Sunteți pe pagina 1din 11

ARTICLE IN PRESS

Control Engineering Practice 13 (2005) 973983


www.elsevier.com/locate/conengprac

Estimation of backlash phenomenon in the


electromechanical actuator
R. Merzouki, J.C. Cadiou
Laboratoire de Robotique de Versailles 10-12, avenue de lEurope, 78140 Velizy, France
Received 4 February 2004; accepted 6 October 2004
Available online 2 December 2004

Abstract

In this paper an adaptive control is developed, based on an estimation of the disturbing torque transmitted inside the dead zone of
an electromechanical system. Disturbing backlash torque is modelled by a continuous and derivable mathematical function
describing an opposite of sigmoid function. Then, nonlinear observer of dead zone amplitude is synthesized. Simulation and
experimental tests applied to an electromechanical actuator are given to support the theoretical demonstrations.
r 2004 Elsevier Ltd. All rights reserved.

Keywords: Backlash effects; Disturbing torque; Electromechanical actuator; Dead zone; Adaptive control

Notation y_ s output reducer velocity (rad/s)


y s output reducer acceleration rad=s2
j0 dead zone amplitude (rad)
w nonlinear transmitted torque (N m)
w~ nonlinear transmitted torque error (N m)
Jm motor inertia N m2 1. Introduction
Js load inertia N m2
K elasticity constant (N m/rad) The presence of mechanical imperfections in robot
N0 reducer constant systems makes it difcult to control them with a high
fs viscous friction coefcient (N m s/rad) degree of accuracy in a closed loop. Among these
ee input position error (rad) imperfections, we nd the backlash phenomenon,
e_e input velocity error (rad/s) represented by a dead zone between the input point
es output position error (rad) system and its output, for example: mechanical gears.
e_s output velocity error (rad/s) Knowing that backlash of weak amplitudes is always
ye input reducer position (rad) present in the electromechanical systems, after long time
y_ e input reducer velocity (rad/s) wear of these systems, the backlash amplitude becomes
y e input reducer acceleration rad=s2 increasingly important, creating a difference between the
ys output reducer position (rad) input and the output of the system.
Mechanical and control solutions are given to reduce
the backlash effect. The mechanical solution consists in
Corresponding author. Laboratoire dAutomatique et Genie
modifying the structure of the mechanical system, which
Informatique et Signal (LAGIS), UMR CNRS 8146, Ecole poly- is expensive. The control solution consists in minimizing
technique de Lille, Avenue Paul Langevin, 59655 Villeneuve dAscq,
France. Tel.: +33 3 28 76 74 86; fax: +33 3 28 76 73 0. the effect of backlash by using a control law that takes
E-mail addresses: merzouki@robot.uvsq.fr, Rochdi.Merzouki@ into account the dead zone disturbance. The latter
polytech-lille.fr (R. Merzouki). solution is less expensive.

0967-0661/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.conengprac.2004.10.016
ARTICLE IN PRESS
974 R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983

