Thus, a decentralized controller C (s) could
be designed to ensure the stability of the system. Nevertheless, due to the inaccurate
mechanical parameters, there would have been a great deterioration of performance. To
improve the performance of the precision mechatronic system, the dynamic decoupling
method proposed in Section 3 can be utilized. In order to estimate the unknown parameters
of the dynamic decoupling controller, a data-driven optimization method M2 is developed
in Section 4.

Table 2. Actual parameters of motor dynamics and transmission delays used in the simulation.

Parameters Values
α1 0.002
α2 0.0018
α3 0.0021
β1 0.0002
β2 0.0001
β3 0.0003

Table 3. Parameters of the force constant polynomial used in the simulation.

Parameters
m1 m2 m3
( N/mi )
α0 11.05 11.54 11.50
α1 −97.35 −108.42 −150.54
α2 −3.47 × 105 −4.67 × 105 −2.30 × 105
α3 1.11 × 108 6.88 × 107 1.38 × 108
α4 7.02 × 1010 9.44 × 1010 2.26 × 1010
α5 −3.28 × 1013 −1.30 × 1013 −3.73 × 1013

To estimate the unknown parameters in M2 , the reference D̃ (s) model was selected as
D (s), and the order used in (18) was chosen as m = 2 in the simulation. By conducting exci-
tation experiments on the closed-loop system using K0 , the dynamic decoupling controller
K2 (s) could be determined. The estimated results are shown in Table 4 with the Q-filter
selected as follows:
1
Q(s) = − 6 2
(41)
7.036 × 10 s + 0.003751s + 1
Energies 2024, 17, 2038 14 of 18

Table 4. The estimated results based on the structure of dynamic decoupling controller.

Parameters Estimate Parameters Estimate Parameters Estimate

τ̂12 4.159 × 10−7 τ̂22 1.909 × 10−7 τ̂32 6.698 × 10−7
τ̂11 0.002198 τ̂21 0.0019 τ̂31 0.002393
0
θ̂11 0.4718 0
θ̂12 −4.9611 0
θ̂13 −0.6164
0
θ̂21 0.3421 0
θ̂22 2.2755 0
θ̂23 −5.3083
0
θ̂31 0.2033 0
θ̂32 3.0487 0
θ̂33 5.2963
0
θ̂1,1 −8.4778 0
θ̂1,2 −30,359
0
θ̂2,1 −9.7282 0
θ̂2,2 −41,652
0
θ̂3,1 −12.7135 0
θ̂3,2 −21,208

4.2. Performance Assessment

To demonstrate the effectiveness of the dynamic decoupling method more intuitively,
some validation experiments in the time domain were carried out. Tracking experiments
for three DoFs were conducted for different systems with the same feedback controller. In
the first tracking experiment, a step signal of 100 µm was exerted in z-DoF as the reference
input. Similarly, a step signal of 100 µrad was exerted in θ x -DoF and θy -DoF as the reference
input in the second and the third tracking experiments, respectively. It should be noted that
in each tracking experiment, the reference inputs of the other two DoFs were set as 0, except
the one tracking the step signal. Therefore, the output responses of these DoFs should be
0 when the system is fully decoupled. However, the outputs of the DOFs that track the
zero signal were much greater than 0 when using M0 and M1 as shown in Figures 7–9.
The outputs were derived from the inaccurate decoupling, which can be referred to as
the cross-talk. On the contrary, the cross-talk of the decoupled system using M2 was
almost negligible. It can be concluded that the decoupling effect of the M2 was the best.
Consequently, to quantify the degree of decoupling, the sum of the maximum value of
cross-talk can be used as the indicator, which can be denoted as

SAC = ∑ ∑ Wji |e j,i | (42)

i = z j ̸ =i

where i, j are selected in the set of {z, θ x , θy }, respectively, Wji is the weighting coefficient,
and e j,i represents the maximum value of the cross-talk of the j-DoF when the i-DoF is
excitated. Based on the maximum value of cross-talk in tracking experiments shown in
Table 5, the SAC of different decoupled systems using different decoupling methods can be
determined as 

 SAC1 = 1.2799 × 10−4

SAC2 = 2.7684 × 10−5 (43)

SAC = 2.5791 × 10−7

3

where the weighting coefficient is chosen as Wji = 1 for i, j ∈ {z, θ x , θy }. SAC1 , SAC2 , and
SAC3 are calculated based on the decoupled system with K0 , K1 , and K2 (s), respectively. It
can be concluded that the dynamic decoupling method proposed in this paper outperforms
the other two static decoupling methods, and at least 99.46% and 97.75% performance
improvements were obtained compared to M1 and M2 , respectively. Consequently, the
effectiveness and the superiority of the proposed method can be clearly demonstrated.
Energies 2024, 17, 2038 15 of 18

200

100

-100
50

-50
80

40

-40

-80
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35

Figure 7. The output responses of tracking experiment of z-DoF.

