Sunteți pe pagina 1din 10

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 1

Approximation-Free Tracking Error


Constraint Control for Lagrangian Systems
Seong Ik Han, Member, IEEE, and Jang Myung Lee, Senior Member, IEEE

 the controller. After the state space approach appeared in the


Abstract—This paper proposes an approximation-free 1960s, model-based linear control techniques such as linear
adaptive tracking error constraint control scheme for quadratic (LQ), linear quadratic Gaussian (LQG), linear
unknown Lagrangian systems. A transformed filtered quadratic Gaussian/loop transfer recovery(LQG/LTR), and
tracking error surface has been constructed to guarantee H that linearize nonlinear Lagrangian systems have been
the transient and steady-state performance of the position
control of the Lagrangian systems combined with adaptive developed. However, because these control schemes are
estimation laws for unknown upper bounds of the designed based on the plant model, an additional robust control
parameters of the Lagrangian systems. To obtain fast error algorithm is required to compensate for unmodeled or varying
convergence, a finite-time control has been adopted in the plant dynamic characteristics. Moreover, tedious modeling and
proposed controller. In addition, the proposed controller is linearization of complex nonlinear systems is performed. To
effective for the constrained positioning of Lagrangian alleviate these problems, fuzzy logic systems (FLSs) and neural
mechanical systems with unknown deadzone input networks (NNs) have been developed to approximate unknown
nonlinearity, without requiring nonlinear approximations complex nonlinear dynamics or unknown parameters. There
or identification processes for unknown parameters of
have been many control applications to servo systems using
dynamics and deadzone. The effectiveness of the
proposed constraint control scheme has been proven by
FLS [1-3] and NN [4-6] techniques. However, in servo
simulation and experimental evaluations for an XY table mechanical systems, the use of FLS-and NN-based
and an articulated manipulator system in the presence of approximation techniques makes the controller structure more
unknown deadzone. complex and leads to large computation times as the order of
the system and the number of fuzzy logic units, neurons, and
Index Terms—Lagrangian system, constrained control, layers are high. To solve these problems, robust adaptive
approximation-free estimation, finite-time control, control methods for robotic systems were developed by Wang
deadzone et al. [7], Li et al. [8],[9], and Andaluz et al.[10]. Using
adaptive laws for the unknown upper bounds of Lagrangian
dynamic parameters, the estimates of unknown parameters are
I. INTRODUCTION obtained so as to provide satisfactory control performance

I NDUSTRIAL mechanical systems including mobile and


industrial robotic systems and several servo systems
without introducing FLS or NN approximations. However,
only simulated investigations have been used to prove validity
of the proposed control strategy. Moreover, the constraint
equipped with electric, hydraulic, and pneumatic energy
conversion actuators are typically described by the Lagrangian control methods for tracking error are not considered.
dynamic equations of motion. Over the last few decades, for These conventional adaptive controls depend on the design
non-model-based control, many proportional-derivative (PD) parameters and certain unknown terms for minimizing the
and proportional -integral- derivative (PID) tuning control norm of the tracking error, and the transient and residual
techniques have been applied to Lagrangian systems, because tracking errors. The size of a residual error can be regulated so
these control methods are easy to design and convenient to that these controllers are designed to guarantee the
implement in real servo systems. However, repeated tuning convergence of a tracking error to a residual set. On the other
processes are required to obtain satisfactory control hand, repeated trial-and-error tests are required to ensure the
performance, although there are empirical gain-tuning rules of prediction of a precise upper bound for the required residual set
and satisfy the transient and steady-state tracking performances
Manuscript received February, 10, 2015; revised July 14, 2015,
by selecting the specific controller parameters. Therefore, in
February, 4, 2016, and June 24, 2016; accepted August 22, 2016. This these control structures, it is difficult to develop a systematic
work was supported by the National Research Foundation of Korea scheme that guarantees a specified control performance.
(NRF) grant funded by the Korean Government (MSIP) Recently, the prescribed performance constraint control
(NRF-2015R1A2A 2A01004457).
Seong Ik Han is with the Department of Electronic Engineering, methods have been actively studied by Rovithakis et al., [11],
Pusan National University, Busan, 46241 Republic of Korea [12], Wang et al.,[13], Na et al., [14], and Han et. al. [15].
(e-mail:snikhan @gmail.com). These techniques provide a prescribed performance function
Jang Myung Lee is with the Department of Electronic Engineering, and transformation function that transform the tracking error of
Pusan National University, Busan, 46241 Republic of Korea (phone:
+82-51 -510-2378; fax:82-51-514-1693; e-mail:jmlee@pusan.ac.kr). an original nonlinear system into a new error in the transformed
system. The transformed error is used as a variable in the design

