energies

Article
Model Reference Adaptive Control and Fuzzy Neural Network
Synchronous Motion Compensator for Gantry Robots
Chin-Sheng Chen and Nien-Tsu Hu *
Graduate Institute of Automation Technology, National Taipei University of Technology, Taipei 10608, Taiwan;
[email protected]
* Correspondence: [email protected]

Abstract: A model reference adaptive control and fuzzy neural network (FNN) synchronous motion
compensator for a gantry robot is presented in this paper. This paper proposes the development and
application of gantry robots with MRAC and FNN online compensators. First, we propose a model
reference adaptive controller (MRAC) under the cascade control method to make the reference model
close to the real model and reduce tracking errors for the single axis. Then, a fuzzy neural network
compensator for the gantry robot is proposed to compensate for the synchronous errors between
the dual servo motors to improve precise movement. In addition, an online parameter training
method is proposed to adjust the parameters of the FNN. Finally, the experimental results show that
the proposed method improves the synchronous errors of the gantry robot and demonstrates the
methodology in this paper. This study also successfully integrates the hardware and successfully
verifies the proposed methods.

Keywords: fuzzy neural network; gantry robot; model reference adaptive controller; online parameter






Citation: Chen, C.-S.; Hu, N.-T.
Model Reference Adaptive Control 1. Introduction
and Fuzzy Neural Network Gantry robots have been widely used in manufacturing industries such as high-
Synchronous Motion Compensator precision motion control, precision manufacturing, circuit assembly, microelectronics, and
for Gantry Robots. Energies 2022, 15, inspection [1–3]. The gantry robot is composed of a manipulator on an overhead system,
123. https://doi.org/10.3390/ and two motors are installed on two parallel linear guides to drive the moving platform.
en15010123 However, due to various factors, such as unbalanced forces on both sides, various dis-
Academic Editors: Frantisek Durovsky turbances in the driving process will cause synchronous errors between the two motors.
and Pavol Makys The consequence of these synchronous errors will not only cause system jitter and affect
the quality of the workpiece but also cause the work process to stop due to overcurrent
Received: 12 November 2021
protection. These undesirable effects are very much in need of control and improvement
Accepted: 21 December 2021
for high-speed and high-precision manufacturing. Therefore, how to effectively control the
Published: 24 December 2021
synchronous errors of a gantry position platform has become a key issue.
Publisher’s Note: MDPI stays neutral Typical methods (see Figure 1) to synchronize the motion in gantry robot control
with regard to jurisdictional claims in systems include (1) the cascade control method, and (2) the parallel control method [4,5].
published maps and institutional affil- Both methods use two control loops to control the motors separately. The first control
iations. method divides the two control loops into a master loop and a slave loop, and the reference
command is only provided to the master loop. The master loop has a master motor, and
the slave loop has a slave motor. If the master loop encounters a disturbance, the slave loop
also reflects the disturbance. The second control method has the two control loops follow
Copyright: © 2021 by the authors.
the same reference position command. Since the inevitable differences between the two
Licensee MDPI, Basel, Switzerland.
subsystems (including motors, motor drives, etc.) are not taken into account, this method
This article is an open access article
distributed under the terms and
usually exhibits poor performance.
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).

Energies 2022, 15, 123. https://doi.org/10.3390/en15010123 https://www.mdpi.com/journal/energies

 Energies 2022, 15, x FOR PEER REVIEW 2 of 18
Energies 2022, 15, 123 2 of 17

Pcmd V1 1
P1 V1 1
P1
S S
Pcmd

V2 1
P2 V2 1
P2
S S

(a) (b)
Figure1.1.Typical
Figure Typicalmethods
methodstotosynchronize
synchronizethe
themotion
motionin
ingantry
gantryrobots:
robots: (a)
(a) cascade
cascade control
control method;
method;
(b) parallel control method.
(b) parallel control method.

Recently,compensators
Recently, compensators[6–9] [6–9]and
andcross-coupling
cross-couplingtechnology
technologyhave haveoften
oftenbeenbeenused
usedin in
machine tools
machine tools andandmultiaxis
multiaxis motion
motionapplications
applications [10,11]
[10,11] totosolve
solvesynchronization
synchronization control control
problemsand
problems andimprove
improvecontrol
controlperformance
performance[12–15].[12–15]. InIn cross-coupling
cross-coupling control,
control, each
each loop
loop
also considers the position and speed errors of another motor to
also considers the position and speed errors of another motor to evaluate its own control evaluate its own control
performance. However,
performance. However, these these methods
methods cannot
cannot provide
provide sufficient
sufficient robustness
robustness becausebecause the the
controlparameters
control parametersare areselected
selectedthrough
throughtrial trialand
anderror.
error.
Fuzzyneural
Fuzzy neuralnetworks
networks(FNNs)(FNNs)[16–26]
[16–26]isisan an intelligent
intelligenttechnology
technologythat thatcombines
combinesthe the
advantages of
advantages of fuzzy
fuzzy logic
logic and
and neural
neural network
network systems.
systems. The FNN system is a straightfor- straightfor-
ward
wardimplementation
implementation of of aa fuzzy
fuzzy inference
inference system
system with
with aa four-layered
four-layered network
network structure.
Generally,
Generally, the advantage of FNN systems lies in that: (1) the FNN system can
the advantage of FNN systems lies in that: (1) the FNN system can automati-
automati-
cally
callyidentify
identifyfuzzy
fuzzy logic
logic rules,
rules,(2)(2)the
theparameters
parametersof ofthe
theFNNFNNsystem
systemhave haveclear
clearphysical
physical
meanings,
meanings,(3) (3)the
theFNNFNNsystem
systemcan canincorporate
incorporatelinguistic
linguisticinformation
information(in (inthe
theform
formof of fuzzy
fuzzy
IF-THEN
IF-THEN rules),
rules), andand(4)(4)the
thedesired
desiredperformance
performancecan canbebeobtained
obtainedunder
underfewer feweradjustable
adjustable
parameters
parameters than than inin neural
neural networks.
networks. Generally,
Generally, FNNs FNNs cancan bebedivided
dividedinto intothetheMamdani
Mamdani
type and Takagi–Sugeno–Kang (TSK) type. Since the TSK
type and Takagi–Sugeno–Kang (TSK) type. Since the TSK model can incorporate mathe- model can incorporate math-
ematical
matical knowledge about the controlled plant and can use control theory to analyse its
knowledge about the controlled plant and can use control theory to analyse its
behaviour,
behaviour,the theTSKTSKmodel
modelisisthe themost
mostcommonly
commonlyused usedFNNFNNmethod.
method.
Recent
Recent research
research on ongantry
gantry robots,
robots,suchsuchas asliterature
literature[27],
[27],discusses
discussesthe thesynchronous
synchronous
control
controlbased
basedononfuzzy fuzzy single
singleneuron
neuronPIDPID cross-coupling
cross-coupling controller, literature
controller, [28] discusses
literature [28] dis-
the suppression of the rotational motion of cross-coupled gantry
cusses the suppression of the rotational motion of cross-coupled gantry stage and litera- stage and literature [29]
proposes a new algorithm to identify the parameters of the synchronous
ture [29] proposes a new algorithm to identify the parameters of the synchronous dual- dual-drive ball
screw gantry system. This paper proposes the development
drive ball screw gantry system. This paper proposes the development and application of and application of gantry
robots
gantrywith
robots MRAC and FNN
with MRAC andonline compensators.
FNN online For the
compensators. Forcontroller design,
the controller a cascade
design, a cas-
control method with an MRAC controller is proposed to ensure
cade control method with an MRAC controller is proposed to ensure the tracking require- the tracking requirements
of single
ments of axis
singlecontrol. Then, Then,
axis control. the purpose
the purpose of theofFNN compensator
the FNN compensatoris to is eliminate the
to eliminate
synchronous
the synchronous errors between
errors between the the
dual servo
dual servomotors.
motors. ToTo improve
improvethe thelearning
learningability
abilityof of
FNN, an online parameter training method is proposed to adjust the parameters of the
FNN, an online parameter training method is proposed to adjust the parameters of the
FNN. This paper has developed and successfully completed the theoretical and technical
FNN. This paper has developed and successfully completed the theoretical and technical
feasibility of the proposed method through various experimental comparisons.
feasibility of the proposed method through various experimental comparisons.
The rest of the paper is organized as follows. The gantry robot servo system description
The rest of the paper is organized as follows. The gantry robot servo system descrip-
is given in Section 2. Section 3 presents the proposed synchronous control methods of
tion is given in Section 2. Section 3 presents the proposed synchronous control methods
gantry robots. The FNN online compensator with two inputs and one output is developed
of gantry robots. The FNN online compensator with two inputs and one output is devel-
to compensate for the synchronous errors. The experimental results are illustrated in
oped to compensate for the synchronous errors. The experimental results are illustrated
Section 4 to demonstrate the methodology proposed in this paper.
in Section 4 to demonstrate the methodology proposed in this paper.
2. The Structure and Mathematical Model of the Gantry Robot System
2. The Structure and Mathematical Model of the Gantry Robot System
Figure 2 shows the gantry robot system used in this paper. It consists of two rotating
servoFigure
motors,2 shows the gantry
guideways, robot
and ball system used in this paper. It consists of two rotating
screws.
servo motors, guideways, and ball screws.
 S e rv o m o to r S e rv o m o to r
2 1

