applied

sciences
Article
A Data-Driven Approach of Takagi-Sugeno Fuzzy Control of
Unknown Nonlinear Systems
Bin Zhang and Yung C. Shin *

School of Mechanical Engineering, Purdue University, West Lafayette, IN 47906, USA; [email protected]
* Correspondence: [email protected]; Tel.: +1-765-494-9775

Abstract: A novel approach to build a Takagi-Sugeno (T-S) fuzzy model of an unknown nonlinear
system from experimental data is presented in the paper. The neuro-fuzzy models or, more specifically,
fuzzy basis function networks (FBFNs) are trained from input–output data to approximate the
nonlinear systems for which analytical mathematical models are not available. Then, the T-S fuzzy
models are derived from the direct linearization of the neuro-fuzzy models. The operating points for
linearization are chosen using the evolutionary strategy to minimize the global approximation error
so that the T-S fuzzy models can closely approximate the original unknown nonlinear system with
a reduced number of linearizations. Based on T-S fuzzy models, optimal controllers are designed
and implemented for a nonlinear two-link flexible joint robot, which demonstrates the possibility of
implementing the well-established model-based optimal control method onto unknown nonlinear
dynamic systems.

Keywords: Takagi-Sugeno model; data-driven system identification; neuro-fuzzy model;

optimal control; flexible robot



1. Introduction


The Takagi-Sugeno (T-S) fuzzy model is a powerful and practical engineering tool for
Citation: Zhang, B.; Shin, Y.C. modeling and control of complex nonlinear systems. It proves to be a universal function
A Data-Driven Approach of approximator that can approximate any smooth nonlinear functions to any degree of
Takagi-Sugeno Fuzzy Control of accuracy [1,2] and is less sensitive to the curse of dimensionality than other fuzzy models [3].
Unknown Nonlinear Systems. Appl. The concept of T-S fuzzy model is similar to the piecewise linear approximation approaches
Sci. 2021, 11, 62. https:// in nonlinear control, which linearizes a system at a set of selected operating points and
dx.doi.org/10.3390/app11010062
designs a local linear feedback controller for each linear model. However, since the overall
control action is switching among the local linear controllers according to system states and
Received: 1 November 2020
thus there is only one local controller active at a certain time in such approaches, it can only
Accepted: 20 December 2020
ensure the stability and performance of the control system at the neighborhood of selected
Published: 23 December 2020
operating points [4], In contrast, the T-S fuzzy model approximates the entire nonlinear
Publisher’s Note: MDPI stays neu-
system by fuzzy inference among local linear models so that the overall control action
tral with regard to jurisdictional claims
can be generated by aggregation of local linear control laws [5]. Therefore, it empowers
in published maps and institutional
a paradigm of designing controllers for local linear models while analyzing stability for
affiliations. the global nonlinear system [6]. The T-S fuzzy-model-based control that blends feedback
controllers from local models is referred to as Parallel Distributed Compensation (PDC)
scheme, in which the stability of the overall control system is assessed through Lyapunov
stability analysis, especially by the Linear Matrix Inequality (LMI) technique [7–12].
Copyright: © 2020 by the authors. Li- To take advantage of T-S fuzzy-model-based control, the identification of T-S fuzzy
censee MDPI, Basel, Switzerland. This model has attracted great research interest. There are two kinds of methods for establishing
article is an open access article distributed
T-S fuzzy models. One is linearizing the original system at a series of operating points
under the terms and conditions of the
when an analytical model of the system is available. The other is the consecutive structure
Creative Commons Attribution (CC BY)
and parameter identification from the data generated by the unknown system [6], which is
license (https://creativecommons.org/
more of interest to us. The structure identification refers to the selection of locations of fuzzy
licenses/by/4.0/).

Appl. Sci. 2021, 11, 62. https://dx.doi.org/10.3390/app11010062 https://www.mdpi.com/journal/applsci

Appl. Sci. 2021, 11, 62 2 of 15

rules based on clustering [13,14]. With the determined antecedent structural parameters,
the T-S fuzzy model transforms into a set of linear models, of which the parameters are
obtained by the recursive least square method [15,16], genetic algorithm [17] or particle
swarm optimization [18]. The objective of data-driven approaches is to minimize the
global prediction error of a T-S fuzzy model. However, they may result in constituent
linear models significantly different from the local linearization of nonlinear systems,
though they may offer good global performance [19]. Hence, the T-S models obtained by
these methods might not be satisfactory for the controller design in the Parallel Distributed
Compensation (PDC) scheme, since the local compensators need to be designed based on
local linear models.
The authors in [2] studied various identification algorithms and concluded that the
constrained and regularized identification method can improve the interpretability of
constituent local models as local linearization, and the locally weighted least square tech-
nique may facilitate the compromise between the local and global accuracy of T-S models.
However, the effectiveness and practicality of this method were demonstrated only by
very simple examples. When considering more complicated higher-order and multivariate
problems, the issues related to interpretability and identifiability will be more pronounced
and difficult to address.
An alternative method to circumvent these difficulties is presented in this paper.
A neuro-fuzzy model referred to as fuzzy basis function networks (FBFN) is adopted to
approximate the unknown nonlinear systems [20]. The Stone–Weierstrass theorem proves
that this kind of neuro-fuzzy model could approximate any real continuous function on
a compact domain arbitrarily well [21]. Then, the T-S fuzzy model is derived from the
linearization of the neuro-fuzzy model at a series of operating points. Therefore, each local
model is close to the local linearization of the original system and, thus, suitable for local
compensator design. In addition, the positions of operating points for linearization are
optimized by the evolutionary strategy to minimize global approximation error so that
the entire T-S fuzzy model is a good global approximation of the original system. A fuzzy
control scheme can be applied to the consequent T-S fuzzy model in which a local optimal
compensator is designed for each of the local affine models and the overall control action
is derived from the fuzzy inferencing of local control actions. The contribution of this
paper is that it presents a practical way to build T-S fuzzy models with both good local and
global approximations by deriving the direct linearization of FBFN models and introducing
the evolutionary strategy for fuzzy rule location selection. Compared with the existing
methods, the method proposed in this paper is less involved compared with nonlinear
controllers while the performance is not compromised and, thus, it is particularly suitable
for controller design of complex unmodeled nonlinear systems in practice.
This paper is organized as follows: Section 2 presents the structure of the T-S fuzzy
model; Section 3 elaborates the proposed T-S fuzzy model identification approach;
Section 4 explains the design of fuzzy optimal controller; Section 5 demonstrates an exam-
ple of T-S model identification and optimal control of a robotic system; Section 6 concludes
the paper.