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 2

of an adaptive controller. The tracking error remains within the Property 3: qT [ M (q )  2C (q, q )]q  0 is always true, based on
prescribed error bound to ensure the stability of the transformed
error system. However, in the constraint method given in the skew-symmetric property between M ( q ) and C (q, q ) .
[11-14], there is a problem, in that the inverse of the Assumption 1 [7]: There exist some unknown finite positive
transformation function to be used to design the controller constants, i  0 , 1  i  4 , and finite nonnegative constants,
might violate the prescribed performance under some specific
5  0, such that q  R n , q  R n , M (q)  1 , C(q, q )  2
constraint condition [15]. On the other hand,
Theodorakopoulos et al. [16],[17] verified recently that the  3 q , G ( q )   4 , and supt  0 Fu  5 .
prescribed performance control can compensate for the The control objectives of the proposed controller are to design a
deadzone input nonlinear effect, which is obstacle for improved control input law such that
control performance in many Lagranian systems, without using 1. The output position vector tracks the desired trajectory
conventional adaptive or neural/fuzzy based approximation
qd (t )  R n and all involved signals in the closed-loop
techniques.
Based on the aforementioned directions, in this paper, a new system remain bounded.
filtered tracking error surface is proposed to guarantee the 2. The tracking errors, e(t )  q (t )  qd (t ) , always remain
prescribed tracking error constraint boundary and deadzone within the prescribed constraint boundaries.
compensation for the Lagrangian systems; it is combined with
robust adaptive control without using a FLS or NN
B. Constrained Tracking Error Boundary Function
approximation for unknown dynamic parameters. Moreover,
deadzone effect is compensated for without requiring any A constrained positive boundary function to prescribe the
additional controller or approximation technique like [16],[17]. tracking performance is selected as follows because this type
Next, finite-time control used in the terminal sliding mode function can regulate conveniently the tracking performances
control is combined with robust adaptive control with unknown such as rising-time, overshoot, and steady-state offset in time-
parameter estimation of the Lagrangian system in order to domain:
obtain a fast- converging response. Therefore, the proposed  (t )  ( 0   ss )exp(at )   ss , (2)
control assures the predefined position tracking error
constraints for the Lagrangian systems while not depending on where 0   ss  0 and  ss  lim  (t ) are design constants
t 
complex approximation methods and not requiring extra
deadzone compensators [18-20]. As application examples, the and e(0)  0 should be satisfied to guarantee the prescribed
control systems of an XY table and an articulated manipulator constraint conditions. 0 and a  0 determine the rising time
system are utilized to show the effectiveness of the proposed
and overshoot performance of the tracking output.  ss
control scheme.
determines the steady-state tracking performance of the given
control system. Each design parameter must be selected
II. SYSTEM DESCRIPTION appropriately according to the control system conditions and
capacity of the actuators. Extreme parameter values may cause
A. Dynamics of Lagrangian Systems high gain control and stability problem. The prescribed bounds
The dynamic equation of the Lagrangian systems with n are guaranteed by the following constraint conditions
degrees-of-freedom is as follows:
  (t )  e(t )   (t ) if e(0)  0 , (3)
M (q )q  C (q, q ) q  G (q )  Fu  u (1) or
  (t )  e(t )   (t ) if e(0)  0 , (4)
n1
where q, q , q  R denote the generalized position, and the
where   diag ( 1 ,...,  n ) and 0  i  1 is a scale factor. Next,
velocity and acceleration vectors, respectively; a positive
a transformed constraint error, i for i  1,..., n , is defined as
definite symmetric matrix, M (q )  R nn , is the moment of
follows:
inertia matrix; C (q, q )  R nn is the centripetal Coriolis
matrix; G (q )  R n1 is the gravity vector; Fu  R n1 is the i (t ) 
ei (t )
, (5)
uncertainty vector including the external disturbance and the i (t )
friction; and u  R n1 is the control input vector. i  hi p  li (1  p) , (6)
Property 1: M (q) and G (q) are uniformly bounded and
uniformly continuous for a uniformly bounded and uniformly where p  1 if ei (t )  0 , and p  0 if ei (t )  0 . The parameters
continuous generalized position vector q . hi and li are defined as follows:
Property 2: C (q, q ) is uniformly bounded and uniformly
continuous for a uniformly bounded and uniformly continuous hi  i (t )
 , if ei (0)  0 , (7)
generalized velocity vector q . li   i i (t )

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 3

hi  i i (t ) 5
1
 , if ei (0)  0 . (8)  r T ( mT  m  u )   k ˆ k . (15)
 li   i (t ) k 1  k

In (3) and (4),  determines the symmetry or not of the The control input and adaptive law can be chosen as follows:
constraint condition. In this paper, we assume that   I to 5
r ˆ k k2
u   Kr     sign(r ) , (16)
consider only the symmetric constraint condition. k 1 r  k   k
Lemma1: The following condition regarding the transformed 2
error defined in (2) holds:  k r  k2
ˆ k    k ˆ k , k  1,...,5 , (17)
r k   k
0  i  1 , t  0, i  1,..., n (9)
where K  diag (k1 ,..., kn )  0  R nn , k  0 and  k  0 are
if and only if 0i ,  ssi , and ai are selected such that they satisfy
constants,   [   q  1 1]T , and   diag ( 1 ,...,
(3) and (4).
Proof: See [15].  n )  0 is a constant matrix. (15) can be written as