Some work has already been carried out on compar- In the second category we nd the works of
able systems. This work can be arranged in three Stepanenko and Sankar (1986) which is based on studies
main categories: those where the main interest is the in relations with the model containing a shock absorber
control of similar systems, those specialized rather in the with impacts. Bapat, Popplewell, and Mclachlan (1983);
dynamic analysis of the system and those where the Popplewell, Bapat, and McLachlan (1983); Li, Rand,
development is mainly dedicated to the simulation of and Moon (1990); Shaw and Shaw (1989) and Shaw and
such systems. Rand (1989) studied the dynamic response of a rigid
For the rst category, the main idea is to conceive body in collisions. They showed the existence of
strategies of control improving the performance of complexes dynamics including various types of periodic
systems subject to shocks with its workspace or to use trajectories and chaos. Pfeiffer and Glocker (1996)
characteristic collisions in order to use them for the carried out a detailed analysis of a rigid body model.
control. The second category studies the dynamic In the third category, the work of Thompson and
equations representing the phenomenon of impact. For Ghaffari (1982) could be also classied.
the third category, the main idea is to study the In this paper, the study presented is based on the
numerical algorithms in order to integrate the systems adaptive compensation of the backlash effect in an
subjected to unilateral constraints. electromechanical actuator, rich in mechanical imper-
In the rst category, we nd mainly work from Tao fections like backlash, viscous friction and elasticity on
and Kokotovic(1993,1995a,b), where they modelled the transmission links. To do so, we propose a mathematical
backlash like a function of hysteresis between the input model of backlash Merzouki, Cadiou, and MSirdi
and output positions of the system. In the latter, (2000, 2004), continuous and derivable. The backlash
collision effects are not considered. The control pro- model describes a null torque inside a dead zone with an
posed consists of a discontinuous signal which remains estimated amplitude and shows a pseudo linearity of the
in contact with the different parts of the system transmitted torque according to the position, outside the
displaying backlash. This idealized system is possible dead zone. This modeling will enable us to establish an
only for systems without inertia. The selected model estimation of disturbing torque after observing the
eliminates all the nonlinear effects produced by colli- dead zone amplitude. The electromechanical system will
sions. Brandenburg and Schafer (1988) studied the thus be controlled due to an adaptive control containing
inuence of the partial compensation of simultaneously two closed loops of PD controller in more of the
acting backlash and Coulomb friction in a two-mass disturbing torque estimator. This work could be located
system. In Azenha and Tenreiro Machado (1996), the in the rst category where smooth collision effects are
authors studied the control of a system with dynamic considered.
backlash. In the model proposed, the effect of the This paper is divided into six sections including the
impact of the gears on their dynamics is taken as a introduction, followed by a description of the electro-
disturbance. Only a numerical simulation is given in this mechanical test bench. Section 3 describes the backlash
work. In the case of Indri and Tornabe (1997), the modeling and Section 4 presents the developed estimator
authors consider the gears with backlash and elasticity for the disturbing backlash torque and the dead
in the articulations. In order to control the system they zone amplitude observer with the proposed adaptive
propose, they studied the behavior of two gears when a control. Sections 5 and 6 present consecutively
PD controller is applied. The properties of the solutions simulation and experimental results, and a comparison
in a closed loop are obtained and the stability of the is given in Section 7. Finally, the conclusion is given in
strategy is proven. A system with completely inelastic Section 8.
impacts is studied by Chalhoub and Zhang (1996). The
technique of control consists in accelerating the con-
trolled gears to decrease the period of disengagement 2. Description of the test bench
when the second gears are not controllable. An
experimental validation is given on a exible beam. The test bench presented in Fig. 1 represents an
Yeh et al. proposed a model compliant to describe the electromechanical system made up of a motor reducer
backlash phenomenon from where an adaptive non- involving an external load. The motor part is controlled
linear controller is designed and a semi-global stability is by a DC motor delivering a relatively important mass
given. Also, among control solutions, we may quote the torque. On the reducer part, the reduction constant is
work of Recker, Kokotovic, Rhode, and Winkelman about 41 in the input and 13 at the output. The test bench is
(1991), where they worked on the adaptive control of made up of mechanical imperfections a mixture of
system with backlash. The idea was to propose an friction, backlash and elasticity. The advantage of this
inverse model of the backlash and to include it in the reducer is that one can vary the amplitudes of the
control scheme in order to dene the desired transmitted mechanical imperfections. Coulomb friction is repre-
torque to mitigate the backlash effect. sented by a contact of different materials with various
ARTICLE IN PRESS
R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983 975

3. Backlash modeling
Electrical Dead Zone
Actuator Mechanism
In this section, a description of a backlash motion is
presented. The known backlash torque model is non-
continuous and nonderivable for Figs. 3 and 4b,
Input transmitted via a dead zone and a exible link of the
Output
mechanical system as in Fig. 4a. In this model, coupling
is null inside the dead and linear outside this zone. The
choice of this model causes a problem during the control
around the noncontinuous areas.
The backlash mechanism could be described as in Fig.
4a. The body1 tries to transmit motion to body2 via a
dead zone of amplitude 2:j 0 : The transmission will be
correct when the two bodies are in contact, in this case
the two positions are identical. Out of the contact, the
transmission will be delayed with the presence of the
dead zone where the relation between the bodies
Incremental positions describes a hysteresis behavior.
Encoders Reducer Part In order to do this, smooth and continuous model of
transmitted torque is proposed, and is given in Fig. 5.
Fig. 1. Test bench. This latter includes a sigmoid function expressed as
 