2. Takagi-Sugeno Fuzzy Model

T-S fuzzy models represent complicated multi-input-multi-output (MIMO) systems
with fuzzy inference rules and local linear models as follows:
.
Rk : IF z1 is F1 k , · · · , zv is Fv k , THEN x = Ak x(t) + Bk u(t) + dk ,

where Rk denotes the kth fuzzy rule, k ∈ {1, 2, · · · , p}, p is the number of fuzzy rules, Fj k
(j = 1, 2, · · · , v) are the input fuzzy sets, x(t) ∈ Rn is the state variable vector, u(t) ∈ Rm is
the input variable vector, z(t): = [z1 , z2 , · · · , zv ] are a subset of measurable or observable
variables in the state and input vectors that are used for fuzzification, and (Ak , Bk , dk )
are the matrices of the kth local model [6]. If the constant bias vector is not null, dk 6= 0,
Appl. Sci. 2021, 11, 62 3 of 15

for some fuzzy inference rules, the corresponding local models are affine models instead of
linear models.
By using the product fuzzy inference, singleton output membership functions, and cen-
troid defuzzifier, a T-S fuzzy model in continuous-time state-space form can be organized as:
.
x = A( µ )x( t ) + B( µ )u( t ) + d( µ ), (1)
p p p
where A(µ) = ∑k=1 µk Ak , B(µ) = ∑k=1 µk Bk , d(µ) = ∑k=1 µk dk and µk denotes the
p
normalized fuzzy membership function µk = ξ k (z)/∑k=1 ξ k (z) with ξ k (z) = ∏iv=1 Φik (zi ).
Φik (zi ) represents the i-th membership function in fuzzy set Fi k of the k-th fuzzy rule.
Due to the fact that the membership functions are nonlinear (e.g., triangular or Gaussian),
the model in Equation (1), as an aggregation of local linear or affine models, is also a
nonlinear model in nature.

3. Identification of Takagi-Sugeno Fuzzy Model

The T-S fuzzy model is expected to attain a good approximation of not only the local
dynamics of the underlying system to facilitate local compensator design, but also the
global dynamics to guarantee the overall control performance. However, considering them
together is difficult because building constituent local linear models from input-output
data is not always straightforward and the tradeoff between the local and global accuracy
should be carefully addressed [2].
In this paper, a novel approach that utilizes the neuro-fuzzy model to obtain a
T-S fuzzy model is proposed. The neuro-fuzzy models are based on the fusion of fuzzy
inference systems and neural networks. Among diverse neuro-fuzzy models, the one
proposed by Wang and Mendel [20], which is also known as fuzzy basis function network
(FBFN), has gained much attention as it has a similar structure to radial basis function
networks (RBFNs) and thus can adopt the training methods that are already established
for the RBFN. It has been proved that the FBFNs can uniformly approximate any real
continuous nonlinear functions on a compact set to arbitrary accuracy [21].
First, from the input-output data, a set of multi-input-single-output FBFNs with
product fuzzy inference, Gaussian membership functions (MFs), and centroid defuzzifier
are trained by the least square algorithm [22] to approximate the nonlinear system whose
analytical mathematical model is not available. Each FBFN can be represented as a fuzzy
system in layered network form (see Figure 1), and is used to approximate the dynamics of
one state variable. For example, the FBFN for the p-th state variable is constructed using
fuzzy rules as:
.
R p l : IF x1 is A1p l , · · · , xn is Anp l and u1 is B1p l , · · · , um is Bmp l , THENx p = wlp ,

where l ∈ {1, 2, · · · , Mp }, and Mp is the number of fuzzy rules for the p-th FBFN.
Through fuzzy inference and defuzzification, the FBFN based on the above fuzzy rules can
be written as:
2 2
xi − x l u j −ul
ip jp
Mp − 12 ( ) − 21 ( )
n σl m σl
∑ wp l ∏ e ip
∏e ujp
. l =1 i =1 j =1
x p = f p (x, u) = 2 2 p = 1, 2, · · · , n , (2)
xi − x l u j −ul
ip jp
Mp n − 12 ( ) − 21 ( )
σl m σl
ip ujp
∑ ∏e ∏e
l =1 i =1 j =1

where x: = [x1 , x2 , · · · , xn ] is the state variable vector, u: = [u1 , u2 , · · · , um ] is the input

vector, and wp l is the weighting factor. The xip l and σip l are the centers and widths of
 where H.O.T denotes the higher order terms in the local model, which w
T
in the following analysis, f  x k , u k    f1  x k , u k  , , f n  x k , u k   , and

Appl. Sci. 2021, 11, 62  f1  x, u  f1  x, u   4 of 15

  
 x1 xn 
2 Ak      
xi − x l  
1 ip
−2( l )
Gaussian MFs e
σ
ip l  f
in fuzzy set Aip , and the  x , ul  l f  x , u 
uip and σu ip aren the centers and widths

n

xi − x l
− 12 ( l )
ip
2
 x1 xn  x  x ,u  u
of Gaussian MFs e
σ
ip in fuzzy set B l .
k k
,
ip

Figure 1. Structure
Figure of the FBFN.
1. Structure of the FBFN.
Next, the T-S fuzzy model is derived from the neuro-fuzzy model by linearizing
EquationBy
(2).taking the partial
Linearization about anderivative of Equation
operating point (2)inwith respect to xq,
(xk , uk ) results one
.
x = Ak (x − xk ) + Bk (u − uM
k )p + f(xk , uk )p + H.O.T, p
M M Mp (3)
l l l l l l l l
, u  in
w a a q a w a q
where H.O.T denotes the higher f  xterms
order the local model, which will be neglected in
p l 1
p p
l 1
p p
l 1
p
l 1
p p p

the following analysis, f(xk , uk ) = [ 


f 1x(xk , uk ), . . . , f n (xk , uk )] T , and 2
Mp
q  l

 ∂ f 1 (x,u)
··· ∂ f 1 (x,u)    a p 
∂x1 ∂xn
 l 1 
.. .. .. (4) ,
 
Ak = 
 . . .

 ,
∂ f n (x,u) ∂ f n (x,u)
where ∂x1 ··· ∂xn x=xk ,u=uk

By taking the partial derivative of Equation (2) 1with l 2

 xi  k respect to xq ,1 one
 u j  kcan
2
 xip   uljpget

n    m   
l 2   ipl  2 l
 ujp 
Mp
p
Mp a  e M 
p M p
 
e 
∑ w p l alp ∑ ailpq1lp − ∑ alp ∑ w p ljalp1qlp
∂ f p (x, u) l =1 l =1 l =1 l =1
= !2 l , (5)
∂xq
l
M
q p l x k   x
qp
p q  ∑ ap 2
l =1 l
qp   .
where
2 2
xi (k)− x l u j (k)−ul
ip jp
Similarly, n − 12 (
σl
) m − 12 (
σl
)
alp = ∏ e ip
∏e ujp
i =1 j =1 (6)
l
xq (k )− xqp
qlp = l 2
(σqp )
Appl. Sci. 2021, 11, 62 5 of 15

Similarly,
 ∂ f 1 (x,u) ∂ f 1 (x,u) 
∂u1 ··· ∂um

Bk =  .. .. .. 
, (7)
 . . .