III. DEIGN OF NON-MODEL-BASED CONTROLLER 5


r ˆ k  k2
V1   r T Kr  r T  mT  m  r T   r
A. Adaptive Control without Constraint Control k 1 r  k   k
2
Consider the following signals 5 r  k k2 5

   k  k ˆ k
k 1 r k   k k 1 k
  qd  e , (10)
2
5 k k2
r 5

  r T Kr  r  mT m     k  k ˆ k
where   diag (1 ,..., n )  0 is a constant matrix. The filtered k 1 r k   k k 1 k
error surface r and its derivative r are defined as 2
5 r  k k2 5

  r T Kr  r  T      k  k ˆ k
r  q   , (11) k 1 r k   k k 1  k
r  q   . (12) 5 5
k
 r T Kr    k  k    k ˆ k
i 1 k 1  k
Based on the terms of the definitions of (11) and (12), (1) can
k 2 5 
5
 2 
be written as   r T Kr    k     k  k  k k 
k 1 2 k i 1  2k 
M (q )r  C (q, q )r  M (q )   C (q, q )   G (q )  Fu  u 5

 min ( K ) r   k  k2  
2
 C (q, q ) r   mT m  u , (13)
k 1 2 k
  V1   , (18)
where  m  [ M (q ) C ( q, q ) G (q ) Fu ]T ,  m  [   1 1]T ,
 m  [ M (q) C (q, q ) G(q) Fu ]T  [ 1 2  3 q 4 where   min[min ( K ),k / 2k ] / max[max ( M (q)),1/ k ]  0
5 ] , and i , i  1,...,5, are unknown finite positive constants 5 
  2 
  [ 1 2 3 4 5 ]T , and      k  k  k k  .
given in Assumption 1. A Lyapunov function candidate can k 1  2k 
be defined as follows:
Multiplying (18) by e t gives
5
1 T 1 2
V1  r M (q)r    k , (14)
2 k 1 2 k
d
dt
 
V1e t   e t . (19)

where  k   k  ˆ k , k  1,...,5, ˆ k are estimates of unknown


  t
 k , and k  0 are constants. The time derivative of (14) with Noting that 0  e t  1 , e  0 and (V1 (0)   /  ) e t

Property 3 and (13) is written as
 V1 (0) and integrating (19) over [0, t ] with the fact that leads
5 to the following inequality
 1  1
V1   r T C (q, q )  M (q )  r    k  k
 2  k 1  k     t  
 r T [ M (q )   C (q, q )   G (q )  Fu  u ] 0  V1 (t )   V1 (0)   e   V1 (0)  , t  0 . (20)
    

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 4

Therefore, all error signals are semiglobally, uniformly, and (ii) when   ˆ , the filtered tracking errors then converge to
ultimately bounded [21] and e  0 and e  0 as t   by the
the regions r f    min(1 , 2 ) of the neighborhood in
Barbalat’s lemma [22].
rf  0 in finite time if  min and  m are selected
B. Adaptive Finite-Time Control without Constraint as  min   / V2 and  min   / V2 m , respectively, where
Control
 
1/ 2 m
The command vector  f and its derivative  f for finite time 1  2 /  min and 2  2 m  /  min .

performance are defined as follows: Proof: Theorem 1 can be verified easily from similar results in
[23],[24] and proof is omitted.
 From Theorem 1, it follows that e and e also converge to
 f  qd  1e  2 e sign(e) , (21) zero or the neighborhood in e  0 and e  0 in finite time.
 f  qd  1e  2 diag ( ei
 1
)e , (22) Remark 1: In (22), to avoid singularity because of 0    1 ,
the following condition is used as
where 1  diag (11 ,..., 1n )  0 and 2  diag (21 ,..., 2n )  0
 e  1 e, if e  0 and e  0
are a constant matrix and 0    1 is constant. The filtered 
  1
error surface r f and its derivative rf for finite time e    e, if e  0 and e  0 , (27)
 0, otherwise e  0
performance are defined as r f  q   f and rf  q   f , 
respectively. A Lyapunov function candidate for finite time where   0 is some small constant.
performance is defined as follows:

1 T 5
1 2 C. Adaptive Error-Constrained Finite-Time Control
V2  r f M (q)rf    k . (23) In this section, a new filtered error surface that guarantees
2 k 1 2 k
the error-constrained condition defined in the section II is
proposed. From (5) and (6), a command error vector with a
The following finite-time-based control and adaptive law are
constraint function  (t ) and its derivative (t ) are defined as
proposed as follows:

5 rf ˆ k  2fk   c  d  
1  2 e sign(e) ,
u   Kr f     rf sign(r f ) , (24)
k 1 rf  fk   k (28)
 1
2 c  d  
1   2 diag ( ei )e , (29)
k rf  2fk
ˆ k    k ˆ k , k  1,...,5 , (25)
r f  fk   k where   [ 1 ,..., n ]T is a transformed error vector,  i 
(1 i )1i , d   1qd  [d1,...,dn ]T , and  di  i1qdi .
where  f  [  f  f q  f T
1 1] . Similarly to (18), it
Defining  q   1q  [ q1 ,...,  qn ]T and  qi  i1qi , we then
follows that
have
2 n  i 1 5
k 2
V2  min ( K ) r f    i r fi   k     q  d , (30)
i 1 k 1 2 k