1  eg:z
C K z  4j 0 ; (1)
1 eg:z
where C is the approximate transmitted torque, z denes
Motorr
Moto e the difference of positions between the input and output
of the reducer part, K represents the constant of rigidity,
j 0 is the dead zone amplitude and g is a constant.
We can decompose the last expression of the
s
transmitted torque C to two parts as follows:
Load
C C0 w (2)

e
Fig. 2. Electromechanical system including mechanical inperfections. Motor

Load s
rigidities. Viscous friction depends on the viscosity of
the lubricant contained between surfaces in contact,
while the backlash is represented by two independent
mechanical parts, whose transmission is carried out via a Fig. 3. Transmitted torque via dead zone and exible links in a motor
dead zone varying between 01and 241. A spring system is reducer system.
placed between the two mechanical parts in order to
make smooth the transmission.
On this test bench, we can measure the input and the Body 1 C
output positions of the reducer axis by using two
incremental encoders as in Fig. 1.
We can describe the three mechanical imperfections velocity j position
0 Z
contained in the test bench by Fig. 2, where the relative Body 2
position of the load will depend on friction between the 2j0
(a) (b)
gears in contact, as well as exible transmission through
a dead zone. Fig. 4. (a) Backlash mechanism, (b) Non continuous backlash model.
ARTICLE IN PRESS
976 R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983

the reducer parts, then an adaptive control based on the


estimation of the disturbing torque and its compensa-
tion is presented.

4.1. Modeling of the electromechanical actuator

Let us consider that static friction is disregarded, then


the mechanical model of the test bench including the
backlash is described by the following system:
J s y s f s y_ s N 0 C;
J m y e f m y_ e C U; 6
where J s ; J m ; f s ; f m are successively the inertias and the
viscous frictions of the reducer and motor parts which
are identied experimentally .
y s ; y e ; y_ s ; y_ e are successively the accelerations and the
Fig. 5. Approximation of transmitted torques. velocities of the reducer and motor parts which are
deduced by derivation of the measured input and output
reducer positions ys and ye :
U is the control torque, C represents the transmitted
torque described in (2).
with C 0 the linear transmitted torque via a exible link Now, the system (6) could be expressed by the
and is written as following expression:

C 0 Kz (3) J s y s f s y_ s KN 0 ye  KN 20 ys N 0 w;
and w is the disturbing and nonlinear transmitted torque J m y e f y_ e U  Kye KN 0 ys  w:
m 7
as expressed by Let us put
g:z
1e
w 4Kj 0 ; es ys  yds ;
1 eg:z
dj 0 ee ye  yde ;
0: 4 z Kye  KN 0 ys ;
dt
Since the real backlash amplitude j 0 is equal to a e z z  zd ;
constant, its variation is reduced to zero. w~ w  w;^ 8
The representation of w depends on the value of the
where yds and yde are successively the desired output and
constant g which depends on the amplitude j 0 : To show
input reducer positions with N 0 represents the reducer
this, it can be observed that expression (4) corresponds
constant. z is the difference between the input and
to the best approximation of the transmitted torque to
output positions and zd denotes its desired value. w^ is
its noncontinuous conguration as shown in Fig. 5.
the estimated disturbing torque with w~ its estimation
From the simulation and experimental tests, we
obtained the best decreasing of w inside j 0 ; j 0  after error.
supposing Therefore, the system (7) could be expressed as
follows:
1
g
2j 0
: (5) J s y s f s y_ s KN 0 ye  KN 20 ys N 0 w;

For the next sections, the parameter g is calculated in off J m y e f y_ e N 1 J s y s f y_ s U:


m 0 s 9
line, with an initial value of j 0 ; then, the estimation of w
is given after observing the evolution of the amplitude j 0
on line. 4.2. The control of the electromechanical actuator with
backlash

4. Adaptive control of the electromechanical actuator The control of the electromechanical actuator de-
scribed in Section 2 is dened by the following steps:
In this section, a mathematical model of the electro-  linearization of the output reducer part equation;
mechanical actuator is given, including the motor and  linearization of the input reducer part equation;
ARTICLE IN PRESS
R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983 977

Fig. 6. Control scheme of the bench test including backlash.