∂ f n (x,u) ∂ f n (x,u)
∂u1 ··· ∂um x=xk ,u=uk

where ∂ f p (x, u)/∂uq has the same expression as Equation (5) except that:

uq (k ) − ulqp
qlp =  2 (8)
l
σuqp

For simplicity, and also by neglecting the higher-order terms, Equation (3) can be
rewritten as
.
x = Ak x + Bk u + dk , (9)
where
dk = f (xk , uk ) − Ak xk − Bk uk . (10)
The affine term dk is non-null, i.e., dk 6= 0 in general even if the operating point is an
equilibrium point other than the origin [23]. Including the affine terms will offer a more
accurate local approximation of the system’s dynamics around the operating points.
Then, the operating points used for linearization need to be chosen carefully to achieve
a good global approximation of the original system. Meanwhile, it is desirable to minimize
the number of linearization so that the number of fuzzy rules in the T-S fuzzy model can
be reduced and the fuzzy controller synthesized in the following section will be more
computationally efficient. To reduce the number of linearization points, it is necessary to
know which variables among z = [z1 , z2 , · · · , zv ] in fuzzy inference are the major sources of
nonlinearity. These variables should be assigned more positions while the variables that are
minor sources can be assigned fewer positions. A measure to roughly recognize the source
of nonlinearity by inspecting the variation of linearized A-matrix is presented in Section 5.
If the i-th variable zi is to be assigned pi positions {zi,1 , . . . , zi,pi }, then by combination of
the ∑iv=1 pi positions, there will be p = ∏iv=1 pi operating points in total, i.e., (xk , uk ) (k = 1,
2, · · · , p).
After the number of design positions pi for each zi is selected, their optimal positions
can be searched by the Evolutionary Strategy (ES) [24] inside the operating range to produce
the p optimal operating points to build the T-S fuzzy model:


Find (x1 , u1 ), (x2 , u2 ), · · · , x p , u p produced by zi,j i=1,...v,1,...,p
v i
u N .
u ∑ [ xi,TS [k]− x. i,M [k]]
n u j =1
to minimize NDEI = ∑ u t N h. .
i .
i =1 ∑ xi,M [k ]− xi,M
j =1

The non-dimensional error index (NDEI) between the responses predicted by the
. .
T-S fuzzy model xi,TS [k ] and the measurement data xi,M [k] is a reliable criterion of the
quality of global approximation and can be used as the objective function to be minimized.
The input-output data pairs used for validation of the neuro-fuzzy model can be reused
here so that no extra experiments need to be conducted. N is the number of pairs in the
. .
validation data set, n is the number of state variables and xi,M is the mean value of xi,M [k ].
The ES optimization is controlled by three parameters: the maximum number of
generations t, the parent population µ, and the offspring population λ. The optimization
starts with initializing a parent pool with µ individuals. Then, in each generation, a pair
of parents is randomly selected to produce an offspring via recombination and mutation,
as explained in [24], until λ offspring have been generated. Each offspring is used to
build a T-S model at the operating point associated with it and the NDEI of this model is
 The ES optimization is controlled by three parameters: the maximum numbe
erations t, the parent population µ, and the offspring population λ. The optimizati
with initializing a parent pool with µ individuals. Then, in each generation, a pai
ents is randomly selected to produce an offspring via recombination and mutatio
Appl. Sci. 2021, 11, 62 6 of 15
plained in [24], until λ offspring have been generated. Each offspring is used to b
S model at the operating point associated with it and the NDEI of this model is r
The best µ offspring will form the parent pool for the next generation. If the optim
recorded. The best µ offspring will form the parent pool for the next generation. If the
of operating points after t generations doesn’t yield a T-S fuzzy model with sat
optimized set of operating points after t generations doesn’t yield a T-S fuzzy model with
global approximation,
satisfactory the number
global approximation, of operating
the number points
of operating p will
points bebe
p will increased,
increased,and the
zation
and will be repeated,
the optimization as illustrated
will be repeated, in Figure
as illustrated 2. 2.
in Figure

Figure
Figure 2. 2. Evolutionary
Evolutionary Strategy
Strategy for selecting
for selecting the points
the operating operating points of
of linearization. linearization.
4. Controller Synthesis for the T-S Fuzzy Model
4. Controller Synthesis for the T-S Fuzzy Model
In order to design a global stabilizing controller for the nonlinear system using the
In order
identified to design
T-S fuzzy model, thea global
parallelstabilizing controller for(PDC)
distributed compensation the nonlinear
framework system
is u
adopted [25]. Local feedback rule is designed as a compensator for
identified T-S fuzzy model, the parallel distributed compensation (PDC) frame each local model and
aadopted
global fuzzy
[25].controller is constructed
Local feedback rulebyisthe aggregation
designed as aofcompensator
local compensators usinglocal mo
for each
the same fuzzy inference system in the T-S fuzzy model [6]: R : IF z1 is F1 , · · · , zv is Fv k ,
k k
a global
u = ufuzzy controller is constructed by the aggregation of local compensato
THEN k , where k ∈ {1, 2, · · · , p}. The fuzzy controller is aggregated as:
the same fuzzy inference system in the T-S fuzzy model [6]: Rk: IF z1 is F1k, ⋯,
p
THEN u = uk, where k ∈ {1, 2, u⋯,=p}. The fuzzy controller is aggregated as:
µ u , ∑ k k (11)
k =1
p
where µk is the normalized membership function same as in Equation (1).
u  k u k
Assume that the fuzzy controller is designed to minimize 
the performance index:
k 1 ,
Z t h
f
i
J= (r − x)T Q(r − x) + uT Ru dt, (12)
where µk is the t0
normalized membership function same as in Equation (1).
where Assume that the
r is the command fuzzy
input controller
vector is designed
to be tracked, x is the statetovector,
minimize
u is thethe performance
input vector, i
t0 is the initial time, tf is the final time and Q and R are symmetric positive definite matrices
to be determined by the designer, then fort f each
J   r local
T controller, theToptimal control action

uk can be derived as [26]:  t0 

x Q  r  x   u Ru dt
,
  −1
uk = Kk (r − x) + BkT Bk BkT dk , (13)

where Kk is given by
Kk = R−1 BkT Pk (14)
Appl. Sci. 2021, 11, 62 7 of 15

Pk is found by solving the Continuous Algebra Riccati Equation:

AkT Pk + Pk Ak − Pk Bk R−1 BkT Pk = −Q. (15)

The stability condition of an affine T-S fuzzy control system based on a quadratic
Lyapunov function is given in [27]. The equilibrium point (x = xk , u = uk ) of the control
system is asymptotically stable in large if there exists a common positive definite matrix
P = PT > 0 and scalars τ ijq ≥ 0 such that:

Gi,j T P + PGi,j < 0 for fuzzy rules i, j ∈ {k|µk (0) 6= 0 }

 n n 
Gi,j T P + PGi,j − ∑ τijq Tijq Pηi,j − ∑ τijq uijq
 q =1 q =1 
< 0 for other fuzzy rules (16)
n n

 T
ηi,j P − ∑ τijq uijq T − ∑ τijq vijq

q =1 q =1

where the Gi,j , ηi,j , Tijq , uijq and vijq are defined as:

{ Ai − Bi K j } + { A j − B j Ki }
Gi,j = 2  −1
{ di − Bi σ j } + { d j − B j σi }

ηi,j = σ j = BTj B j BTj d j (17)
2
xT Tijq x + 2uijq T x + vijq ≤ 0

Equation (16) belongs to bilinear matrix inequalities (BMIs) and can be solved in an
iterative LMI manner. The details of the iterative linear matrix inequality (ILMI) algorithm
can be found in [27]. The stability condition in Equation (16) can be integrated into the ES
optimization procedure shown in Figure 2 as a constraint so that the fuzzy control system
designed based on the optimized operating point set is guaranteed to be stable.
The fuzzy controller is implemented in a full state-feedback manner. If some state
variables are not measurable during operation, then an observer needs to be designed for
each local model and the fuzzy observer is constructed by aggregation of local observers
with a fuzzy inference system. In [28], it has been proved that the separation principle for
linear systems also holds for T-S fuzzy systems, and thus the fuzzy controller and fuzzy
observer can be designed independently.

5. Application Example
In this section, the data-driven T-S fuzzy model identification and control are implemented
on a flexible two-link joint robot to demonstrate the effectiveness of the proposed approach.
Flexible robot manipulators possess various advantages over the rigid ones: they require
less material, allow higher manipulation speed while consume less power, and are safer
to operate due to reduced inertia. However, controlling flexible robot manipulators for
precise positioning could a challenging task because of the high precision required for
positioning, oscillation due to flexibility, highly nonlinear and distributed dynamics of the
system, as well as the difficulty in establishing an accurate model [29]. The picture and the
schematic of the two-link flexible-joint robot manipulator to be dealt with in this paper are
shown in Figure 3.
The robot is described by 8 state variables: θ 1 , angle of the 1st link; θ 2 , angle of the 2nd
link; θ 3 , angle of the 1st motor; θ 4 , angle of the 2nd motor; and the four angular velocities.
There are two input variables: T1 , the torque of the 1st motor and T2 , the torque of the 2nd
motor. The state vector and input vector of this system are defined as:

 T h .T
iT  T
θ= θ1 θ2 θ3 θ4 x= θT θ u= T1 T2 .
 quire less material, allow higher manipulation speed while consume less power, and are
safer to operate due to reduced inertia. However, controlling flexible robot manipulators
for precise positioning could a challenging task because of the high precision required for
positioning, oscillation due to flexibility, highly nonlinear and distributed dynamics of the
Appl. Sci. 2021, 11, 62 system, as well as the difficulty in establishing an accurate model [29]. The picture 8 ofand
15
the schematic of the two-link flexible-joint robot manipulator to be dealt with in this paper
are shown in Figure 3.

Figure3.3.The
Figure Thetwo-link
two-linkflexible-joint
flexible-jointrobot
robotmanipulator:
manipulator:physical
physicalphotograph
photograph(left),
(left),top-view
top-viewschematic
schematicdiagram
diagram(right).
(right).

The
Thenonlinear equation of
robot is described bymotion
8 state of this robot
variables: θ1,can be expressed
angle of the 1st as [30]:
link; θ2, angle of the
2nd link; θ3, angle of the 1st ..motor; θ4,. angle of
  . the 2nd
 . motor; and the four angular veloc-

ities. There are two input )θ + V θ,Tθ1, the
M(θvariables: + Ctorque
θ + D of θ the+ Kθ = T, and T2, the torque
1st motor (18)
of
the 2nd motor. The state vector and  input
.  vector of this system are defined as:
where M(θ) is the inertia matrix, V θ, θ is the vector of Coriolis and centrifugal functions,
θ  1  2 3  4  x  θT θ T   .u T1 T2  .
T T T

C is the viscous damping coefficient matrix, D θ is the Coulombic friction vector, K is the
stiffness
Thecoefficient
nonlinearmatrix,
equation and is the input
ofTmotion of thistorque
robot vector.
can The inertiaas
" be expressed matrix
[30]: is given by:
#
   J6  J5
M1 (θ) M12 0 J4 + r J3 + 0
M(θ) = M12 = M21 T = M2 = r2
M21 M2 M  θ 0
θ  V 0θ, θ  Cθ  D θ  Kθ
   0  T ,J4 + rJ62 (18)
(19)
2 2



p1 + m2 a2 +
J2 + 2(l1 m2 a2 ) cos(θ2 ) m2 a2 + J2 + (l1 m2 a
2 ) cos(θ2 )
M1 (θ) = 2
2 a2 + J2M+θ( l1is
mwhere m2the
a2 )inertia
cos(θ2 )matrix, V θ, θ m a22 +vector
J2
    is2the of Coriolis and centrifugal func-
where miisisthe
theviscous
lumpeddamping
mass of components, Ji is the θ is theofCoulombic
D moment inertia of components, li
tions, C coefficient matrix,   friction vector,
is the length of links, and a1 and a2 denote the offset from the center of gravity of the first
K issecond
and the stiffness
link tocoefficient matrix,
the first and andjoint,
second T is the input torque
respectively. vector. The
In addition, b1 inertia matrix is
is the distance
given by:
between the second motor and the first joint, and r is the gear ratio of the chain drives.
The vector of the Coriolis and centrifugal functions is
  . . .  
−(l1 m2 a2 ) 2θ 1 θ 2 + θ 2 2 sin(θ2 )
 .  . 
V θ, θ = 
 (l1 m2 a2 )θ 1 2 sin(θ2 ) ,