  2 ( i 1)/2 k
2 n 5
 min ( K ) rf    i rfi   k2   where   [1,..., n ]T and   diag (1,...,n ) . The constrained
i 1 k 1 2k filtered error surfaces rc and its derivative rc are defined as
m
  V2   mV2   , (26)
rc  q   c , (31)
m m
where  m  2 min[ i / max(max (M (q)) )] and  m  ( i min
rc  q  c . (32)
1) / 2 .
Theorem 1 (finite-time stability): Consider the closed-loop of a Based on the terms of (28)-(32), the dynamics in (1) can be
system consisting of the plant in (1), and following the control written as
and adaptive laws in (24) and (25). If Assumption 1 holds, then
the following conditions are satisfied:
M (q)rc  M (q)q  M (q) c
(i) when   ˆ , the filtered tracking errors converge to zero in
a finite time.   1[C (q, q )q  G (q)  Fu  u ]  M (q) q  M (q) c

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 5

 C (q, q )[rc  c   1q]   1[G(q)  Fu  u ] Remark 3: The achievement of the properties (3) and (4)
provides that the desired tracking error performance
 M (q ) q  M ( q ) c specifications can be explicitly determined by appropriately
T
 C (q, q )rc   mcmc  1u , (33) selecting the parameters of the function  ; maximum
overshoot, minimum convergence rate, and maximum
steady-state error are determined by combination
where  mc  [ M (q ) C (q, q ) G ( q ) Fu ]T ,  mc  [ c  c  1
of 0 and a , a , and  ss , respectively. If the evolution of the
1]T ,  mc  [ M (q) C (q, q ) G (q) Fu ]T  [ c1 c 2 tracking error violates the constrained boundaries or required
c3 q c4 c5 ] , ci  i , i  1,..., 4, Fu  c5 , Fu   1 Fu control effort is not reasonable, the values of K ,  , and  as
well as  , which are specific to the application consideration,
 1C(q, q)q  M (q) q ,  q  [ qi ,..., qn]T , and  qi 
need to be adjusted.
i1 (qii   qii ) i2 (qii   qii2 ) . Considering the The graphical presentation for the error constraint in (28) and
following Lyapunov function candidate (31) is illustrated in Fig. 1.

5
1 T 1
V2  rc M (q) rc   ck2 , (34)
2 k 1 2ck

its time derivative with the result in (34) can be written as

5
1 1
V2  rcT C (q, q )rc  rcT M (q )rc    ck ck
2 k 1 ck

 rcT [ M (q ) c  C (q, q )  c   1G ( q )  Fu   1u ]


5
1
 rcT ( mc
T
mc   1u )   ck ˆ ck . (35)
k 1 ck
Fig. 1. Graphical presentation for the error constraint

The control input and adaptive laws are selected as follows:


D. Adaptive Error-Constrained Finite-Time Control for a
Lagrangian System with Deadzone Input Nonlinearity
 5
r ˆ  2  
u    Krc   c ck ck   rc sign(rc )  , (36) The recent tracking error constraint control schemes [16],
 k 1 rc ck   ck  [17] proved analytically the fact that if a prescribed error
2 2 performance is satisfied, the deadzone input nonlinear effect
ck rc ck
ˆ ck   ˆ ck , k  1,...,5 ,
 ck (37) can be compensated for its absence without requiring
rc ck   ck knowledge of deadzone characteristics, furthermore, either
utilizing adaptive observation techniques or neural/fuzzy
approximation scheme, which have been adopted frequently for
where c  [ c c q c  1 1]T , ck
  0 , and  ck  0
removing negative deadzone effect [18-20]. In this direction, it
are constants. Considering (36) and (37), we can obtain will be shown via both simulation and experiment that the
proposed error-constrained scheme has also aforementioned
5  
 1
5 
ck  2 property provided the prescribed constrained performance is
V2  rcT Krc   rc   ck2    ck  ck  ck ck  satisfied. However, in the previous studies [16],[17] only
k 1 2ck i 1  2ck  simulation demonstrations were presented. The dynamic

  2 m 
ck
n 5
equation of the Lagrangian systems with deadzone input
  i rfi 
2
 min ( K ) rc ck2   c , (38)
i 1 k 1 2 ck
nonlinearity is written as
5
where  c    ck  ck  ck
k 1
2
 ck / 2ck . Therefore, finite-time  M (q)q  C (q, q ) q  G (q )  Fu  D(u ) . (40)

stability is obtained, similarly to the case of Theorem 1. The deadzone function D : R  R is defined as
Remark 2: In order to satisfy the given constrained tracking
performance,  and  should be selected such that
 gl (u ) , u  bl
D (u )   0,  bl  u  br . (41)
  (0)  e(0) , if q(0)  qd (0)
 . (39)  g r (u ) , u  br
   0, if q(0)  qd (0)