 adaptive compensation of the disturbing backlash So, the second equation of system (9) will become
torque using its corresponding estimation with the
Jm f
backlash amplitude observer. z m z_ J m N 0 J s N 1
0 ys
K K
The control diagram of the adaptive control is given in f m N 0 f s N 1 _
0 ys U: 14
Fig. 6.
This control diagram is represented by two closed Thus, we dene a control law U for the global system
loops. The rst is based on a PD controller, which expressed as follows:
deduces the desired input position of the reducer. The  
Jm d f
second, based also on a PD controller, is joined the U z m z_d
K K
estimator of the disturbing torque in order to deduce the
d d
control signal. J m N 0 J s N 1 1 _
0 ys f m N 0 f s N 0 ys ; 15
 k1 ez  K D1 e_e  K P1 ee u; 16

4.2.1. Linearization of the output reducer part equation where K P1 and K D1 are the PD controller constants C 1
In order to linearize the rst equation of system (9), of Fig. 6.
we replace the z and w values by their estimated values. The desired input yde represents the output PD con-
So the following equation is obtained: troller C 2 of Fig. 6, dened as follows:

yde K D2 e_s K P2 es (17)


J s y s f s y_ s N 0 ez zd w~ w:
^ (10)
with K P2 and K D2 representing the controller C 2
So the desired value zd of z is chosen to be equal to
constants of Fig. 6.
d
_ d u represents a control signal dened in function of the
zd N 1 ^
0 J s ys f s ys  w: (11) disturbing torque estimation error w~ and position errors.
If we replace expression (11) in (10), we will obtain Then after replacing expression of U in Eq. (14), we
obtain
J s es f s e_ s N 0 ez w
~ (12)    
Jm f m K D1 K P1
ez e_ z k1 ez
which is the equation deduced after linearization of the K K K K
reducer part model. J m N 0 J s N 1 es f m N 0 f s N 1
0 0 K D1 N 0 _
es
K P1 N 0 es u 18
4.2.2. Linearization of the input reducer part equation
Now, to linearize the second equation of system (9),
once put
4.2.3. Estimation of disturbing backlash torque
z The model in (4) needs an observer of an unknown
ye N 0 ys ;
K state j 0 in order to compensate the disturbing backlash
z_ torque. The observer of state ^|0 is used to estimate the
y_ e N 0 y_ s ;
K disturbing torque w in w:^
z This observer is chosen under the supposition that the
y e N 0 y s : 13
K model parameters are known, it is given by the following
ARTICLE IN PRESS
978 R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983

expression:    
0 N 0 0
C K P1 ; E ;
1  eg:z K P1 N 0 k1 1
w^ 4K^|0 ; K
1 eg:z      
d^|0 es e_ s es
k2 ; 19  ; _ ;  :
dt ez e_z ez
where K and g are supposed to be known, k2 The control law dened by u is chosen as follows:
k21 k22 40 a constant vector and T es ez : The
Eu E w~ Hs (24)
expression k2  in the observer represents a correction
of the position error. then we dene the Qs a matrix function, by the
following equation:

4.2.4. Stability analysis Qs 4KAs2 Bs C  Hs1 D E; (25)


Knowing that the disturbing backlash torque model where s is the Laplace constant and Hs a state
of (4) is symmetrical around its origin z 0; so we are matrix which is chosen to have a positive real function
studying rst the positive case, then we could generalize Qs:
for the negative one. Now, we divide our study in two Thus, from (23)(25) we obtain the relation between 
parts depending on the input and output positions of the and j 0
system according to the dead zone.
 Qsj~0 : (26)
If we take
4.2.4.1. Outside the dead zone. Let us take the case of a
positive displacement of z with z4j 0 ; so the expression e_ ae bj~0 ;
of w in (4) and w^ in (19) will be reduced to the following  ce 27
representation:
as a state representation of Qs function. So e
w 4Kj 0 ; corresponds to an intermediate variable between the
w^ 4K^|0 : 20 input j~0 and the output : As long as Qs is strictly
positive real and according to the KalmanYakubovitch
The estimation error of the disturbing backlash torque lemma, there exists two matrices P PT 40 and M
w~ w  w^ and the observation error of the dead zone M T 40 so that
~
amplitude ddtj 0 djdt0  d^dt|0 will take the following expres-
sion: aT P Pa M;
Pb cT : 28
w~ 4K j~0 ;
Let us take now
dj~0
k2 : 21 2
dt V eT Pe k1 ~
2 j0 (29)
So, the global system after linearization in (12) and (18) as a Lyapunov function, is taken positive denite and its
is expressed in the following system equations: equilibrium state is e 0 and j~0 0:
We shall now calculate the rst derivation of the
J s es f s e_ s N 0 ez w;
~
    function V
Jm f m K D1 K P1
ez e_z k1 ez ~
K K K K V_ e_T Pe eT Pe_ 2k1 ~ dj 0 ;
2 j0
dt
J m N 0 J s N 1 es f m N 0 f s N 1
0 0 K D1 N 0 _
es
T dj~0 ~
K P1 N 0 es u 22 V_ eT aT  j~0 bT Pe eT Pae  j~0 b 2k1 j ;
2
dt 0
which can be expressed by the following equation: V_ eT Me  2j~0 2k1 ~
2 k 2 j 0 ;