(20)
 0 
0

and the viscous damping matrix can be written as

c1 + c5 − cr5
 
 c2 + c6 − cr6 
C= , (21)
 − cr5 c3 + c5
r2

− cr6 c4 + c6
r2
Appl. Sci. 2021, 11, 62 9 of 15

where ci is the viscous friction coefficients. The vector of Coulombic friction is given by
 .  
d1 sign θ 1
 .  
.   d2 sign θ 2 

D θ =   .  , (22)
 d3 sign θ 3 

 .  
d4 sign θ 4

where di is the friction torque at each joint. The matrix of stiffness coefficients is given by

− kr5
 
k5
 k6 − kr6 
K =  k5 , (23)
 
k5
 −r r2

− kr6 k6
r2

where ki denotes the coefficients of the torsional springs at the flexible joints. The torque
vector is  T
T = 0 0 T1 T2 . (24)
Table 1 lists the estimated values of the robot’s physical parameters. However,
deriving the mathematical model and obtaining an accurate estimation of each parameter is
quite difficult and time-consuming. The purpose of this paper is to develop a methodology
of T-S model construction and controller design when the analytical mathematical model
detailed above is not available. Hence, the Equations (18)–(24) and the parameters in Table 1
are only used to validate the fuzzy controller via simulation.

Table 1. Estimated Values of the Robot’s Physical Parameters [31].

Parameter Value Parameter Value

p1 0.1402 kgm2 /rad c4 1.4975 × 10−3 Nms/rad
2
m2 a2 + J2 0.01962 kgm2 /rad c5 0.005 Nms/rad
l1 m2 a2 0.02338 kgm2 /rad c6 8.128 × 10−3 Nms/rad
J3 4.1751 × 10−5 kgm2 /rad k5 2.848 Nm/rad
J4 7.5429 × 10−4 kgm2 /rad k6 2.848 Nm/rad
J5 0.025 kgm2 /rad d1 0.01987 Nm
J6 0.025 kgm2 /rad d2 0.0323 Nm
c1 0.04 Nms/rad d3 0.0053 Nm
c2 0.02143 Nms/rad d4 0.0271Nm
c3 1.8937 × 10−4 Nms/rad r 5

In experiments, four encoders are mounted to measure the four arm and motor angles,
and the angular velocities and accelerations are obtained from the angular position signals
by the central finite difference method:
. ..
θi [k+1]−θi [k ] θi [k +1]−2θi [k ]+θi [k −1]
θi = ∆T θi = , (25)
∆T 2

where θ i [k], k = 1, 2, · · · are the values of discrete angular position measurements and
∆T = 0.001 s is the sampling time. To reduce noises in the velocity and acceleration
signals that mainly originate from the quantization of the position signal, a zero-phase low-
pass Butterworth filter with 150 Hz cutoff frequency was applied to the position signals.
To avoid the transient effect from filtering in both directions when initializing the filter
states, experiments were started with a 0.5 s rest period. This period and the last 0.5 s of
the experiment were removed from the data set.
To obtain the neuro-fuzzy model of the robot, the motors were excited by sine sweep
torques. A combination of a 5-s 1 Hz sine wave and a subsequent 5-s sine sweep signal
with an initial frequency of 1 Hz and a final frequency of 5 Hz was used to excite the
FOR PEER REVIEW 11 of 16

Appl. Sci. 2021, 11, 62 10 of 15

shoulder motor and elbow motor, as shown in Figure 4. Then, 2000 training data are
evenly drawn from the collected input-output pairs and used to train the neuro-fuzzy
model. 500 testingshoulder
data thatmotor
areand elbow motor,
different fromasthe
shown in Figure
training 4. Then,
data 2000 training
are used to test data are evenly
the accu-
drawn from the collected input-output pairs and used to train the neuro-fuzzy model.
racy of the neuro-fuzzy model
500 testing dataand thedifferent
that are subsequent T-Straining
from the fuzzydata
models. The
are used T-Sthe
to test fuzzy mod-
accuracy of the
els in this paper are all constructed
neuro-fuzzy model and with triangularT-S
the subsequent membership
fuzzy models.functions duemodels
The T-S fuzzy to their
in this
simplicity. paper are all constructed with triangular membership functions due to their simplicity.

shoulder motor torque elbow motor torque

0.4 0.4

0.2 0.2

T2 (Nm)
T1 (Nm)

0 0

-0.2 -0.2

-0.4 -0.4
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time (s) Time (s)

Figure
Figure 4. Sine
Sinesweep
sweep signals
signals to motors.
to motors.

To determine the proper number of operating points for linearization, the grade of the
To determinenonlinearity
the properofnumber
each state of operating
variable points
is evaluated as for linearization,
follows: the isgrade
first, the system of at
linearized
the nonlinearity ofthe each state variable
equilibrium point x is
= 0,evaluated as follows:
where the result is Ax =0 . first,
Then the system is
by randomly linearized
changing one state
at the equilibriumvariable
point xi =in 0,
thewhere therange
operating result
andisfixA . Then
allx=0the by randomly
other state changing
variables, a series one Axi
of matrices
are obtained. They are compared to A in terms of the 2-norm
state variable xi in the operating range and fix all the other state variables, a series of ma-
x =0 of difference. The results
are summarized in Table 2. The idea is that if the state variable xi is linear or affine in the
trices Axi are obtained. They are compared to Ax=0 in terms of the 2-norm of difference. The
system, then identical A matrices should be obtained at different positions of xi . Otherwise,
.
results are summarized in Table 2. The idea is that if the state variable xi is linear or affine
xi should be regarded as nonlinear. From Table 2, it can be seen that the variables θ2 , θ 1 and
.
in the system, then θ 2 identical
are sourcesA of matrices
nonlinearityshould
aroundbe theobtained
origin, among at different
which θ 2 ispositions of xi.
the most significant
Otherwise, xi shouldone.beTheregarded as nonlinear.
other variables From
can be treated Table 2, it can be seen that the vari-
as linear.
ables  2 , 1 and 2 are sources of nonlinearity around the origin, among which θ2 is
Table 2. Inspection of the source of nonlinearity.
the most significant one. The other variables can be treated as linear.
xi θ1 θ2 θ3 θ4 θ1dot θ2dot θ3dot θ4dot
mean
Table 2. Inspection of the kAxi −of
source Ax=0 k 0
nonlinearity. 85.89 0 0 3.67 9.13 0 0