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 6

The constants bl , br  0 are not necessarily equal and known. depend on the signal rc . In this respect, the proposed constraint
Their values determines the dead-band size. The functions gl : control scheme is highly robust against the dead-band size, i.e.,
( , bl ]  R , g r : [bl , )  R are locally Lipschitz, strictly the value of b  0 . However, as the dead-band size increases,
the price to be paid is the increase of the corresponding control
increasing and unknown to guarantee that D (u ) is locally
effort only by adjusting the value of K and  .
Lipschitz as well. The constraint control problem under
deadzone input nonlinearity is that the error-constrained
IV. APPLICATION EXAMPLES
condition defined in the section II is satisfied by using the
control scheme proposed in (36) without considering any
additional deadzone compensator. In this section, a solution for In this section, simulated and experimental evaluations of the
this problem is presented simply. More rigorous verifications proposed control scheme applied to the XY table and the
for compensating deadzone effect by the prescribed error articulated manipulator systems are described.
control scheme were presented in [17]. Defining F2  M (q) c A. Simulation and Experiment for the XY Table System
1
C (q, q )  c   G (q )  Fu , (35) is written as Fig. 1 shows the structure of the XY table control system, for
which the dynamic equation is written as
V2   rcT [ M (q ) c  C (q, q )  c   1G ( q )  Fu   1 D (u )]
5
M ( q)q  C (q, q)q Fu  D(u) , (43)
1
  ck ˆ ck
k 1 ck where deadzone appears due to misalignment between the
5
1 motor axis and the ball-screw. In the XY table system, the
 rcT  F2   1 D(u )     ck ˆ ck gravity dynamics do not appear in (43) because the XY table
  ck k 1
5
lies on the horizontal plane. In this system, M ( q) contains the
1
 rcT  1   F2  D(u )    ck ˆck . (42) unknown moment of inertia of each axis, which contains the
 k 1 ck inertia of the servo motors. The friction dynamics of the
ball-screws, linear motion guides, and servo motors are
Owing to the boundedness of M ( q) , C (q, q ) , G (q) from the included in Fu . The controller is designed not to include any
properties1, 2 and the boundeness by construction of c ,  c , dynamics or identify any of the parameters of the XY table
 with assumption of the boundedness of ck , Fu , it follows system. The specifications of the ball-screw, servo motor, and
sensor are given in Table 1. Three controllers are designed to
that F2 is bounded as well. Then, the existence of an unknown evaluate the proposed control scheme: the filtered-error-based
positive constant  * such that  F2   * , t  [0, ) is adaptive controller without finite-time and constraint control
guaranteed. By continuity of D and the monotonicity of (WO-FC), the filtered-error-based adaptive controller without
functions g r and gl , there always exists an unknown positive constraint control (WO-C), and the proposed controller
(Proposed). The WO-FC controller is designed as follows:
constant d * such that u  d *  D(u )   * .Thus, V   2 c

u  d * or, rc  
1
K1d *
5
r ˆ k k2
when equivalently when u   Kr     tanh(r ) . (44)
* k 1 r k   k
because fro u   rc K  d ,
5
1
where K  K   ˆckck
2
/ ( rc ck  ck )   rc .
k 1
Therefore, rc and u are bounded. The boundedness of signal
1
rc implies that rc  max{rc (0),  K1d * } .
Remark 4: Similar to the statements of [16], to obtain insight on
the operation of the proposed constraint scheme, deadzone is
assumed to have the symmetric and linear slope that
br  bl  b , g r (u )  u  b , and gl (u )  u  b . It is not difficult Fig. 2. Structure of the XY table control system

to verify that d *   *  b and thus rc  max{rc (0), TABLE I


XY TABLE SPECIFICATION
1
 K1 ( *  b)} . Because of the independence of  * on b , it Component Specification
is straightforward that in the presence of the deadzone, Servo motor type HP-KP23
i.e., b  0 , rc may become larger; even so, the properties of Servo amp. type MR-J3-20B
Lead of ball-screw 10mm (X, Y axis)
the proposed constraint controller remain unaltered, since it Motor power 200W (X-axis), 100W(Y-axis)
only requires rc  L and the constraint control action is mainly
Resolution of linear encoder 262144p/rev (X, Y axis)

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 7

The designed controllers in Fig. 2 are implemented in the


Matlab RTI system via a MF624 board with 0.001 sec of
sampling time. The prescribed performance functions are
selected for the following two cases:  x  (3  0.08)e 9t
0.08 (mm) and  y  (3  0.1)e 9t 0.1 ( mm) . The
parameters for the control and adaptive laws are 11  40 ,
21  30 , 12  10 , 22  10 , k1  50 , k2  30 ,  1  30 ,
 2  20 ,  1  0.6 ,  2  0.6 , cx1  0.5 , cx 2  1.5 ,
cx3  0.8 , cx5  1 , cy1  0.5 , cy 2  1 , cy 3  1 , cx5  0.9 ,
 1  0.005 , cx
cx  2  0.002 , cx
 3  0.002 , cx
 5  0.001 ,
 2  0.0025 , cy 3  0.005 , cx5  0.001 ,
 1  0.005 , cy
cy
 x  0.1 , and  y  0.1 . The position commands are selected
as xd  2sin(0.8 t )(mm) and yd  2sin(0.8t)(mm) .
Fig. 3 shows the position tracking results via simulation
(Figs. 3 (a)-(f)) and experiment (Figs. 3 (g)-(k)) along x axis
and y axis, respectively. As shown in Figs. 3 (a), (b), (d), and
(e), the position tracking errors of the WO-FC and WO-C
schemes violate the prescribed constraint condition and the
control performances are sensitive to the variations of the
dead-band in the control input, whereas as shown in Figs. 3 (c)
and (f), the tracking errors of the proposed control system
remain within the constraint boundaries and are robust to the
variations of the dead-band in the control input.
Experimental results given in Figs. 3 (g) and (h) show that
the proposed scheme satisfies the prescribed control
performance. On the other hand, the WO-FC and WO-C
schemes don’t satisfy the prescribed performance. As stated in
the section III, the tracking errors of the proposed control
scheme remains within the prescribed boundary regardless to
deadzone compensation via the inverse deadzone method as
shown in Fig. 3 (i) though the inverse deadzone was used. The
root mean square (RMS) values for the tracking error of
WO-FC, WO-C, and proposed control scheme are presented in
Table II. The RMS error sizes of the WO-C and proposed
systems are small, reaching a maximum of 34% and 24% that