A B_ C Dw~ Eu; (23) V_ eT Mep0:

where So, the V function is decreasing and the equilibrium


    state e 0 and j~0 0 is asymptotically stable and
Js 0 N0 according to (21), we notice that w~ ! 0:
A Jm ; D ;
J m N 0 J s N 1
0 K 0
4.2.4.2. Inside the dead zone. In this case z is limited
!
fs 0 between zero and j 0 positions, Eq. (4) of the backlash
B fm ; torque model could be approached via a Taylor series
f mN 0 f s N 1
0 K D1 N 0 K KKD1 around z 0:
ARTICLE IN PRESS
R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983 979

If the function wz admits, around jz  ajoR of the 5. Simulation results


point a, a development in series according to the powers
of z  a; it will have a form of In the simulation tests, we carry out a follow-up of a

wa null position and a null velocity trajectories, applied to
_
wz wa waz  a z  a2 the test bench of Fig. 1 where the corresponding
2!
constants are given in Table 1. For that, we impose a
wn a
z  an : 30 null value of the output position initialized by value
2! 1.57 rad for 4 s. The value of j 0 given in Table 1
For a 0 the Taylor series called also Maclaurin0 s represents the nominal value taken in the experimental
series. Equality (30) is true if for jz  ajoR the bench test.
remainder of the Taylor series In Fig. 7 we notice the follow-up of the desired output
 X wk a  position signal, before and after the compensation of the
n
Rn z wz  wa z  a ! 0: (31) disturbing backlash torque. The case before compensa-
k!
tion corresponds to the application of only a PD
Using this theorem, the expression of the disturbing controller and the case after compensation means the
torque model of (4), around z 0 is now expressed by application of a PD controller plus a compensation of
the disturbing backlash torque. The convergence to-
w 4Kj 0 z;
wards the zero value is ensured with a transient period
dj 0 of approximately 13 s, longer than of a PD controller
0: 32
dt with a compensation of the disturbing torque estimated
it looks like a decreasing linear function of z as it is at 5.2 s. We also notice that at approximate transitions
shown in Fig. 5. of 0 and 4 s, the signal varies quickly with a relative high
According to the relative observer of (19), we obtain peak before compensation. In Fig. 8, a superposition of
the error signals for output position follow-up is
w^ 4K^|0 z;
presented. We notice the time of convergence of the
d^|0 two signals before and after compensation as well as the
k2  33
dt transient periods which are also described by the
with an estimation and observation errors velocity curves of Fig. 9. Figs. 1012 successively show
the follow-up of the input reducer position before and
w~ 4K j~0 z;
after compensation with the corresponding errors and
dj~0 velocity signals. We notice a faster convergence with
k2  34
dt smooth transitions after compensation of the disturbing
where z corresponds to the position difference between backlash torque. This smooth transition is mainly due to
the input and the output of the reducer part, containing the proposed backlash model which is continuous with
a dead zone area. So its maximum value Z maxz smooth transitions. Fig. 13 shows the hysteresis effect
cons tan t is highly limited by j 0 : In this case between the input and output reducer positions, before
and after adaptive compensation. It is noted that the
~  4KZ j~0 :
wp (35) hysteresis phenomenon thickness is reduced after
Let us take, for demonstration purposes, the maximum compensating the disturbing torque and the two
~ i.e.
value of w; positions meet quickly and converge towards the desired
positions. In Fig. 14, a linearization of the disturbing
w~ w~ max 4KZ j~0 : (36) backlash torque after applying the adaptive compensa-
So, the global system (22) could be expressed by the tion is presented.
following equation:
 Qsj~0 ; (37)
Table 1
where Qs in this case is dened as follows: Simulation model characteristics