Energies 2022,
Energies 15,15,
2022, x FOR
123 PEER REVIEW 3 of3 18
of 17

S e rv o m o to r S e rv o m o to r
2 G u id e - w a y 1
G u id e - w a y
2 1
B a ll- s c r e w 2

B a ll- s c r e w 1

A x is 1 A x is 2

Figure 2. The hardware structure

G u id e - w ay ofGthe
u i dgantry
e - w a y robot.
2 1
B a ll- s c r e w 2
In traditional cascade control, the inner current control loop usually has a high gain
B a ll- s c r e w 1
to minimize the current error over the system operating range. The bandwidth of the cur-
A x i sloop
rent 1 is usually well over 2 kHz, and the effect A x i sof2 back EMF is eliminated. The elec-

tronic dynamics are so fast that the

Figure 2. The hardware structure of the gantry robot. transfer function of the AC servo drive can be treated
Figure 2. The hardware structure of the gantry robot.
as a constant current gain. For simplicity, the system equations of each axis of the gantry
robot can In traditional
be shown cascade
as control, the inner current control loop usually has a high gain to
In traditional cascade control, the inner current control loop usually has a high gain
minimize the current error over the system operating range. The bandwidth of the current
to loop
minimize the current error over the vi + Di vi operating
M system + FLi = K tirange. ui The bandwidth of the cur- (1)
is usually well over 2 kHz, andi the effect of back EMF is eliminated. The electronic
rent loop
dynamics is usually
are well over
soequivalent
fast that the 2 kHz,
transfer and the
function effect of
of the D backAC is EMF
servo is eliminated.
drive can be The elec-
treatedfric-
as a
where M i is the mass of the mechanism; the equivalent viscous
tronic
constantdynamicscurrent aregain.
so fast
Forthat the transfer
simplicity, the systemfunction of thei AC
equations servoaxis
of each drive
of can be treated
the gantry robot
tion;
as K is the
a constant torque constant; FLi is the
current external
systemdisturbance
equations ofterm; eachand
axisuiof
is the
the gantry
control
can betishown as gain. For simplicity, the
effort.can
robot Then, the undermined
be shown as plant .
Mis i vi + Di vi + FLi = Kti ui (1)
Mthe v
vi=+mechanism;
Di vi i + FLi == D 1
K ti uisi the equivalent viscous friction;
i (s i)
where Mi is the equivalent massGof (1)Kti
(2)
K ti ui − FLi M iis + Di
is the torque constant;
where M i is the equivalent F is the external disturbance term;
Li mass of the mechanism; D is the iequivalent viscous fric- and u is the control effort. Then,
i
the The velocity PI
undermined controller
plant is (shown in Figure 3) is used to eliminate external disturb-
tion; K
ances. Then,ti
is the torque constant;
the identified plant from F Li is the
the control effort to the term;
external disturbance andresponse
velocity ui is the control
can be
effort.
simplified Then,tothe undermined plant is vi 1
Gi (s) = = (2)
vG Kti uiv− FLiK M1i s + K Di
i
i=( sK)ti=Gi ( s ) = =ti = i
i
(2)
(3)
The velocity PI controller (shown ui in ui − M
K tiFigure FLii s3)+isD is +
Miused τ iDsto
i+ 1
eliminate external disturbances.
Then,
The the
Then, identified
velocity
the input plant from(shown
PI controller
command, the
showncontrol effort
Figureto
ininFigure the
3)4a,
is is velocity
used response can
to eliminate
simultaneously be
external
fed intosimplifiedaxisto
disturb-
each
ances.
as vcmd1Then, and vthecmd2,identified plant from
and the responses of the
Axes control
1 and effort2 are shown to the velocity
in Figureresponse
4b. The candatabe of
simplified vi Kti Ki
Figure 4a,btoare utilized to identify = Kthe ti iG parameters
( s ) = in Equation
= (3); then, the identified re-
(3)
ui Mi s + Di τi s + 1
sults of each axis are indicated vi
in Table 1. K ti Ki
= K ti Gi ( s ) = = (3)
ui M i s + Di τ i s + 1
Then, the input command, shown in Figure 4a, is simultaneously fed into each axis
as vcmd1 and vcmd2, and the responses of Axes 1 and 2 are shown in Figure 4b. The data of
vcmd1 + － u1 v1
PI K t1
Figure 4a,b are utilized to identify G 1in Equation (3); then, the identified re-
the parameters
sults of each axis are indicated in Table 1.
F L1
Mechanism Velocity
Velocity
coupling Response
Command
F L2
vcmd1 +－ u1 v1
PI K t1 G1
vcmd2 u2 v2
+
PI K t2 G2
－
F L1
Mechanism Velocity
Velocity
coupling Response
Command
F L2
Figure 3. The PI controller loop for parameter identification.
Figure 3. The PI controller loop for parameter identification.
vcmd2 u v2
+
Then, thePIinput2 command,
K t2 shown in Figure
G 2 4a, is simultaneously fed into each axis
－
as vcmd1 and vcmd2 , and the responses of Axes 1 and 2 are shown in Figure 4b. The data
of Figure 4a,b are utilized to identify the parameters in Equation (3); then, the identified
results of each axis are indicated in Table 1.
Figure 3. The PI controller loop for parameter identification.
 Energies 2022, 15, x FOR PEER REVIEW 4 of 18
Energies 2022, 15, 123 4 of 17

60
50 Vcmd V1
X: 0.001
Y: 48.8 V2
40
40
30
V e lo city(m m /se c)

Velocity(m m /sec)
20 20
10

0 0

-10

-20 -20

-30
-40
-40 X: 6
Y: -48.8

-50
-60
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(sec) Time(sec)
(a) (b)
Figure
Figure4.4.(a)
(a)Velocity
Velocity command;
command; (b)
(b) velocity
velocity response.
response.

Table 1. Identified parameters of gantry stage.