0.8

0.4

-0.4

-0.8
200

100

-100
8

-4

-8
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35

Figure 8. The output responses of tracking experiment of θ x -DoF.

Energies 2024, 17, 2038 16 of 18

1.5
1
0.5
0
-0.5
-1
-1.5
10

-10
200

100

-100
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35

Figure 9. The output responses of tracking experiment of θy -DoF.

Table 5. The maximum value of cross-talk in tracking experiments with different decoupling con-
trollers (matrix).

Excitated
ez (µm) eθx (µrad) eθy (µrad)
DoF
None 1 47.53 62.87
z None 13.58 5.096
None 0.07180 0.1148
0.6832 None 6.056
θx 0.3564 None 5.220
0.003698 None 0.02546
1.430 9.417 None
θy 0.5249 2.902 None
0.007016 0.03507 None
1 “None” indicates that the data is not relevant to the content of the text.

5. Conclusions
In this paper, a novel dynamic decoupling method was developed based on the
compensation for varying motor dynamics and transmission delays. The key essence of the
proposed method lies in the structured dynamic component included in the decoupling
controller. What is more, to estimate the unknown parameters, an on-line data-driven
optimization algorithm was presented, where only inputs and outputs of the plant need
to be measured in a single experiment. Both theoretical analysis and experimental results
confirm that the proposed structured dynamic decoupling approach can achieve accurate
decoupling; thus, the interactions between multiple DOFs in the mechatronic system can
be eliminated significantly. Consequently, compared to the conventional static decoupling
methods, a 97.75% performance improvement can be obtained by the proposed dynamic
decoupling method.
Energies 2024, 17, 2038 17 of 18

Author Contributions: Conceptualization, K.L.; methodology, K.L.; writing—original draft, K.L.

and F.S.; software, K.L.; validation, K.L.; investigation, K.L. and Y.L.; supervision, Y.L. and J.T.;
project administration, Y.L. and J.T. All authors have read and agreed to the published version of
the manuscript.
Funding: This research was funded by the National Natural Science Foundation of China (52075132
and 52105546) and the Opening Foundation of State Key Lab of Digital Manufacturing Equipment &
Technology, China (DMETKF2020024).
Data Availability Statement: All data used are contained within the article.
Conflicts of Interest: The authors declare no conflicts of interest.

Abbreviations
The following abbreviations are used in this manuscript:

CoM Center of mass

DoF Degree of freedom
DoFs Degrees of freedom
ETF Equivalent transfer function
MIMO Multiple-input multiple output
SAC Sum of the absolute value of the maximum value of cross-talk
SISO Single-input single-output
VRFT Virtual reference feedback tuning