xi Twoθsets 1 θ2
of operating θ3 are θ
points 4
selectedθfor θ2dot theθfirst
1dot comparison: 3dot set θ 4dot
consists of
27
mean A xi  A x=0 . points based
0 on the
85.89 combination
0. of uniformly
0 chosen
3.67 positions
9.13 θ 2 =0[ − 0.7, 0, 0.7]
0 (rad),
θ 1 = [−1.0, 0, 1.0] (rad/s) and θ 2 = [−1.0, 0, 1.0] (rad/s); the second set consists of only
12 points, the positions are chosen by the ES optimization: θ 2 = [−0.9575, −0.0181, 0.9479]
.
Two sets of operating
(rad), θ 1 = [− points
0.9034, are selected
0.5891] for θ.comparison:
(rad/s) and the first set consists of 27
2 = [−1.5, 1.5] (rad/s). θ 2 is assigned three positions
points based on the combination
while the other two ofare
uniformly
assigned onlychosen positions
two positions as θ = [−0.7,
θ 22 is 0, significant
the most 0.7] (rad),source 1 of
nonlinearity according to Table 2.
= [−1.0, 0, 1.0] (rad/s) and 2 = 5,[−1.0,
In Figure it can0,be1.0]
seen(rad/s);
that thethe second
outputs set consists
predicted by the T-Soffuzzy
onlymodels
12 points,are very
the positions are chosen by the ES optimization: θ2 = [−0.9575, −0.0181, 0.9479] (rad),  1two
close to the testing experimental data. The NDEIs of the neuro-fuzzy model and the  = T-S
fuzzy models, which are used as the criteria of global approximation, are listed in Table 3.
[−0.9034, 0.5891] (rad/s)
It can beand seen2that= the
[−1.5,
T-S 1.5]
fuzzy(rad/s).
modelsθderived
2 is assigned
from thethree positions
neuro-fuzzy modelwhile the
can provide
other two are assignedan accurate
onlyapproximation
two positions of the
as original
θ2 is the nonlinear system. The source
most significant 12 pointsof T-Snonline-
fuzzy model
arity according to with
Table optimized
2. operating points gives an even better approximation than the 27 points one
with uniformly distributed operating points.
In Figure 5, it can be seen that the outputs predicted by the T-S fuzzy models are very
close to the testing experimental data. The NDEIs of the neuro-fuzzy model and the two
T-S fuzzy models, which are used as the criteria of global approximation, are listed in
Table 3. It can be seen that the T-S fuzzy models derived from the neuro-fuzzy model can
provide an accurate approximation of the original nonlinear system. The 12 points T-S
fuzzy model with optimized operating points gives an even better approximation than
the 27 points one with uniformly distributed operating points.
 T-S Fuzzy Model with
Neuro-fuzzy T-S Fuzzy Model with
Uniformly Distributed 27
Model Optimized 12 Points
Points
NDEI 0.0267 0.0697 0.0477
Appl. Sci. 2021, 11, 62 11 of 15

Figure5.5.Experimental
Figure Experimentaldata
dataand
andpredicted
predictedoutputs
outputsfrom
fromthe
theT-S
T-Sfuzzy
fuzzymodels.
models.

ToNDEIs
Table 3. of the
test the neuro-fuzzy
controller model and the
performance, T-S fuzzy models.
reference commands to be tracked are given
by
T-S Fuzzy Model with
T-S Fuzzy Model with
Neuro-Fuzzy Model Uniformly Distributed
 2
0.4t 27 Points t  0.4 Optimized 12 Points

NDEI 0.0267 0.064  0.32  t 0.06970.4  0.4  t  1.875 0.0477
r1  r2   2 (26)
 0.6  0.4  t  2.275  1.875  t  2.275

To test the controller performance, the
0.6reference commands to be tracked are given by
t  2.275
 .

2 t ≤ 0.4
5,0.4t

Because the gear ratio is the command inputs for θ3 and θ4 are given by

0.064 + 0.32(t − 0.4) 0.4 < t ≤ 1.875
r θ1 = r θ2 = r3 (t 5−r12.275
r4 )2 5r1.875 . (26)
0.6 − 0.4
 2
. < t ≤ 2.275 (27)


0.6 t > 2.275
The command inputs for the angular velocities are
Because the gear ratio is 5, thecommand 0.8t inputs for tθ3 and
0.4 θ 4 are given by
 0.32 0.4  t  1.875

r1  r2 rθ3 = 5rθ1 rθ4 = 5rθ2 . (27)
0.8  t  2.275  1.875  t  2.275 (28)
The command inputs for the angular  0velocities aret  2.275
r3  5 r1 r4  5r2

 0.8t t ≤ 0.4
0.32 0.4 < t ≤ 1.875

The performance
rθ. =index
rθ. =is defined as:
1 2 
 − 0.8 ( t − 2.275 ) 1.875 < t ≤ 2.275 (28)
0 t > 2.275

rθ. = 5rθ. rθ. = 5rθ.
3 1 4 2

 h The performance index is defined as: 

2
2 i h
2
2 i
 10, 000 θ1 − rθ1
Z 4 + θ 2 − r θ2 + θ 3 − r θ3 + θ 4 − r θ4 

J= . 2  . 2    . 2  . dt (29)
 2 
0  +1000 θ 1 − rθ
 . + θ 2 − rθ . + θ 3 − rθ . + θ 4 − rθ
. + 0.1T12 + 0.1T22 

1 2 3 4
 where e1 and e2 are tracking error of θ1 and θ2. The PID gains were based on [32], in w
a PID controller was designed for this robot manipulator. Figure 6 shows the simul
Appl. Sci. 2021, 11, 62
responses of the flexible robot system with the fuzzy controllers and PID controller, 12 of 15
w
the mean square errors (MSEs) of command tracking are compared in Table 4. The
formances of the fuzzy controllers are evidently better than that of the PID controller.
PID responses
The fuzzyexhibit
optimallarger
control overshoot
systems derivedandfrom
longer settling
the T-S time, which
fuzzy models with 27 leads
and to la
MSE.12 The
pointsresults of the 12using
were simulated points T-S fuzzy model
the mathematical model are almost with
in MATLAB the same
initial as the results o
condition
x = 0. To obtain an assessment of the proposed control algorithm, it is necessary
27 points one, as the non-significant difference of modeling errors may only yield tr to compare
thison
effect paradigm to alternative
the control system.control
Since techniques.
the numberTherefore, a PID controlleriswas
of linearization also designed:
reduced, the 12 poin
.
S fuzzy model is more computationally