Table 1. Identified parameters of gantry stage.
Gain Value Time Constant Value
Gain Value Time Constant Value
K1 (mm
K1 ( smm ) ) 393.79
393.79 ττ11(s)
(s) 0.06
0.06
smm⋅ volt
·volt
K2 ( s·volt ) 384.49 τ2 (s) 0.06
K 2 (mm ) 384.49 τ 2 (s) 0.06
s⋅ volt
3. Proposed Control System
3.1. Controller Design for The Single Sxis
3. Proposed Control System
To reduce the tracking error of the single axis position, we adopt the cascade control
3.1. Controller Design
thefor The Single Sxis as shown in Figure 5. In this method, G pi z−1 is


method to design single axis control,
the discrete model
To reduce of a single
the tracking axisofplant,
error whichaxis
the single canposition,
be represented
we adoptas the cascade control
method to design the single axis control, as shown in Figure 5. In this method, Gpi ( z −1 ) is
  
1 − e − Ts K

bz−1
−1 i
the discrete model of aGsingle
p z axis = plant,
Z which· can be represented
= as (4)
s τi s + 1 1 + az−1
1 − e −Ts K i  bz −1
where ( )
G p z −1 = Z  ⋅
τB·iTs =
s + 1  1 + az −1
(4)
 s
−
a = −e J
where − B·JTs
 (5)
Energies 2022, 15, x FOR PEER REVIEW
b = (Kt /B) 1 − e
B ⋅Ts 5 of 18
−
a = −e J
(5)
and Ts is the sampling time.
b = (K t B )1 − e
− B⋅Ts 
J

 
and TPsdi is the sampling time. u V Pai
K vii −1 Ts
K ppi G pi ( z )
1 − z −1 1 − z −1

Gp( z ) K vpi
K pp
K vp
K vi
i

Figure 5. Position
Figure loop
5. Position control
loop of of
control single axis.
single axis.

The inner velocity loop and outer position loop use the integral Kvii, proportional Kvpi,
and proportional Kppi controllers. To make the reference model close to the real model, we
adopted the adaptive control to let the reference model approach to real model. At pre-
sent, there are two main design architectures in the adaptive control. One is the self-tuning
controller (STC), and the other is the model reference adaptive controller (MRAC). Here,
 K pp
K vp
K vi
i
Energies 2022, 15, 123 5 of 17
Figure 5. Position loop control of single axis.

TheTheinner velocity
inner looploop
velocity and outer position
and outer loop use
position looptheuse
integral Kvii, proportional
the integral Kvpi,
Kvii , proportional
andKproportional K controllers. To make the reference model close to
vpi , and proportional Kppi controllers. To make the reference model close to the
ppi the real model, wereal
adopted
model, the
weadaptive
adoptedcontrol to let the
the adaptive reference
control model
to let the approach
reference to approach
model real model. to At
realpre-
model.
sent,
Atthere are two
present, main
there are design
two main architectures in the adaptive
design architectures in thecontrol.
adaptive One is the self-tuning
control. One is the self-
controller
tuning (STC), and(STC),
controller the otherandisthe
theother
model reference
is the modeladaptive
referencecontroller (MRAC). Here,
adaptive controller (MRAC).
we Here,
adopted wethe MRACthe
adopted [30],
MRACas shown in Figure
[30], as shown 6. inThe basic
Figure 6. concept of concept
The basic MRAC isoftoMRAC plan is
the to
performance of the control system in a reference model, and the design
plan the performance of the control system in a reference model, and the design of the of the entire
feedback
entire control
feedbacksystem
controlis tosystem
matchisthetoplanned
match the reference
planned model to achieve
reference modelthe toexpected
achieve the
expected
system system response.
response.

Reference model
bn z −1 Vm
1 + an z −1
+ Verr
-
d
u + V
bz −1
+ + + 1 + az − 1
ucomp

+ +
K2 K1
+
K3 Sgn

Reference model adaptive compensator

Figure 6. The model reference adaptive control method for the single axis system.
Figure 6. The model reference adaptive control method for the single axis system.
From Figure 6, the adaptive force ucomp can be obtained by the error Verr, and the
From Figure
reference 6, the
output Vmadaptive force uof
is composed cannominal
the
comp be obtained by the error
parameters Verr,band
an and the refer-
n . The adaptive
enceparameters
output Vm is composed of the nominal parameters a and b . The adaptive parameters
K1 , K2 , and K3 are derived from the Lyapunov stability criterion. The detailed
n n

K1, K2, and K3 are

derivation derived
process, from
please seethe Lyapunov
Appendix A. stability criterion. The detailed derivation
process, please see Appendix A.
N
K1 ( N ) = B1 N∑ Verr (i ) · V (i ) + C1 · Verr ( N ) · V ( N )

K1 ( N ) = B1 i=0Verr (i ) ⋅ V (i ) + C1 ⋅ Verr ( N ) ⋅ V ( N )
Ni = 0
K2 ( N ) = B2 ∑ Verr (i ) × u(i ) + C2 × Verr ( N ) × u( N ) (6)
i =0
N
K3 ( N ) = B3 ∑ Verr (i ) × sgn(V (i )) + C3 × Verr ( N ) × sgn(V (i ))
i =0

where the positive constants are Bj and Cj , and j = 1, 2, 3, could be well tuned under the
model reference adaptive control method. Therefore, the transfer function of the velocity
inner loop can be represented as

Kvpi bTs z
TFv (z) = (7)
z2 + ( a − 1 + Kvpi bTS + Kvii bTs )z + (− a − Kvii bTs )

We can use the pole-placement method to design the controller for this second-order
system. Let the two parameters ξ and ωn be similar to the damping ratio and natu-
Energies 2022, 15, 123 6 of 17

ral frequency of the standard second-order system; then, the parameters of the velocity
controller are
1+e−2ξωn Ts −2e−ξωn TS cos(ωd Ts )
Kp = bTs
(8)
(e−2ξωn Ts + a)
Kd = −b
The transfer function of the outer position loop can be simplified as

Pai K ppi Ts
=
(9)
Pdi z + K ppi Ts − 1

Here, the bandwidth of the position loop can be well designed according to the rule
of cascade control, and then the parameter Kppi can be easily obtained. For more details,
please refer to [30]. Here, we adopt the important design results of [30].

3.2. FNN Synchronous Motion Compensator

Although the abovementioned single axis controller can reduce the single axis position
tracking error, the synchronous error between the dual motors is caused by various factors,
such as unbalanced forces on both sides, various disturbances in the driving process and
environmental uncertainty. This is an unavoidable situation. Therefore, we developed an
FNN online compensator combined with MRAC, as shown in Figure 7. This FNN online
compensator can compensate for the synchronous error online. Here, two methods are
proposed: (1) parallel control and (2) parallel master–slave control. When the linear guides
are not parallel with respect to each other axis in installation, the second method 7will
Energies 2022, 15, x FOR PEER REVIEW be
of 18
applied to avoid the mechanical coupling force being yielded by achieving the synchronous
motion in position.

K13 K13

K12 K11 K12 K11

Pcmd Vc1 P1 P1
G p1 ( z −1 ) Ts Pcmd Vc1 Ts
G p1 ( z −1 )
1 − z −1 1 − z −1
∏

∏
∏

∏
∏

∏
2
T

2
T
∏

∏
∏


∏

∏

∏
T 1
∏

T 1
∏
∏

∏
∏

Vc 2 Ts P2
G p 2 ( z −1 ) Vc 2 Ts P2
1 − z −1 G p 2 ( z −1 )
1 − z −1

K 22 K 21
K 22 K 21
K 23
K 23

(a) (b)
Figure
Figure 7.
7. The proposedsynchronous
The proposed synchronouscontrol
control methods
methods of gantry
of the the gantry
robot:robot: (a) parallel
(a) parallel control control
method;
method; (b) parallel master–slave control
(b) parallel master–slave control method. method.