Fig. 3. Simulation and experimental results of the XY table system. 1)


Simulation results for the dead-band variation: (a) Tracking errors of
WO-FC scheme in the x-axis. (b) Tracking errors of WO-C scheme in
the x-axis. (c) Tracking errors of Proposed scheme in the x-axis. (d)
Tracking errors of WO-FC scheme in the y-axis. (e) Tracking errors of

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 8

WO-C scheme in the y-axis. (f) Tracking errors of Proposed scheme in  1  60 ,  2  40 ,  1  0.2 ,  2  0.2 , c11  0.2 , c12  0.2 ,
the y-axis. 2) Experimental results: (g) Tracking error of WO-FC, WO-C,
and Proposed schemes in the x-axis. (h) WO-FC, WO-C, and Proposed cx3  0.8 , c14  1 , c15  0.8 , c 21  1.5 , c 22  2 ,
schemes in the y-axis. (i)Tracking errors using the Proposed scheme
with(W)/without(WO) deadzone compensation in the x-axis. (j) The c 23  2 , c14  3 , c 25  3 , c11  0.01 , c12  0.01 ,
control inputs of each scheme in the x-axis. c13  0.001 , c14  0.035 c15  0.025 , c 21  0.02 ,
of the WO-FC system because they use finite-time and c 22  0.01 , c 23  0.001 , c 24  0.03 , c 25  0.025 1  0.1 ,
constraint control implemented in the controller. As mentioned and  2  0.1 . The joint position commands are selected
previously and shown in Figs. 3 (j), no excessive chattering
as qd 1  0.1sin(0.8 t ) ( rad ) and qd 2  0.1sin(0.8 t ) (rad ) .
appears in the control inputs.
The designed controllers are implemented using the Matlab
TABLE II RTI system via a MF624 board and the sampling time is
RMS TRACKING ERROR OF THE XY TABLE SYSTEM selected as 0.001 sec.
Axis WO-FC WO-C Proposed
Fig. 5 shows the simulated (Figs. 5 (a)-(f)) and experimental
X-axis 0.128 mm (100 %) 0.094 mm (73 %) 0.043 mm (34%) (Figs. 5 (g)-(n)) results for the manipulator system. The joint
Y-axis 0.152 mm (100 %) 0.087 mm (57 %) 0.037 mm (24%)
position tracking errors for the variations of the dead-band in
the control input are shown in Figs. 5 (a)-(f) according to the
B. Simulation and Experiment for the Articulated joint position sine-wave commands. In the proposed scheme,
Manipulator System the tracking errors remain within the prescribed bound and, and
As a second application example, the articulated manipulator the constraint performance is satisfied in spite of the variations
of the dead-band. On the other hand, the tracking errors of the
shown in Fig. 4 is selected. Among the four links of the robot
WO-FC and the WO-C systems violate the prescribed error
manipulator, only two links (upper arm = link1, forearm=link2)
bounds and the tracking performances of both schemes are
were selected and the manipulator parameters are presented in
sensitive to the variations of the dead-band. In the experimental
Table III. From (1), the dynamic parameters for two degree-of results shown in Figs 5. (g)-(j), the proposed scheme satisfies
- freedom links of the manipulator are described as M (q )q  the prescribed tracking performance, while other systems do
C(q, q)q  G(q)  Fu  D(u) , where Fu contains the unknown not. Similarly to the XY table system, the proposed control
joint friction and the external disturbance. The prescribed scheme has robustness to deadzone as shown in Figs. 5 (k) and
(l), where the tracking errors remains within the constraint
constraint functions are selected as 1  (0.3 0.006)e 5t boundary regardless of deadzone compensation via the inverse
0.006(rad) and 2  (0.3  0.005)e5t  0.005(rad ) . deadzone method. Figs. 5 (m) and (n) show the control inputs
with no excessive chattering problem.
Next, the experiment for the step input command was
executed in order to ascertain finite-time convergence
performance of the WO-C and the proposed control scheme.
The command inputs are selected as qd1  0.2  0.1e3t  0.05e6t
( rad ) and qd 2  0.1 0.06e3t  0.015e12t ( rad ) with the initial
condition q1 (0)  0.18( rad ) and q2 (0)  0.14 (rad ) . The
Fig. 4. Photograph and diagram of the articulated manipulator control
system. constraint functions were selected as 1 (t )  (0.3  0.0005)e 5t
TABLE III 0.0005(rad ) and 2 (t )  (0.3  0.0005)e 5t 0.0005(rad ) .
MANIPULATOR PARAMETERS Figs. 6 (a) and (b) show the tracking output of each control
Symbol Parameter Value system, where the WO-C and the proposed systems show faster
response than the WO-FC system because of finite time control.
m1, m2 mass of link1 and 2 12.1kg , 3.59kg However, the WO-C system still violates the constraint bound,
L1, L2 mass of link1 and 2 0.3m, 0.41m but the proposed control scheme satisfies the constraint
ni gear ratio of reduction gear 65.5 conditions as shown in Figs. 6 (c) and (d). The RMS error sizes
Rmi resistance of motor 0.8294  of the WO-C and proposed systems are small, reaching a
torque constant 0.0182 Nm/A
maximum of 17% and 29% that of the WO-FC system because
kti
they use finite-time and constraint control implemented in the
kbi back emf constant 0.0182 V/rad/sec
controller.