Qs 4KZAs2 Bs C  Hs1 D E: (38) Controller Parameters Sys. Parameters


Let us take V as a Lyapunov function dened above, K P1 1 J s 7:5
positive denite and its equilibrium state is e 0 and K P2 10 f s 16
j~0 0: K D1 0.01 N0 3
So, calculating its derivative as it is detailed in the K D2 10 J m 9:72  104
preceding section, the function V is decreasing and the k1 1 f m 4:3  104
k21 0.01 j 0 0:16
equilibrium state e 0 and j~0 0 is asymptotically
k22 0.5 K1
stable and according to (36) we deduce that w~ ! 0:
ARTICLE IN PRESS
980 R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983

Fig. 7. Desired and real output signals before and after compensation. Fig. 10. Desired and real input signals before and after compensation.

Fig. 8. Output position error signals before and after compensation. Fig. 11. Input position error signals before and after compensation.

Fig. 9. Output velocity signals before and after compensation. Fig. 12. Input velocity signals before and after compensation.
ARTICLE IN PRESS
R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983 981

Table 2
Experimental model characteristics

Controller Parameters Sys. Parameters

K P1 1 N0 3
K P2 10 j 0 0:16
K D1 0.01 K1
K D2 10
k1 1
k21 0.01
k22 0.5

Fig. 13. Backlash hysterisis behavior before and after compensation.

Fig. 15. Output position before and after adaptive compensation.

Fig. 14. Linearization of the distributing torque after compensation.

6. Experimental results

The experimental constants of test bench Fig. 1 are


given in Table 2.
In these tests, the electromechanical actuator is
required to move from the initial static output position
ys 0 p=2 rad; and output velocity y_ s 0 0 rad=s to
the origin ys 0 0 rad; y_ s 0 0 rad=s:
Fig. 16. Output velocity before and after adaptive compensation.
Fig. 15 represents the tracking of the desired output
position before (i.e. when the control is made by only a
PD controller) and after applying the adaptive compen-
sation. The static position error is about 0.32 rad before compensation, undesired oscillations around y_ s 0
compensation and it is eliminated after adding the 0 rad=s are present and represent the nonlinearities
estimator-compensator of the disturbing backlash tor- effects. These imperfections are compensated after
que. Fig. 16 shows the output velocity signals, before applying the adaptive control of the disturbing backlash
and after the adaptive compensation. So, before torque. Finally, the control signals before and after
ARTICLE IN PRESS
982 R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983

8. Conclusion

The presence of backlash in mechanical system makes


it difcult to control it with high accuracy. So, we could
reduce the effects of the latter imperfection by estimat-
ing the necessary disturbing torque inside the dead zone,
then we can construct an adaptive control, which allows
to compensate the backlash effects by compensating this
estimated disturbing backlash torque. For that we have
chosen a nonlinear and continuous mathematical model
for the disturbing torque with an adaptive observer of
the dead zone amplitude. A good choice of adaptive
control parameters make an asymptotic convergence of
global system to equilibrium state, as it has been shown
in the simulation and experimental results.