Adopting the concept of fuzzy neural network technology, the proposed FNN com-
pensator for MRAC can be constructed, as shown in Figure 8.

y 4
o


 K 22 K 21
K 23
K 23

(a) (b)
Energies 2022, 15, 123 7 of 17
Figure 7. The proposed synchronous control methods of the gantry robot: (a) parallel control
method; (b) parallel master–slave control method.

Adopting
Adopting the
the concept
concept of
of fuzzy
fuzzy neural
neural network
network technology,
technology, the
the proposed
proposed FNN
FNN com-
com-
pensator for MRAC can be constructed, as shown in Figure
pensator for MRAC can be constructed, as shown in Figure 8.8.

y 4
o


w 4
ko
o

y 3
k

∏ ∏ ∏ ∏ ∏ ∏ ∏ ∏ ∏
x 3
k
k
w 3
k

y 2
ij

x 2
i
j
y 1
i

T1
T2
x 1
i
i
x =e
1
1 s
x = e
1
2 s

Figure 8.
Figure The proposed
8. The proposed FNN
FNN compensator
compensator for
for MRAC.
MRAC.

Next, we
Next, we introduce
introduce the
the important
important concepts
concepts of
of an
an FNN.
FNN. An
An FNN
FNN is
is a
a network
network with
with
fuzzy inference characteristics implemented by a four-layer neural network.
fuzzy inference characteristics implemented by a four-layer neural network. The following
The follow-
will describe the structure, corresponding operation and learning process of an FNN.
ing will describe the structure, corresponding operation and learning process of an FNN.
Layer 1: Input layer
Each node in this layer represents the input node of each input linguistic variable
and corresponds to an input variable. This means that the nodes in this layer are only
responsible for passing the input signal to the next linguistic layer. There are two input
variables of the FNN compensator in our proposed synchronous control method. One is
the position synchronous error x11 = es = P1 − P2 , and the other is the velocity synchronous
.
error x21 = es = V1 − V2 . P1 , P2 are the position responses, and V 1 and V 2 are the velocity
responses corresponding to Axes 1 and 2, respectively. Therefore, the node output of this
layer is as follows:
y1i ( N ) = Ti · xi1 ( N ), i = 1, 2 (10)

Layer 2: Linguistic layer (Membership layer)

The nodes in layer 2 are called membership nodes, and the role of this layer is to
perform the membership function of each node. The Gaussian function is used here as a
membership function. Then,
" 2#
( xi2 − mij )
y2ij ( N ) = exp − , i = 1, 2 ; j = 1, 2, . . . , M (11)
(σij )2

where mij and σij denote the mean and the standard deviation, respectively, of the Gaussian
functions of the jth term of the ith input linguistic variable; M is the number of rules.

Layer 3: Rule layer

The rule nodes are located in layer 3, which includes the rule layer and fuzzy inference
mechanism. For each layer 3 node, there is at most one previous link from the layer 2 node
Energies 2022, 15, 123 8 of 17

of the language variable. The nodes in this layer are denoted by ∏, which are multiplied
by the input signal from layer 2. Then, for the jth rule node,

2
y3k ( N ) = ∏ w3k · y2ij ( N ) i = 1, 2 j = 1, 2, . . . , M (12)
i =1

where y3k and w3k represent the output and weight of the rule layer, respectively. Here, w3k is
designed to be 1.

Layer 4: Output layer

The layer 4 contains output variable nodes. This layer performs defuzzification to
obtain the numerical output y4o . The operation of layer 4 is

y4o ( N ) = ∑ w4ko · y3k (13)

k

where y4o ( N ) is the output of the proposed FNN compensator for MRAC and the link weight
w4ko is the output strength. In this paper, M is set to 3, which means that the linguistic layer
has 6 nodes, and the rule layer has 9 nodes.

3.3. On-Line Learning Algorithm

The parameter learning algorithm is based on a supervised learning law to train the
system. This method is the same as the derivation of the back propagation algorithm,
adjusting the link weight in the output layer to minimize the given energy function.

1 2 1. 2
E= es + es (14)
2 2
Next, we describe the update laws of the parameters in the FNN. First, the error term
to be propagated is given by
. .
∂E 1 ∂E ∂es 1 ∂E ∂es 1 ∂es 1 . ∂es
δo4 = − =− − . = − es 4 − es 4 (15)
∂y4o 2 ∂es ∂y4o 2 ∂es ∂y4o 2 ∂yo 2 ∂yo
.
The exact calculation of the Jacobian of systems ∂es /∂y4o and ∂es /∂y4o , which are
contained in ∂E/∂y4o , cannot be determined due to the uncertainties of the plant dynamics,
such as parameter variations and external disturbances. To overcome this problem and to
increase the online learning rate of the network parameters, the derivatives ∂es /∂y4o and
. .
∂es /∂y4o are approximated by the ratio of the signs of the changes in es and es with respect to
y4o , respectively. In this study, the compensated force y4o is used to reduce the synchronous
.
errors of position es and velocity es . Therefore,
 
∂es ∂es
∂y4o
= sgn 4 = −1
.  ∂y. o  (16)
∂es ∂es
∂y 4 = sgn ∂y 4 = − 1
o o

then, Equation (15) can be rewritten as

.
1 ∂es 1 . ∂es 1 .

δo4 = − es 4 − es 4 = es + es (17)
2 ∂yo 2 ∂yo 2

and the update of w4ko is

 " 4 #
∂E ∂E ∂yo
∆w4ko = − ηw 4 = − ηw 4 = ηw δo4 y3k (18)
∂wko ∂yo ∂w4ko
 Energies 2022, 15, 123 9 of 17

where ηw is the learning rate parameter of the link weights. Then, the weights are
updated as
w4ko ( N + 1) = w4ko ( N ) + ∆w4ko (19)

3.4. Stability Analysis

Refer to [31–33], based on the discrete Lyapunov function analysis, we consider the
energy function (14) as the discrete Lyapunov function. The change in the Lyapunov
function can be written as
∆V ( N ) = V ( N + 1) − V ( N ) (20)
then, according to [31–33], Equation (20) will be derived as follows.

Energies 2022, 15, x FOR PEER REVIEW V ( N + 1) = V ( N ) + ∆V ( N ) 10 of 18

" # " # " #

M M M 2
∂V ( N ) ∂V ( N ) ∂V ( N ) ∂V ( N )
≈ V(N) + ∑ ∂w4ko
∆wko + ∑
4
∂w3k
∆wk + ∑ ∑
3
∂mij
∆mij +
∂σij
∆σij (21)
j =1 j =1 j =1 i =1
V ( N )ε V ( N )ε
V ( N + 1) ≈ ε (η w + ηθ + η m + ησ ) = + +
M By the 4design of4 the   Mrate
learning 2
parameters [31], the convergence of Equation (21)
4 
 jcan
4 2
((∂V ( N ) ∂yo )(∂yo ∂wko )) + ε 
be guaranteed. Here, we
4
 have
 4 4 3 3 3 3 3 2
((∂V ( N ) ∂yo )(∂yo ∂wk yk )(∂wk yk ∂wk )) + ε 
 j =1omitted some mathematical processes, and  listed the
=1 i =1
important result as follows:
V ( N )ε V ( N )ε
+ (22)
 ε(ηw + ηθ + ηm +4 ησ ) 4= 2M 2
V ( N + 1) ≈ M 2 V ( N )ε
  M 2
+ M 2 V ( N )ε
 #+
 