T1
 
0.4e1 +efficient.
R
0.1e1 + 0.1 e1 dt

= . R , (30)
T2 0.4e2 + 0.1e2 + e2 dt
Table 4. MSEs of command tracking with the designed controllers.
where e1 and e2 are tracking error of θ 1 and θ 2 . The PID gains were based on [32],
in which a PID controller
T-S Fuzzy Model withwas designed
27 Pointsfor thisT-S
robotFuzzy
manipulator.
ModelFigure
with6 12
shows the sim- PID
Points
ulated responses of the flexible robot system with the fuzzy controllers and PID controller,
θ1 2.95 × 10−4 2.98 × 10−4 1.20 × 1
while the mean square errors (MSEs) of command tracking are compared in Table 4.
θ2 The performances 7.31 × 10fuzzy controllers are evidently better
of the
−4 8.06than
× 10that of the PID con-1.22 × 1
−4

θ3 troller. The PID 0.0169

responses exhibit larger overshoot and longer0.0170 time, which leads 0.074
settling
θ4 to larger MSE. The0.0084
results of the 12 points T-S fuzzy model are almost the same as the 0.026
0.0097
results of the 27 points one, as the non-significant difference of modeling errors may only
yield trivial effect on the control system. Since the number of linearization is reduced,
the 12 points T-S fuzzy model is more computationally efficient.

Figure 6. Simulation results of the fuzzy optimal controllers for the flexible robot.

Table 4. MSEs of command tracking with the designed controllers.

T-S Fuzzy Model T-S Fuzzy Model

PID
with 27 Points with 12 Points
θ1 2.95 × 10− 4 2.98 × 10− 4 1.20 × 10− 3
θ2 7.31 × 10− 4 8.06 × 10− 4 1.22 × 10− 3
θ3 0.0169 0.0170 0.0744
θ4 0.0084 0.0097 0.0268

The fuzzy optimal controllers were then implemented on the robot. The experimen-
tal results are shown in Figure 7. The achieved tracking performances are good and
close to the simulation results, except for some lags and oscillations along the trajectories.
 Figure 6. Simulation results of the fuzzy optimal controllers for the flexible robot.

The fuzzy optimal controllers were then implemented on the robot. The experime
Appl. Sci. 2021, 11, 62 13 of 15
results are shown in Figure 7. The achieved tracking performances are good and clo
the simulation results, except for some lags and oscillations along the trajectories. T
oscillations are caused by the flexible joints of the robot, which are very difficult to c
These oscillations are caused by the flexible joints of the robot, which are very difficult to
pletely eliminate.
completely eliminate.

Figure 7. Experimental results of the fuzzy optimal controllers for the flexible robot.
Figure 7. Experimental results of the fuzzy optimal controllers for the flexible robot.

6. Conclusions
6. Conclusions
An explicit procedure of establishing the T-S fuzzy models of unknown nonlinear
An explicit
systems proceduredata
from experimental of isestablishing
presented: (1) the T-S constituent
the local fuzzy models modelsofareunknown
obtained nonli
systems from
by direct experimental
linearization data is presented:
of neuro-fuzzy (1) the
models trained fromlocal constituent
data so that theymodels
are closeare obta
by direct linearization
approximations of neuro-fuzzy
to the local linearization ofmodels trained
the original from
nonlinear data(2)
systems; soThe
that they are close
operating
points for linearization are optimized using the evolutionary
proximations to the local linearization of the original nonlinear systems; (2) strategy to achieve good
The opera
global approximation with a reduced number of linearization. The controller design based
points for linearization are optimized using the evolutionary strategy to achieve g
on T-S fuzzy model has also been discussed. The derived fuzzy optimal controller was
global approximation
applied with a reduced
to a nonlinear flexible-joint number
robot system andofcompared
linearization.
with theThe controller
alternative design b
control
on T-S fuzzywhich
technique, model has also been
demonstrated discussed.ofThe
the effectiveness derivedmethod
the proposed fuzzy foroptimal controller
controlling
nonlinear dynamic systems whose analytical models are not available.
applied to a nonlinear flexible-joint robot system and compared with the alternative Compared with
trolthe existing T-S model methods, the proposed method could more effectively address
technique, which demonstrated the effectiveness of the proposed method for con
the tradeoff between local and global approximation for complex systems. Moreover,
lingsince
nonlinear dynamic methods
the computational systems ofwhose analytical
the proposed methodmodels are on
are based nota available.
set of well- Comp
withdeveloped
the existing T-S model
tools (e.g., methods,
FBFN training, the proposed
ES optimization), method
it can could
be readily more
used effectively
for data-
dress the system
driven tradeoff betweenand
identification local and global
controller design approximation for complex
in realistic applications. The futuresystems.
work Mo
might include studying the effect of uncertainty in data and the
ver, since the computational methods of the proposed method are based on a set of w modeling errors of the
neuro-fuzzy and T-S fuzzy models, such that the robustness of the T-S fuzzy control scheme
developed tools (e.g., FBFN training, ES optimization), it can be readily used for d
can be more explicitly addressed.
driven system identification and controller design in realistic applications. The fu
work might
Author include studying
Contributions: the effect
Conceptualization, of uncertainty
B.Z. and in data
Y.C.S.; methodology, B.Z. and the modeling
and Y.C.S.; software, erro
B.Z.; validation, B.Z.; formal analysis, B.Z.; investigation, B.Z.; resources, Y.C.S.; data curation, B.Z.;
the neuro-fuzzy and T-S fuzzy models, such that the robustness of the T-S fuzzy con
writing—original draft preparation, B.Z.; writing—review and editing, Y.C.S.; visualization, B.Z.;
scheme can be
supervision, more
Y.C.S.; explicitly
project addressed.
administration, Y.C.S.; funding acquisition, Y.C.S. All authors have read
and agreed to the published version of the manuscript.
Author Contributions: Conceptualization, B.Z. and Y.C.S.; methodology, B.Z. and Y.C.S.; softw
B.Z.;Funding: This research
validation, received
B.Z.; formal no external
analysis, funding.
B.Z.; investigation, B.Z.; resources, Y.C.S.; data curation,
writing—original draft preparation, B.Z.; writing—review and editing, Y.C.S.; visualization,
supervision, Y.C.S.; project administration, Y.C.S.; funding acquisition, Y.C.S. All authors have
and agreed to the published version of the manuscript.
Appl. Sci. 2021, 11, 62 14 of 15