References
1. Incremona, G.P.; Ferrara, A.; Magni, L. MPC for Robot Manipulators With Integral Sliding Modes Generation. IEEE-ASME Trans.
Mechatron. 2017, 3, 1299–1307. [CrossRef]
2. Zheng, Y.; Zhou, Z.; Huang, H. A multi-frequency MIMO control method for the 6DOF micro-vibration exciting system. Acta
Astronaut. 2020, 170, 552–569. [CrossRef]
3. Hanifzadegan, M.; Nagamune, R. Contouring Control of CNC Machine Tools Based on Linear Parameter-Varying Controllers.
IEEE-ASME Trans. Mechatron. 2016, 5, 2522–2530. [CrossRef]
4. Jeon, T.; Kim, D.; Song, Y.; Paek, I. Design and Validation of Demanded Power Point Tracking Control Algorithm for MIMO
Controllers in Wind Turbines. Energies 2021, 18, 5818. [CrossRef]
5. Ma, J.; Cheng, Z.; Zhu, H.; Li, X.; Tomizuka, M.; Lee, T.H. Convex Parameterization and Optimization for Robust Tracking of a
Magnetically Levitated Planar Positioning System. IEEE Trans. Ind. Electron. 2022, 4, 3798–3809. [CrossRef]
6. Song, F.; Liu, Y.; Jin, W.; Tan, J.; He, W. Data-Driven Feedforward Learning With Force Ripple Compensation for Wafer Stages: A
Variable-Gain Robust Approach. IEEE Trans. Neural Netw. Learn. Syst. 2022, 4, 1594–1608. [CrossRef] [PubMed]
7. Schitter, G. Advanced Mechatronics for Precision Engineering and Mechatronic Imaging Systems. IFAC Pap. 2015, 1, 942.
[CrossRef]
8. Butler, H. Position Control in Lithographic Equipment An Enabler for Current-Day Chip Manufacturing. IEEE Control Syst. Mag.
2011, 5, 28–47.
9. Heertjes, M.; Hennekens, D.; Steinbuch, M. MIMO feed-forward design in wafer scanners using a gradient approximation-based
algorithm. Control Eng. Pract. 2010, 5, 495–506. [CrossRef]
10. Barros, C.P.B.; Butler, H.; van de Wijdeven, J.; Tóth, R. On feedforward control of piezoelectric dual-stage actuator systems.
In Proceedings of the 2021 60th IEEE Conference on Decision and Control (CDC), Austin, TX, USA, 14–17 December 2021;
pp. 5588–5594.
11. Poot, M.; Portegies, J.; Mooren, N.; van Haren, M.; van Meer, M.; Oomen, T. Gaussian Processes for Advanced Motion Control.
IEEJ J. Ind. Appl. 2022, 3, 396–407.
12. Li, B.; Zhou, X.; Ning, Z.; Guan, X.; Yiu, K.-F.C. Dynamic event-triggered security control for networked control systems with
cyber-attacks: A model predictive control approach. Inf. Sci. 2022, 612, 384–398. [CrossRef]
13. Zhao, D.; Xia, L.; Dang, H.; Wu, Z.; Li, H. Design and control of air supply system for PEMFC UAV based on dynamic decoupling
strategy. Energy Convers. Manag. 2022, 253, 115159. [CrossRef]
14. Hagglund, T.; Shinde, S.; Theorin, A.; Thomsen, U. An industrial control loop decoupler for process control applications. Control.
Eng. Pract. 2022, 123, 105138. [CrossRef]
15. Heertjes, M.; van Engelen, A. Minimizing cross-talk in high-precision motion systems using data-based dynamic decoupling.
Control Eng. Pract. 2011, 12, 1423–1432. [CrossRef]
16. Shen, Y.; Cai, W.-J.; Li, S. Normalized decoupling control for high-dimensional MIMO processes for application in room
temperature control HVAC systems. Control Eng. Pract. 2010, 6, 652–664. [CrossRef]
Energies 2024, 17, 2038 18 of 18

17. Luan, X.; Chen, Q.; Liu, F. Equivalent Transfer Function based Multi-loop PI Control for High Dimensional Multivariable Systems.
Int. J. Control Autom. Syst. 2015, 2, 346–352. [CrossRef]
18. Wu, G.; Sun, H.; Zhang, X.; Egea-Alvarez, A.; Zhao, B.; Xu, S.; Wang, S.; Zhou, X. Parameter Design Oriented Analysis of the
Current Control Stability of the Weak-Grid-Tied VSC. IEEE Trans. Power Deliv. 2021, 3, 1458–1470. [CrossRef]
19. Luan, X.; Chen, Q.; Liu, F. Centralized PI control for high dimensional multivariable systems based on equivalent transfer
function. ISA Trans. 2014, 5, 1554–1561. [CrossRef] [PubMed]
20. Rahideh, A.; Bajodah, A.H.; Shaheed, M.H. Real time adaptive nonlinear model inversion control of a twin rotor MIMO system
using neural networks. Eng. Appl. Artif. Intell. 2012, 6, 1289–1297. [CrossRef]
21. van Dael, M.; Witvoet, G.; Swinkels, B.; Oomen, T. Systematic feedback control design for scattered light noise mitigation in
Virgo’s MultiSAS. In Proceedings of the 2022 IEEE 17th International Conference on Advanced Motion Control (AMC), Padova,
Italy, 18–20 February 2022; pp. 300–305.
22. Wang, X.; Yang, B.; Zhu, Y. Optimization of current distribution coefficients to decouple the 6-DOF fine stage of lithographic
equipment. Optik 2016, 20, 9896–9904. [CrossRef]
23. Campestrini, L.; Eckhard, D.; Gevers, M.; Bazanella, A.S. Virtual Reference Feedback Tuning for non-minimum phase plants.
Automatica 2011, 8, 1778–1784. [CrossRef]
24. Wang, Y.; Zhang, J.; Zhang, H.; Xie, X. Adaptive Fuzzy Output-Constrained Control for Nonlinear Stochastic Systems with Input
Delay and Unknown Control Coefficients. IEEE Trans. Cybern. 2021, 11, 5279–5290. [CrossRef] [PubMed]

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.