" # "
2 4 4 2 2 2
4 ((∂V ( N ) ∂yo )(∂yo ∂4yij∑)(((∂y∂V
ij ( ∂
Nm) ))
/∂y
ij o
4 )(+
∂yε 
4 /∂w
o

44)) 2
+ ε ((
4 ∂ V
∑ ∑ ( N )
(( ∂
∂Vy(
o N)(
) ∂y
/∂y 4 )(∂
oo y
∂y )(
4 /∂w
ijo ∂y3 y3 )(
k ij k
∂σ
∂w 3y)) +
3 /∂w
ijk k ε 2 + ε
3 ))

 j =1 i =1   j =1 i =1 
ko k
j = 1 j = 1 i = 1
V ( N )ε V ( N )ε
" # + " # (22)
V (N ) V (N ) 4 V∑( N
M 2
) V (N )
∑ ((+∂V ( N )/∂y=4o )( 4 2 2 2 M
4 ∑ ∑
2 2
((∂V ( N )/∂y4o )(∂y4o /∂y2ij )(∂y2ij /∂σij )) +ε
< + + j =1 i =1
) ij )(∂yij /∂mij )) +ε
( No /∂y
V ∂y j =1 i =1
4 4 4 4 V(N) V(N) V(N) V(N)
4 < 4 + + + = V(N)
From Equation (22), it means that4the synchronous error of the gantry robot will grad-
4
From Equation
ually converge to zero. (22), it means that the synchronous error of the gantry robot will
gradually converge to zero.
4. Experimental Results
4. Experimental Results
Figure 9 shows the experimental system of the gantry robot control system in this
Figure 9 shows the experimental system of the gantry robot control system in this
study. In the experimental system, the single axis controller and the proposed FNN online
study. In the experimental system, the single axis controller and the proposed FNN online
compensator for MRAC are implemented in the PC.
compensator for MRAC are implemented in the PC.

Figure
Figure 9. 9.
TheThe hardware
hardware setup
setup of of
thethe PC-based
PC-based control
control system
system inin
thethe gantry
gantry robot.
robot.

InIn this
this study, msecsampling
study,a a1 1msec samplingrate
rateis isadopted
adoptedforforthe
theencoder
encoder interface
interface and
and the
the
execution of the control algorithm. For comparison, the results of experiments
execution of the control algorithm. For comparison, the results of experiments for low- for low-
frequency (1/5 Hz) and high-frequency (2/3 Hz) sinusoid position commands with the
same strokes ( ± 36.6 mm) are used to verify the synchronous control performance of cas-
cade synchronous control, parallel synchronous control without a synchronous compen-
sator and the proposed control methods in Figure 7. Here, we set the parameters of the
 Energies 2022, 15, 123 10 of 17

frequency (1/5 Hz) and high-frequency (2/3 Hz) sinusoid position commands with the
same strokes (±36.6 mm) are used to verify the synchronous control performance of cascade
synchronous control, parallel synchronous control without a synchronous compensator
and the proposed control methods in Figure 7. Here, we set the parameters of the proposed
FNN compensator as w3k = 1 and M = 3, which means that the linguistic layer has 6 nodes,
and the rule layer has 9 nodes. Furthermore, the learning rate ηw is designed to 0.001 to
let the convergence of Equation (21) can be guaranteed. To provide an overall evaluation,
Energies 2022, 15, x FOR PEER REVIEW two performance indices, the sum of absolute synchronous error and root mean square
11 of 18
synchronous error, are defined as

ESAE = ∑ |es |
1 r Ni2
E RMS =  ( es1) ∑ (es )2
ERMSN i=N N
(23)
i
i
Ni
Figure 10 shows the synchronous error of the cascade synchronous control method,
Figure1a,
shown in Figure 10 by
shows the the
feeding synchronous
low-frequencyerrorcommand
of the cascade synchronous
(1/5 Hz). The maximumcontrol method,
syn-
shown
chronous in Figure
error 1a, by feeding
is approximately ± 0.92
the low-frequency
mm even when command (1/5 Hz).
the federate is low.TheInmaximum
this
method,synchronous
the controlerror is approximately
efforts, depicted in Figure ±0.9211, mm eventhat
indicate when
theythearefederate is low. inIn this
not consistent
the phase and magnitude due to the servo lag between the master and slave axes not
method, the control efforts, depicted in Figure 11, indicate that they are andconsistent
that
in the
the effect phase
will yieldand magnitude
a large synchronousdue toerror.
the servo
Hence,lagthis
between
methodtheismaster and slave
not suitable axes and
for gan-
that synchronous
try robot the effect willcontrol.
yield a large
In thesynchronous error. Hence,
following discussion, the this method isofnot
performance thesuitable
two for
gantry
proposed robot synchronous
methods will be comparedcontrol.under
In thedifferent
following discussion,
conditions: (1)the performance
without of the two
compensa-
proposed methods will be compared under different conditions:
tion; (2) with FNN compensation; and (3) with FNN and MRAC compensation. (1) without compensation;
(2) with FNN compensation; and (3) with FNN and MRAC compensation.

1
Synchronous Error
0.8

0.6
Synchronous Error(mm)

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
0 5 10 15
Time(sec)

FigureFigure
10. The10. The synchronous
synchronous error
error of the of the cascade
cascade synchronous
synchronous controlcontrol method.
method.

Figure 11. Control efforts of the cascade synchronous control method.

 -0.6

-0.8

-1
0 5 10 15
Time(sec)
Energies 2022, 15, 123 11 of 17

Figure 10. The synchronous error of the cascade synchronous control method.

11.Control
Figure11.
Figure Controlefforts
effortsof
ofthe
thecascade
cascadesynchronous
synchronouscontrol
controlmethod.
method.

4.1.
4.1. Parallel
Parallel Synchronous
SynchronousControl
Control
There
There are two kinds ofofposition
are two kinds positioncommands
commands mentioned
mentioned above
above fedfed to parallel
to the the parallel
Energies 2022, 15, x FOR PEER REVIEW 12 of syn-
18
synchronous control method to test the synchronous performance. For
chronous control method to test the synchronous performance. For low-frequency com- low-frequency
commands, the synchronous error and cost function shown in Equation (14) are depicted in
mands, the synchronous error and cost function shown in Equation (14) are depicted in
Figure 12a,b. Figure 13a,b shows the case of a high-frequency command. When a constant
Figure 12a,b. Figure 13a,b shows the case of a high-frequency command. When a constant
corresponding to low-N-m
disturbance 0.02375 and is
high-frequency
applied to Axis commands, respectively,
2 during the period of 2andto 3the synchronous
s and 1.2 to 1.6 s
disturbance 0.02375 N-m is applied to Axis 2 during the period of 2 to 3 s and 1.2 to 1.6 s
errors will be magnified
corresponding without
to low- and a compensator.
high-frequency Figure 14a,b
commands, shows the
respectively, andsynchronous
the synchronous er-
rors with
errors respect
will to low- and
be magnified high-frequency
without commands.
a compensator. FigureIn contrast,
14a,b shows these figures also
the synchronous
show
errorsthat therespect
with synchronous
to low-errors will be suppressed
and high-frequency to be similar
commands. to the condition
In contrast, without
these figures also
disturbances by synchronous compensators. The detailed results of the two
show that the synchronous errors will be suppressed to be similar to the condition withoutperformance
indices in Equation
disturbances (20) are shown
by synchronous in Tables 2The
compensators. anddetailed
3. The results
sampling points
of the twobetween
performancethe
dotted
indices lines shown in(20)
in Equation Figure
are 14a,b,
shownwhere disturbances
in Tables are applied,
2 and 3. The sampling arepoints
calculated
betweenin the
the
case of “with
dotted lines disturbances”.
shown in Figure 14a,b, where disturbances are applied, are calculated in the
case of “with disturbances”.
0.25 0.014
without compensation without compensation
with FNN compensation with FNN compensation
0.2 with FNN+MRAC compensation 0.012 with FNN+MRAC compensation
Synchronous Error(m m )

0.15 0.01
Error F un ctio n(E)

0.1 0.008

0.05 0.006

0 0.004

-0.05 0.002

-0.1 0
0 5 10 15 0 5 10 15
Time(sec) Time(sec)

(a) (b)
Figure
Figure12.
12.(a)
(a)The
Thesynchronous
synchronouserror
errorunder
underlow-frequency
low-frequencycommands;
commands;(b)
(b)the
thecost
costfunction
functionunder
under
low-frequency commands.
low-frequency commands.