Data Availability Statement: The data presented in this study are available on request from the
corresponding author, Upon reasonable request, and the code used during the study proprietary.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Fantuzzi, C.; Rovatti, R. On the approximation capabilities of the homogeneous Takagi-Sugeno model. In Proceedings of the Fifth
IEEE International Conference on Fuzzy Systems, New Orleans, LA, USA, 11 September 1996; pp. 1067–1072.
2. Johansen, T.A.; Shorten, R.; Murray, S. On the interpretation and identification of dynamic Takagi–Sugeno models. IEEE Trans.
Fuzzy Syst. 2000, 8, 297–313. [CrossRef]
3. Sugeno, M. On stability of fuzzy systems expressed by fuzzy rules with singleton consequents. IEEE Trans. Fuzzy Syst. 1999,
7, 201–224. [CrossRef]
4. Khalil, H.K.; Grizzle, J.W. Nonlinear Systems, 3rd ed.; Prentice Hall: Upper Saddle River, NJ, USA, 2002.
5. Takagi, T.; Sugeno, M. Fuzzy identification of systems and its applications to modeling and control. IEEE Trans. Syst. Man Cybern.
1985, 15, 116–132. [CrossRef]
6. Feng, G. A Survey on Analysis and Design of Model-Based Fuzzy Control Systems. IEEE Trans. Fuzzy Syst. 2006,
14, 676–697. [CrossRef]
7. Sala, A.; Arino, C. Asymptotically necessary and sufficient conditions for stability and performance in fuzzy control: Applications
of Polya’s theorem. Fuzzy Sets Syst. 2007, 158, 2671–2686. [CrossRef]
8. Chen, C.W. Application of fuzzy-model-based control to nonlinear structural systems with time delay: An LMI method.
J. Vib. Control 2010, 16, 1651–1672. [CrossRef]
9. Lam, H.K. LMI-Based Stability Analysis for Fuzzy-Model-Based Control Systems Using Artificial T–S Fuzzy Model. IEEE Trans.
Fuzzy Syst. 2011, 19, 505–513. [CrossRef]
10. Lam, H.K.; Lauber, J. Membership-function-dependent stability analysis of fuzzy model based control systems using fuzzy
Lyapunov functions. Inf. Sci. 2013, 232, 253–266. [CrossRef]
11. Li, P.W.; Zhang, W. Towards a unified stability analysis of continuous-time TS model-based fuzzy control systems. Int. J. Model.
Identif. Control 2019, 31, 113–123. [CrossRef]
12. Zhao, T.; Huang, M.B.; Dian, S.Y. Robust stability and stabilization conditions for nonlinear networked control systems with
network-induced delay via TS fuzzy model. IEEE Trans. Fuzzy Syst. 2019. [CrossRef]
13. Li, C.; Zhou, J.; Fu, B.; Kou, P.; Xiao, J. T–S fuzzy model identification with a gravitational search-based hyperplane clustering
algorithm. IEEE Trans. Fuzzy Syst. 2011, 20, 305–317. [CrossRef]
14. Zou, W.; Li, C.; Zhang, N. A T–S fuzzy model identification approach based on a modified inter type-2 FRCM algorithm.
IEEE Trans. Fuzzy Syst. 2017, 26, 1104–1113. [CrossRef]
15. Stefan, J.; Christoph, H.; Nikolaus, K. Total least squares in fuzzy system identification: An application to an industrial engine.
Eng. Appl. Artif. Intell. 2008, 21, 1277–1288.
16. Li, L.Q.; Wang, X.L.; Xie, W.X.; Liu, Z.X. A novel recursive TS fuzzy semantic modeling approach for discrete state-space systems.
Neurocomputing 2019, 340, 222–232. [CrossRef]
17. Roy, A.; Briczinski, S.J.; Doherty, J.F.; Mathews, J.D. Genetic algorithm based parameter estimation technique for fragmenting
radar meteor head echoes. IEEE Geosci. Remote Sens. Lett. 2009, 6, 363–367. [CrossRef]
18. Lin, G.H.; Zhao, K.Y.; Wan, Q. Takagi-Sugeno fuzzy model identification using coevolution particle swarm optimization with
multi-strategy. Appl. Intell. 2016, 45, 187–197. [CrossRef]
19. Yen, J.; Wang, L.; Gillespie, C.W. Improving the interpretability of TSK fuzzy models by combining global learning and local
learning. IEEE Trans. Fuzzy Syst. 1998, 6, 530–537. [CrossRef]
20. Wang, L.X.; Mendel, J.M. Fuzzy basis functions, universal approximation, and orthogonal least Squares learning. IEEE Trans.
Neural Netw. 1992, 3, 807–814. [CrossRef]
21. Kosko, B. Fuzzy systems as universal approximators. IEEE Trans. Comput. 1994, 43, 1329–1333. [CrossRef]
22. Lee, C.W.; Shin, Y.C. Construction of fuzzy systems using least-squares method and genetic algorithm. Fuzzy Sets Syst. 2003,
137, 297–323. [CrossRef]
23. Teixeira, M.; Żak, S.H. Stabilizing controller design for uncertain nonlinear systems using fuzzy models. IEEE Trans. Fuzzy Syst.
1999, 7, 133–142. [CrossRef]
24. Lee, C.W.; Shin, Y.C. Evolutionary modelling and optimization of grinding processes. Int. J. Prod. Res. 2000, 38, 2787–2813. [CrossRef]
25. Ho, W.H.; Chen, S.H.; Chen, I.T.; Chou, J.H.; Shu, C.C. Design of stable and quadratic-optimal static output feedback controllers
for TS fuzzy-model-based control systems: An integrative computational approach. Int. J. Innov. Comput. Inf. Control 2012,
8, 403–418.
26. Kirk, D.E. Optimal Control Theory; Prentice Hall: Englewood Cliffs, NJ, USA, 1970.
27. Kim, E.; Kim, S. Stability analysis and synthesis for an affine fuzzy control system via LMI and ILMI: A continuous case.
IEEE Trans. Fuzzy Syst. 2002, 10, 391–400.
28. Ma, X.J.; Sun, Z.Q.; He, Y.Y. Analysis and design of fuzzy controller and fuzzy observer. IEEE Trans. Fuzzy Syst. 1998, 6, 41–51.
29. Mohamed, Z.; Tokhi, M.O. Command shaping techniques for vibration control of a flexible robot manipulator. Mechatronics 2004,
14, 69–90. [CrossRef]
Appl. Sci. 2021, 11, 62 15 of 15

30. Nho, H.C. An Experimental and Theoretical Study of Various Control Approaches to Flexible-Joint Robot Manipulator Undergoing
Payload Changes. Ph.D. Thesis, Purdue University, West Lafayette, IN, USA, 2004.
31. Chatlatanagulchai, W. Backstepping Intelligent Control Applied to a Flexible-Joint Robot Manipulator. Ph.D. Thesis,
Purdue University, West Lafayette, IN, USA, 2006.
32. Chatlatanagulchai, W.; Meckl, P.H. Model-independent control of a flexible-joint robot manipulator. J. Dyn. Syst. Meas. Control
2009, 131, 041003. [CrossRef]