Fig. 17. Control signal before and after adaptive compensation. References

Azenha, A., & Tenreiro Machado, J. A., 1996. Variable structure


control of systems with nonlinear friction and dynamic backlash.
compensation of the backlash effects are shown in 2nd Porteguess conference on automatic control (vol. 1)
Fig. 17. After compensation, the control signal is (pp. 237242), Porto, Portugal, September 1996.
cleaner than its equivalent before compensation due to Bapat, C. N., Popplewell, N., & Mclachlan, K. (1983). Stable periodic
motions of an impact-pair. Journal of Sound and Vibration, 8e`(1),
the adaptive compensation of the backlash disturbance 1940.
effects. According to the experimental results, we Brandenburg, G., & Schafer, U. (1988). Inuence and partial
notice that the position tracking is ensured with a time compensation of simultaneously acting backlash and Coulomb
limit after applying the adaptive backlash control. friction in a speed and position controlled elastic two-mass system.
The signal control of Fig. 17 converge to its ICED 88, proceedings of international conference on electrical drives.
(pp. 112), Romania.
minimal value, compared with the situation before Chalhoub, N. G., & Zhang, X. (1996). Modeling and control of
compensation. backlash in the drive mechanism of a radially rotating beam.
Journal of Dynamic Systems, Measurement and Control, 118,
158161.
7. Comparison of simulation and experimental results Indri, M., & Tornabe`, A. (1997). Application of a PD controller on
two mating gears with elasticity and backlash. Conference on
decision and control. San Diego, USA, December 1997.
After comparing the corresponding results of simula- Li, G. X., Rand, R. H., & Moon, F. C. (1990). Bifurcations and chaos
tion and experimentation, we can make the following in a forced zero-stiffness impact oscillator. Journal of Nonlinear
comments: Mechanics, 25(4), 417432.
 Convergence to the desired output position is ensured Merzouki, R., Cadiou, J. C., & MSirdi, N. K. (2000). Compensation
of backlash effects in an electrical actuator. The IASTED
by a PD controller before the compensation with a international conference on intelligent systems and control
low transient period in the case of simulation as (IASTED2000), Hawaii, USA August 1416.
shown in Fig. 7, but it is not ensured in the case of the Merzouki, R., Cadiou, J. C., & MSirdi, N. K. (2004). Compensation
real system in Fig. 15, where a static error is present. of friction and backlash effects in an electrical actuator. Journal of
Systems and Control Engineering, 218(12), 7584.
So, this difference is mainly due to the dynamic system
Pfeiffer, F., & Glocker, C. (1996). Multibody dynamics with unilateral
model and according to the tracking, we have to contacts. New York: Wiley.
include the equivalent of an integrator to minimize Popplewell, N., Bapat, C. N., & McLachlan, K. (1983). Stable periodic
this static error. This is what the adaptive compensa- vibroimpacts of an oscillator. Journal of Sound and Vibration,
tion sets out to verify. 87(1), 4159.
 After applying the adaptive compensation, conver- Recker, D. A., Kokotovic, P. V., Rhode, D., & Winkelman, J. (1991).
Adaptive nonlinear control of system containing a dead zone.
gence to the desired output position is faster in the Proceedings of the 30th conference on decision and control (vol. 3)
case of experimentation with clean tracking as is (pp. 21112115) Brighton, England, December 1991.
shown in Figs. 15 and 16. For the simulation case, Shaw, J., & Rand, S. R. H. (1989). The transition to chaos in a simple
convergence is made after a small perturbed transition mechanical system. Journal of Nonlinear Mechanics, 24(1), 4156.
shown in Figs. 7 and 9. We can justify this perturbed Shaw, J., & Shaw, S. W. (1989). The onset of chaos in a two-degree-of-
freedom impacting system. ASME Journal of Applied Mechanics,
transition by considering the mechanical imperfec- 56, 168174.
tions present in the simulation system model such us Stepanenko, Y., & Sankar, T. S. (1986). Vibro-impact analysis of
friction and exibility. control systems with mechanical clearance and its application to
ARTICLE IN PRESS
R. Merzouki, J.C. Cadiou / Control Engineering Practice 13 (2005) 973983 983

robotics actuators. Journal of Dynamic Systems, Measurement and Tao, G., & Kokotovic, P. V. (1995b). Contiuous-time adaptive control
Control, 108, 916. of systems with unknown backlash. IEEE Transactions on
Tao, G., & Kokotovic, P. V. (1993). Adaptive control of systems with Automatic Control, 40(2), 10831087.
backlash. Automatica, 29(2), 323335. Thompson, J. M. T., & Ghaffari, R. (1982). Chaos after period-
Tao, G., & Kokotovic, P. V. (1995a). Adaptive control of plants with doubling bifurcations in the resonance of an impact oscillator.
unknown hysteresis. IEEE Transactions on Automatic Control, Physical Letters, 91A(1), 58.
40(2), 200212.

S-ar putea să vă placă și