0.15 0.012
without compensation without compensation
with FNN compensation with FNN compensation
0.1 with FNN+MRAC compensation with FNN+MRAC compensation
0.01
Synchro nous Error(m m )

0.05
0.008
Erro r F unction(E)

0
0.006
-0.05

0.004
-0.1
 -0.1 0
0 5 10 15 0 5 10 15
Time(sec) Time(sec)

(a) (b)
Figure 12. (a) The synchronous error under low-frequency commands; (b) the cost function under
Energies 2022, 15, 123 12 of 17
low-frequency commands.

0.15 0.012
without compensation without compensation
with FNN compensation with FNN compensation
0.1 with FNN+MRAC compensation with FNN+MRAC compensation
0.01
Synchro nous Error(m m )

0.05
0.008

Erro r F unction(E)
0
0.006
-0.05

0.004
-0.1

0.002
-0.15

-0.2 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Time(sec) Time(sec)

(a) (b)
Energies 2022, 15, x FOR PEER REVIEW 13 of 18
Figure
Figure13.
13.(a)
(a)The
Thesynchronous
synchronouserror
errorunder
underhigh-frequency
high-frequencycommands;
commands;(b)
(b)the
thecost
costfunction
functionunder
under
high-frequency commands.
high-frequency commands.

0.1 0.15
without compensation without compensation
0.08 with FNN compensation with FNN compensation
with FNN+MRAC compensation 0.1 with FNN+MRAC compensation
0.06
Synchro nous Error(m m )

Synchro nous Error(m m )

0.05
0.04

0.02 0

0 -0.05

-0.02
-0.1
-0.04
-0.15
-0.06

-0.08 -0.2
0 5 10 15 0 1 2 3 4 5 6 7
Time(sec) Time(sec)

(a) (b)
Figure14.
Figure 14.(a)
(a)The
Thesynchronous
synchronouserror
errorunder
underlow-frequency
low-frequencycommands
commandswith
withdisturbances.
disturbances.(b)
(b)the
the
synchronous error under high-frequency commands with disturbances.
synchronous error under high-frequency commands with disturbances.

Table 2. Synchronous performance under low-frequency commands.

Table 2. Synchronous performance under low-frequency commands.
Performance No
Performance No FNN FNN + MRAC
Index (mm) Compensation FNN FNN + MRAC
Index (mm) Compensation
without
E SAE
ESAE 294.204
294.204 195.394
195.394 157.399
157.399
without
disturbances
disturbances ERMS 0.0359 0.0235 0.0198
E RMS 0.0359 0.0235 0.0198
with ESAE 30.125 25.126 23.142
disturbances E SAE
ERMS 0.0387
30.125 0.0241
25.126 0.0204
23.142
with disturbances
E RMS 0.0387 0.0241 0.0204
Table 3. Synchronous performance under low-frequency commands.

Table 3. SynchronousPerformance
performance under low-frequency
No commands.
FNN FNN + MRAC
Index (mm) Compensation
Performance
without ESAE No578.168
Compensation 357.550
FNN FNN + MRAC
307.116
Index (mm)
disturbances ERMS 0.0606 0.0383 0.0330
E 578.168 357.550 307.116
without
with ESAESAE 91.225 71.114 63.453
disturbances
disturbances ERMS 0.0918 0.0709 0.0644
E RMS 0.0606 0.0383 0.0330

E SAE 91.225 71.114 63.453

with disturbances
E RMS 0.0918 0.0709 0.0644

4.2. Parallel Master–Slave Synchronous Control

Energies 2022, 15, 123 13 of 17

4.2. Parallel Master–Slave Synchronous Control

There are also two kinds of position commands mentioned above fed to the parallel
master–slave synchronous control method to test the synchronous performance. For low-
frequency commands, the synchronous error and cost function shown in Equation (14) are
depicted in Figure 15a,b. Figure 16a,b shows the case of a high-frequency command. When
a constant disturbance 0.02375 N-m is applied to Axis 2 during 2 to 3 s and 1.2 to 1.6 s
corresponding to low- and high-frequency commands, respectively, and the synchronous
errors will be magnified without a compensator. Figure 17a,b shows the synchronous
errors with respect to low- and high-frequency commands. In contrast, these figures also
show that the synchronous errors will be suppressed to be similar to the condition without
disturbances by synchronous compensators. The detailed results of the two performance
Energies 2022,
Energies 2022, 15,
15, xx FOR
FOR PEER
PEER REVIEW
REVIEW 14 of
14 of 18
indices in Equation (23) are shown in Tables 4 and 5. The sampling points between the18
dotted lines shown in Figure 17a,b, where the disturbances are applied, are calculated in
the case of “with disturbances”.
0.25
0.25 0.02
0.02
without compensation
without compensation without compensation
without compensation
with FNN
with FNN compensation
compensation 0.018
0.018 with FNN
with FNN compensation
compensation
0.2
0.2 with FNN+MRAC compensation
with FNN+MRAC compensation with FNN+MRAC compensation
with FNN+MRAC compensation
0.016
0.016
Error(mmm))

0.15
0.15 0.014
0.014
nous Error(m

unction(E)
Errorr FFunction(E)

0.012
0.012
0.1
0.1
0.01
0.01
Synchronous

0.05
0.05 0.008
0.008
Synchro

Erro

0.006
0.006
00
0.004
0.004
-0.05
-0.05
0.002
0.002

-0.1
-0.1 000
00 55 10
10 15
15 0 55 10
10 15
15
Time(sec)
Time(sec) Time(sec)
Time(sec)

(a) (b)
Figure 15. (a) The synchronous error under low-frequency commands; (b) the cost function under
Figure 15. (a) The synchronous error under low-frequency commands; (b) the cost function under
low-frequency commands.
low-frequency commands.

0.15
0.15 0.012
0.012
without compensation
without compensation without compensation
without compensation
with FNN
with FNN compensation
compensation with FNN
with FNN compensation
compensation
0.1
0.1 with FNN+MRAC compensation
with FNN+MRAC compensation with FNN+MRAC compensation
with FNN+MRAC compensation
0.01
0.01
Error(mmm))

0.05
0.05
0.008
nous Error(m

unction(E)

0.008
Errorr FFunction(E)

00
0.006
0.006
Synchronous

-0.05
-0.05
Synchro

Erro

0.004
0.004
-0.1
-0.1

0.002
0.002
-0.15
-0.15

-0.2
-0.2 000
00 11 22 33 44 55 66 77 0 11 22 33 44 55 66 77
Time(sec)
Time(sec) Time(sec)
Time(sec)

(a) (b)
Figure16.
Figure 16.(a)
(a)The
Thesynchronous
synchronouserror
errorunder
underhigh-frequency
high-frequencycommands;
commands;(b)
(b)the
thecost
costfunction
functionunder
under
high-frequency commands.
high-frequency commands.