Similarly to the case of the XY table system, three controllers V. CONCLUSION


are used to evaluate the proposed control scheme, namely, the A robust adaptive control system that satisfies the prescribed
WO-FC, the WO-C, and the proposed controller (Proposed). performance, a fast convergence time, and robustness to
The parameters of the control and adaptive laws given are input nonlinearity has been developed for the Lagrangian
11  80 , 12  30 , 21  60 , 22  20 , k1  80 , k2  60 , mechanical system without using FLS or NN approximations,
or identifying the unknown system parameters without/with

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 9

unknown deadzone parameters. A new constrained filtered


tracking error surface is proposed to ensure that the given
tracking error constraint conditions are met. The proposed
control scheme satisfies the prescribed tracking performance
without requiring any intelligent approximation techniques or
identification of unknown dynamics as well as deadzone input
nonlinearity. In addition, a finite-time control term is
introduced to provide fast error convergence. Therefore, the
designed controller has a simple structure and can be
conveniently applied to positioning control for the Lagrangian
systems. Simulated and experimental demonstrations of the XY
table and articulated manipulator systems with deadzone in the
control input verify the satisfactory prescribed position
tracking performance of the proposed control scheme.

Fig. 5. Position tracking results for the articulated manipulator system


for the sine-wave command. 1) Simulation for the variations of the
dead-band: (a) Tracking error of WO-FC scheme in the link 1. (b)
Tracking error of WO-C scheme in the link 1. (c) Tracking error of
Proposed scheme in the link 1. (d) Tracking error of WO-FC scheme in
the link 2. (e) Tracking error of WO-C scheme in the link 2. (f) Tracking
error of Proposed scheme in the link 2. 2) Experiment: (g) Tracking
errors of WO-FC, WO-C, and Proposed schemes in link1. (h) Tracking
errors of WO-FC, WO-C, and Proposed schemes in link2. (i) Tracking
errors in link 1 using Proposed scheme with(W)/without(WO) deadzone
compensation. (j) Tracking errors in link 2 of Proposed scheme
with(W)/without(WO) deadzone compensation. (k) Control inputs of
WO-FC, WO-C, and Proposed schemes in link 1. (l) Control inputs of
WO-FC, WO-C, and Proposed schemes in link 2.

TABLE IV
RMS TRACKING ERROR FOR SINE-WAVE COMMAND OF THE ROBOT SYSTEM
Link WO-FC WO-C Proposed
Link1 0.012 rad (100 %) 0.006 rad (50 %) 0.002 rad (17%)
Link2 0.013 rad (100 %) 0.008 rad (86 %) 0.003 rad (29%)

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2016.2613990, IEEE
Transactions on Industrial Electronics