0.15
0.15 0.2
0.2
without compensation
without compensation without compensation
without compensation
with FNN
with FNN compensation
compensation with FNN
with FNN compensation
compensation
0.1
0.1 with FNN+MRAC
with FNN+MRAC compensation
compensation 0.15
0.15 with FNN+MRAC
with FNN+MRAC compensation
compensation
Error(mmm))

Error(mmm))

0.05
0.05 0.1
0.1
Synchronous Error(m

Synchronous Error(m

00 0.05
0.05
Synchronous

Synchronous

-0.05
-0.05 00

-0.1
-0.1 -0.05
-0.05
 -0.2 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Time(sec) Time(sec)

(a) (b)
Energies 2022, 15, 123 Figure 16. (a) The synchronous error under high-frequency commands; (b) the cost function
14 under
of 17
high-frequency commands.

0.15 0.2
without compensation without compensation
with FNN compensation with FNN compensation
0.1 with FNN+MRAC compensation 0.15 with FNN+MRAC compensation
Synchronous Error(m m )

Synchronous Error(m m )
0.05 0.1

0 0.05

-0.05 0

-0.1 -0.05

-0.15 -0.1

-0.2 -0.15
0 5 10 15 0 1 2 3 4 5 6 7
Time(sec) Time(sec)

(a) (b)
Figure17.
Figure 17.(a)
(a)The
Thesynchronous
synchronouserror
errorunder
underlow-frequency
low-frequencycommands
commandswithwithdisturbances;
disturbances;(b)
(b)the
the
synchronouserror
synchronous errorunder
underhigh-frequency
high-frequency commands
commands with
with disturbances.
disturbances.

Table 4. Synchronous performance under high-frequency commands.

Table 4. Synchronous performance under high-frequency commands.

Performance No
FNN FNN + MRAC
Index (mm) Compensation
without ESAE 208.891 124.671 71.494
disturbances ERMS 0.0768 0.0461 0.0278
with ESAE 48.253 29.221 15.223
disturbances ERMS 0.117 0.0653 0.0341

Table 5. Synchronous performance under high-frequency commands.

Performance No
FNN FNN + MRAC
Index (mm) Compensation
without ESAE 262.844 189.101 129.021
disturbances ERMS 0.0914 0.0667 0.0449
with ESAE 52.573 35.891 26.432
disturbances ERMS 0.129 0.0892 0.0654

In addition, we use the parameter settings of [34] to realize the performance of the
PID compensator, as shown in Figure 18. The synchronous errors under different control
schemes are shown in Table 6.
 with disturbances
E RMS 0.129 0.0892 0.0654

In addition, we use the parameter settings of [34] to realize the performance of the
Energies 2022, 15, 123 PID compensator, as shown in Figure 18. The synchronous errors under different15control
of 17
schemes are shown in Table 6.

(a) (b)

(c)
Figure
Figure 18.18.
TheThe performance
performance ofof the
the PID
PID compensator:
compensator: (a)(a) the
the tracking
tracking error
error ofof Axis
Axis 1; 1;
(b)(b) the
the tracking
tracking
error of Axis 2; (c) the synchronous error.
error of Axis 2; (c) the synchronous error.

Table 6. Synchronous performance under different control schemes.

Table 6. Synchronous performance under different control schemes.

Performance Index (mm) No Compensation FNN FNN + MRAC

ERMS 0.04 0.024 0.02

5. Conclusions
This paper has proposed MRAC controllers and FNN online compensators for a gantry
robot. We successfully completed the theoretical and technical feasibility of the proposed
method through various experimental comparisons. From Tables 2–5, we demonstrate
the advantages of our proposed method (FNN + MRAC) for the synchronous errors and
the design can enhance robustness to uncertainty. In addition, this study also successfully
integrates the hardware and successfully verifies the proposed methods. For the future
research direction, because this paper does not analyze and deal with the influence of
friction, the analysis and compensation of friction will be the future development direction.

Author Contributions: Conceptualization, C.-S.C.; methodology, C.-S.C.; software, C.-S.C. and N.-
T.H.; validation, N.-T.H.; formal analysis, N.-T.H.; investigation, N.-T.H.; resources, C.-S.C.; data
curation, C.-S.C. and N.-T.H.; writing—original draft preparation, N.-T.H.; writing—review and
editing, N.-T.H.; visualization, C.-S.C.; supervision, C.-S.C.; project administration, C.-S.C.; funding
acquisition, C.-S.C. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by Ministry of Science and Technology of the Republic of China,
Taiwan, for financially/partially supporting, grant number MOST 102-2221-E-027-079.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the
study.
Energies 2022, 15, 123 16 of 17

Data Availability Statement: Not applicable.

Conflicts of Interest: The author declares no conflict of interest.

Appendix A
According to the development in [30], the continuous Lyapunov function is selected as

3
1
L = e2 + ∑ ( xi + β i egi )2 (A1)
i =1
α i

. 3
where the αi , β i are arbitrary positive constants and e = −( Bn /Jn )e + ∑ xi gi . For more
i =1
parameter description, please see [35]. Reference [30] has shown that the purpose of the
terms β i egi (i = 1, 2, 3) in Equation (A1) will make the adaptive process converge faster.
Therefore, the time derivative of Equation (A1) can be calculated and finally obtained
as follows
. . 3 n h. io 3
L = 2ee + 2 ∑ 1
+ β i egi ) xi + β i dtd (egi ) = −2 BJnn e2 + 2e ∑ xi gi +
αi ( x i
i =1 i =1 (A2)
3 n h. io
1 d
2 ∑ α ( xi + β i egi ) xi + β i dt (egi )
i
i =1

. . .
then, xi is designed as xi = − KJvt K i = −αi egi − β i dt
d
(egi ), so that Equation (A2) is negative
definite and the response of the plant is consistent with the reference model. Therefore,
Equation (A3) can be obtained as follows

. 3
Bn 2
L = −2 e − 2 ∑ β i (egi )2 (A3)
Jn i =1

.
which is negative definite for all e. We divide both sides of the adaptive law xi by − KJvt and
integrate, then the following adaptive laws can be obtained as
Z t
K1 = B1 e · ωdt + C1 eω (A4)
0
Z t
K2 = B2 e · zdt + C2 ez (A5)
0
Z t
K3 = B3 e · sgn(ω )dt + C3 esgn(ω ) (A6)
0
Equations (A4)–(A5) are continuous modes. After converting to discrete modes, Equations (A4)–(A5)
are Equation (6), respectively.

References
1. Luo, R.C.; Perng, Y.W. Adaptive skew force free model-based synchronous control and tool center point calibration for a hybrid
6-DoF gantry-robot machine. IEEE/ASME Trans. Mechatron. 2020, 25, 964–976. [CrossRef]
2. Tan, K.K.; Lee, T.H.; Huang, S. Precision Motion Control; Springer: Berlin/Heidelberg, Germany, 2008.
3. Li, C.; Li, C.; Chen, Z.; Yao, B. Advanced synchronization control of a dual-linear-motor-driven gantry with rotational dynamics.
IEEE Trans. Ind. Electron. 2018, 65, 7526–7535. [CrossRef]
4. Koren, Y. Cross-coupled biaxial computer control for manufacture systems. ASME J. Dyn. Syst. Meas. Control 1980, 102, 265–272.
[CrossRef]
5. Giam, T.S.; Tan, K.K.; Huang, S. Precision coordinated control of multi-axis gantry stages. ISA Trans. 2007, 46, 399–409. [CrossRef]
[PubMed]
6. Lu, L.; Chen, Z.; Yao, B.; Wang, Q. Desired compensation adaptive robust control of a linear-motor-driven precision industrial
gantry with improved cogging force compensation. IEEE/AMSE Trans. Mechatron. 2008, 13, 617–624. [CrossRef]
7. Chen, S.L.; Lin, W.M.; Chang, T.H. Tracking control for a synchronized dual parallel linear motor machine tool. Proc. Inst. Mech.
Eng. Part I: J. Syst. Control Eng. 2008, 222, 851–862. [CrossRef]
Energies 2022, 15, 123 17 of 17

8. Tsai, W.C.; Wu, C.H.; Cheng, M.Y. Tracking accuracy improvement based on adaptive nonlinear sliding mode control.
IEEE/ASME Trans. Mechatron. 2021, 26, 179–190. [CrossRef]
9. Zhao, J.; Mou, Q.; Zhu, C.; Chen, Z.; Li, J. Study on a double-sided permanent-magnet linear synchronous motor with reversed
slots. IEEE/AMSE Trans. Mechatron. 2021, 26, 3–12. [CrossRef]
10. Wu, X.; She, J.; Yu, L.; Dong, H.; Zhang, W.A. Contour tracking control of networked motion control system using improved
equivalent-input-disturbance approach. IEEE Trans. Ind. Electron. 2021, 68, 5155–5165. [CrossRef]
11. Chen, C.S.; Fan, Y.H.; Tseng, S. Position command shaping control in a retrofitted milling machine. Int. J. Mach. Tools Manuf. 2006,
46, 293–303. [CrossRef]
12. Sun, D. Position synchronization of multiple motion axes with adaptive coupling control. Automatica 2003, 39, 997–1005.
[CrossRef]
13. Wang, Y.W.; Zhang, W.A.; Yu, L. GESO-based position synchronization control of networked multiaxis motion system.
IEEE Trans. Ind. Inform. 2020, 16, 248–257. [CrossRef]
14. Chen, Z.; Li, C.; Yao, B.; Yuan, M.; Yang, C. Integrated coordinated/synchronized contouring control of a dual-linear-motor-driven
gantry. IEEE Trans. Ind. Electron. 2020, 67, 3944–3954. [CrossRef]
15. Hu, C.; Yao, B.; Wang, Q. Coordinated adaptive robust contouring controller design for an industrial biaxial precision gantry.
IEEE/ASME Trans. Mechatron. 2010, 15, 728–735.
16. Yemelyanov, V.; Chernyi, S.; Yemelyanova, N.; Varadarajan, V. Application of neural networks to forecast changes in the technical
condition of critical production facilities. Comput. Electr. Eng. 2021, 93, 107225. [CrossRef]
17. Demby’s, J.; Gao, Y.; DeSouza, G.N. A Study on solving the inverse kinematics of serial robots using artificial neural network and
fuzzy neural network. In Proceedings of the 2019 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), New Orleans,
LA, USA, 23–26 June 2019; pp. 1–6.
18. Chetyrbok, P.V. Fuzzy neural networks for classification of problems. In Proceedings of the 2021 XXIV International Conference
on Soft Computing and Measurements (SCM), St. Petersburg, Russia, 26–28 May 2021; pp. 133–135.
19. Wang, L.X. A Course in Fuzzy Systems and Control; Prentice-Hall Press: Hoboken, NJ, USA, 1997.
20. Xiao, L.; Zhang, Z.; Li, S. Solving time-varying system of nonlinear equations by finite-time recurrent neural networks with
application to motion tracking of robot manipulators. IEEE Trans. Syst. Man Cybern. Syst. 2019, 49, 2210–2220. [CrossRef]
21. Wan, P.; Sun, D.; Zhao, M.; Huang, S. Multistability for almost-periodic solutions of Takagi–Sugeno fuzzy neural networks with
nonmonotonic discontinuous activation functions and time-varying delays. IEEE Trans. Fuzzy Syst. 2021, 29, 400–414. [CrossRef]
22. Lin, C.J.; Chin, C.C. Prediction and identification using wavelet-based recurrent fuzzy neural networks. IEEE Trans. Syst. Man
Cybern. Part B Cybern. 2004, 34, 2144–2154. [CrossRef]
23. Wang, J.; Luo, W.; Liu, J.; Wu, L. Adaptive type-2 FNN-based dynamic sliding mode control of DC–DC boost converters.
IEEE Trans. Syst. Man Cybern. Syst. 2021, 51, 2246–2257. [CrossRef]
24. Han, M.; Zhong, K.; Qiu, T.; Han, B. Interval Type-2 fuzzy neural networks for chaotic time series prediction: A concise overview.
IEEE Trans. Cybern. 2019, 49, 2720–2731. [CrossRef]
25. Na, J.; Wang, S.; Liu, Y.J.; Huang, Y.; Ren, X. Finite-time convergence adaptive neural network control for nonlinear servo systems.
IEEE Trans. Cybern. 2020, 50, 2568–2579. [CrossRef] [PubMed]
26. Kayacan, E.; Kayacan, E.; Ramon, H.; Saeys, W. Adaptive neuro- fuzzy control of a spherical rolling robot using sliding-mode-
control- theory-based online learning algorithm. IEEE Trans. Cybern. 2013, 43, 170–179. [CrossRef]
27. Dou, H.; Lu, H.; Wang, S.; Liu, Q.; Wang, D.; Meng, L. Research on synchronous control of gantry-type dual-driving feed system
based on fuzzy single neuron PID cross-coupled controller. In Proceedings of the IEEE 5th Advanced Information Technology,
Electronic and Automation Control Conference (IAEAC), Chongqing, China, 12–14 March 2021; pp. 1107–1112.
28. Jung, H.; Oh, S. Disturbance observer based decoupling control to suppress rotational motion of cross-coupled gantry
stage. In Proceedings of the IEEE 28th International Symposium on Industrial Electronics (ISIE), Vancouver, BC, Canada,
12–14 June 2019; pp. 503–508.
29. Chen, C.S.; Chen, S.K.; Chen, L.Y. Disturbance observer-based modeling and parameter identification for synchronous dual-drive
ball screw gantry stage. IEEE ASME Trans. Mechatron. 2019, 24, 2839–2849. [CrossRef]
30. Gilbart, J.W.; Winston, G.C. Adaptive compensation for an optical tracking telescope. Automatica 1974, 10, 125–131. [CrossRef]
31. Chou, P.H.; Chen, C.S.; Lin, F.J. DSP-based synchronous control of dual linear motors via Sugeno type fuzzy neural network
compensator. J. Franklin Inst. 2012, 349, 792–812. [CrossRef]
32. Tan, K.H.; Lin, F.J.; Chen, J.H. DC-link voltage regulation using RPFNN-AMF for three-phase active power filter. IEEE Access
2018, 6, 37454–37463. [CrossRef]
33. Yoo, S.J.; Choi, Y.H.; Park, J.B. Generalized predictive control based on self-recurrent wavelet neural network for stable path
tracking of mobile robots: Adaptive learning rates approach. IEEE Trans. Circuits Syst. 2006, 53, 1381–1395.
34. Wu, K.Z. Iterative Learning Command Shaping and Cross-Coupled PIDNN Synchronous Controller for H-Type Gantry Stage.
Master’s Thesis, National Taipei University of Technology, Taipei, Taiwan, 2014.
35. Yang, T.H. Fuzzy Neural Network Synchronous Control of Gantry Stage. Master’s Thesis, National Taipei University of
Technology, Taipei, Taiwan, 2011.