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 10

[11] C. P. Benchlioulis and G. A. Rovithakis, “Robust adaptive control of


feedback linearizable MIMO nonlinear systems with prescribed
performance,” IEEE Trans. Auto. Control., vol. 53, no. 9, pp. 2090-2099.
2008.
[12] C. P. Benchlioulis and G. A. Rovithakis, “Adaptive control with
guaranteed transient and steady state tracking error bounds for strict
feedback systems,” Automatica, vol. 45, pp. 532-538, 2009.
[13] W. Wang and C. Wen, “Adaptive actuator failure compensation control of
uncertain nonlinear systems with guaranteed transient performance,”
Automatica, vol. 46, pp. 2082-2091, 2009.
[14] J. Na, Q. Chen, X. Ren, and Y. Guo, “Adaptive prescribed performance
motion control of servo mechanisms with friction compensation,” IEEE
Trans. Ind. Electron., vol. 61, no. 1, Jan. 2014.
[15] S. I. Han and J. M. Lee, “Partial tracking error constrained fuzzy dynamic
surface control for a strict feedback nonlinear dynamic system,” IEEE
Trans. Fuzzy Sys., vol. 22, no. 5, pp. 1049-1061, Oct. 2014.
[16] A. Theodorakopoulos, and G. A. Rovithakis, “Prescribed performance
control of strict feedback systems with deadzone input nonlinearity,” 52nd
IEEE Confer. On Decision and Control, pp. 1774-1779, Dec. 2013.
[17] A. Theodorakopoulos, and G. A. Rovithakis, “Guaranteeing preselected
tracking quality for uncertain strict-feedback with deadzone input
nonlinearity and disturbance via low-complexity control,” Automatica,
vol. 54, pp. 135-145, 2015.
[18] S. Y. Oh and D. J. Park, “Design of new adaptive fuzzy logic controller
for nonlinear plants with unknown or time-varying deadzones,” IEEE
Trans. Fuzzy Sys., vol. 6, no. 4, pp. 482-491, 1998.
[19] R. R. Selmic and F. L. Lewis, “Deadzone compensation in motion control
systems using neural networks,” IEEE Trans. Auto. Control., vol. 45, no.
4, pp. 602-613, 2000.
[20] C. Hu, B. Yao, and Q. Wang, “Adaptive robust precision motion control
of systems with unknown deadzones: a case study with comparative
experiments,” IEEE Trans. Ind. Electron., vol. 58, no. 6, May 2011.
Fig. 6. Position tracking results of the articulated manipulator system for [21] S. S. Ge, and C. Wang, “Adaptive neural control of uncertain MIMO
the step command. (a) Tracking output in link1. (b) Tracking output in nonlinear system,” IEEE Trans. Neural Net., vol. 15, no. 3, May 2004.
link2. (c) Tracking errors in link1. (d) Tracking errors in link2. [22] J. J. Slotine, and W. Li, Applied nonlinear control, New Jersey: Prentice
Hall, 1991.
[23] S. Yu, X. Yu, B. Shirinzadeh, and Z. Man, “Continuous finite-time
REFERENCES control for robotic manipulators with terminal sliding mode,” Automatica,
[1] Y. C. Lu, and J. S. Chen, “A self-organizing fuzzy sliding mode controller vol. 41, pp. 1957–1964, 2005.
design for a class of nonlinear servo systems,” IEEE Trans. Ind. Electron., [24] A. M. Zou, K. D. Kumar, Z. G. Hou, and X. Liu, “Finite-time attitude
vol. 41, no. 5, pp. 492-502, Oct. 1994. tracking control for spacecraft sliding mode and Chebyshev neural
[2] T, Das and I. N. Kar, “Design and implementation of an adaptive fuzzy network,” IEEE Trans. System, Man, and Cyber. Part B: Cyber., vol. 41,
logic-based controller for wheeled mobile robots,” IEEE Trans. Control. no. 4, pp. 950-963, Aug. 2011.
System Tech., vol. 14, no. 3, pp. 501-510, May 2006.
[3] R. Precup, R. C. David, E. M. Petriu, S. Preitl, and M. B. Radac, “Fuzzy
control systems with reduced parametric sensitivity based on simulated Seong Ik Han (M’12) received the B.S. and M.S.
annealing,” IEEE Trans. Ind. Electron., vol. 59, no. 8, pp. 3049-3061, degrees in mechanical engineering from Pusan
Aug. 2012. National University, Busan, Korea, in 1987 and 1989,
[4] F. F. M. El-Sousy, “Intelligent optimal recurrent wavelet Elman neural respectively, and the Ph. D. in mechanical design
network control system for permanent magnetic synchronous motor servo engineering from Pusan National University, Busan,
drive,” IEEE Trans. Indust. Inform., vol. 9, no. 4, pp. 1986-2003, Nov. in 1995. From 1995 to 2009, he was an associate
2013. professor of electrical automation of Suncheon First
[5] N. Nikdel, P. Nikdel, M. A. Badamchizadeh, and I. Hassanzadeh, “Using College, Korea. Now he is with the Department of
neural network model predictive control for controlling shape memory electronic engineering, Pusan National University,
alloy-based manipulator,” IEEE Trans. Ind. Electron., vol. 61, no. 3, pp. Korea. His research interests include intelligent
1394-1401, Mar. 2014. control, nonlinear control, robotic control, vehicle
[6] R. J. Lian, “Adaptive self-organizing fuzzy sliding-mode radial basis system control, and steel process control.
-function neural-network controller for robotic systems,” IEEE Trans. Ind.
Electron., vol. 61, no. 3, pp. 1493-1503, Mar. 2014. Jang Myung Lee (SM’03) received the B.S. and M.S.
[7] Z. P.Wang, S. S. Ge, and T. H. Lee, “Robust motion/force control of degree in electronic engineering from Seoul National
uncertain holonomic/nonholonomic mechanical systems,” IEEE Trans. University, Seoul, Korea, in 1980 and 1982,
Mechatronics, vol. 9, no. 1, pp. 118-123, Mar. 2004. respectively, and the Ph. D. degree in computer
[8] Z. Li, S. S. Ge, and A. Ming, “Adaptive robust motion/force control of engineering from the University of Southern
holonomic-constrained nonholonomic mobile manipulators,” IEEE Trans. California, Los Angeles, in 1990.
Sys. Man and Cyber.-Part B: Cyber., vol. 37, no. 3, June pp. 607-616, Since 1992, he has been a Professor with Pusan
2007. National University, Busan, Korea. He was the Leader
[9] Z. Li, P. Y. Tao, S. S. Ge, M. Adams, and W. S. Wijesoma, “Robust of the “Brain Korea 21 Project” of Pusan National
adaptive control of cooperating mobile manipulators with relative motion,” University. His research interests include intelligent
IEEE Trans. Sys. Man and Cyber.-Part B: Cyber., vol. 39, no. 1, pp. robotics, advanced control algorithm, and specialized
103-116, Feb. 2009. environment navigation/localization. Prof. Lee was the former president of the
[10] V. Andaluz, F. Roberti, J. M. Toibero, and R. Carelli, “Adaptive unified Korean Robotics Society.
motion control of mobile manipulators,” Control Engineering Prac., vol.
20, pp. 1337-1352, 2012.

0278-0046 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

S-ar putea să vă placă și