Você está na página 1de 11

International Journal of Non-Linear Mechanics 42 (2007) 459469

www.elsevier.com/locate/nlm
Non-linear control of friction-induced self-excited vibration
S. Chatterjee

Department of Mechanical Engineering, Bengal Engineering and Science University, Shibpur, P.O. Botanic Garden, Howrah 711 103, West Bengal, India
Received 21 September 2006; received in revised form 9 January 2007; accepted 15 January 2007
Abstract
The present paper introduces a new method of controlling friction-driven self-excited vibration. The control law is primarily derived using
the Lyapunovs second method. A single degree-of-freedom oscillator on a moving belt represents the primary model of the system. The control
action is achieved by modulating the normal load at the frictional interface based on the state of the oscillatory system. The basic mechanism
of the control action utilises subcritical Hopf bifurcation of the equilibrium followed by cyclic-fold bifurcation (of limit cycle oscillations) to
globally stabilise the equilibrium. The basic mechanism is qualitatively independent of the exact model of friction. Different models of friction,
like, algebraic model, LuGre model and switch model with time-dependent static friction are considered to substantiate the above claim. An
approximate method for estimating the critical value of the control parameter that ensures global stability of the equilibrium is also proposed.
2007 Elsevier Ltd. All rights reserved.
Keywords: Friction; Self-excited vibration; Hopf bifurcation; Limit cycle
1. Introduction
Friction-induced self-excited vibration is very common
in mechanical and electromechanical systems. One of the
major causes of friction-induced vibration is identied with
the velocity-weakening characteristics of friction, which
is also known as the Stribeck effect (this is somewhat
confusing, because Stribeck effect is originally associated with
hydrodynamic lubrication). Because of the drooping nature of
friction force in the low velocity range, the formal dissipative
nature of friction is lost and energy is generated inside the
system leading to unstable equilibrium. However, at relatively
higher velocity range, friction force either rises or remains con-
stant with velocity of sliding. Therefore, in this regime of slid-
ing, friction force regains its usual dissipative characteristics.
Thus, in the steady state, the system is thrown into a bounded
periodic vibration, the so-called limit cycle oscillation.
In most systems, such vibrations are highly undesirable. For
example, self-excited vibration in mechanical brakes causes
noise related discomfort. Another example is the controlled
positioning systems that have now become an integral part of a

Tel.: +91 033 2668 4561.


E-mail address: shy@mech.becs.ac.in.
0020-7462/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijnonlinmec.2007.01.015
large number of industrial production systems as well as con-
sumer electronics product. Friction often causes high level of
undesirable positioning inaccuracies by inducing self-excited
oscillations around the desired position. Thus, it is not surpris-
ing that researchers, over the past few years, have put seri-
ous efforts towards understanding the phenomena and nding
means of controlling such unwanted oscillations.
Literature on controlling friction-induced vibration is vast.
Heckl and Abrahams [1] propose an active control scheme
for quenching friction-induced self-excited oscillation. In their
scheme, tangential (in the direction of slipping) control force
is generated by driving an actuator with phase-shifted, ampli-
ed velocity signal sensed by a transducer. Elmer [2] proposes
time-delayed state feedback method. Unlike the method in [1],
Elmers scheme employs normal load modulation based on the
time-delayed state feedback. Das and Mallik [3] consider PD
control using time-delayed velocity and displacement feed-
back. Here, control force acts in the slipping direction. Popp
et al. [4] discuss various active and passive means of controlling
friction-induced oscillations. These include inclined passive
dynamic vibration absorber, active control of normal load, etc.
It is shown that stickslip oscillations can be quenched by mod-
ulating the normal load by an amount proportional to the phase-
shifted displacement or acceleration. An optimal phase shift is
460 S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469
Nomenclature
a, b parameters of the normal form equation of
Hopf bifurcation
c viscous damping coefcient
d, e parameters of the normal form equation of
Hopf bifurcation
f friction force per unit normal load
J Jacobian of the linearised ow around the
equilibrium
g, h non-linear part of ow
K oscillator stiffness
M oscillator mass
N, N
0
, N normal load, nominal normal load, normal
load variation
s function describing the Stribeck effect in Lu-
Gre model
t non-dimensional time
V Lyapunov function
v
0
non-dimensional belt velocity
v
s
non-dimensional Stribeck velocity: a param-
eter of Gaussian model of kinetic friction
X =
y
x
0
non-dimensional displacement of the oscilla-
tor mass
X

equilibrium non-dimensional displacement


of the oscillator under friction force

X state vector of the switch model


x
0
reference displacement
y displacement of the oscillator mass
y = {y
1
y
2
}
T
state vector of the rescaled equation with
(0, 0) as the equilibrium point
z non-dimensional bristle displacement
: =N/N
0
normal load modulation factor
:
c
critical value of : at cyclic-fold point
:
HB
critical value of : at Hopf bifurcation point
[, ,, j, t
stick
,
p, j
s
, o
parameters of the switch model
c boundary layer thickness
?1 a large positive quantity
generalised bifurcation parameter
z slope of the normal force with respect to ve-
locity in the boundary layer

1,2
eigenvalues of the Jacobian matrix J
j
0
, j minimumdynamic coefcient of friction, dif-
ference between the static and dynamic co-
efcients of friction
o
0
, o
1
,
o

1
, o
2
, v
d
LuGre model parameters
shown to be equal to 90

. Normal load modulation can also be


achieved by passive means, where an additional degree of free-
dom that couples the tangential and normal forces, is provided
by appending an inclined vibration absorber. Rozman et al. [5]
discuss methods of controlling chaotic stickslip oscillations
governed by frictional forces in thin (nano-scale) sheared liq-
uid between two solid surfaces. Their algorithm depends on the
necessary existence of an unstable sliding limit cycle. Efcacy
of normal and tangential dither in controlling friction-induced
vibration has also been explored. Thomsen [6] and Chatterjee
et al. [7] investigate the use of high-frequency tangential exci-
tation (the so-called dither) in controlling self-excited oscilla-
tions. Feeny and Moon [8] conducted experiments on the use
of tangential dither in controlling stickslip chaos driven by
frictional forces.
A detail review of friction-induced limit cycling and friction
compensation in controlled positioning mechanical systems can
be found in a survey paper by Armstrong-Helouvry et al. [9].
Some recent works [10,11] also address the issue of friction
compensation in mechatronic systems. These works discuss
different algorithms of model based PD and PID control of
friction related oscillations and inaccuracies.
The present paper proposes a new method of controlling
friction-induced self-excited vibration. It is assumed that self-
excitation is caused by the velocity-weakening characteristics
of friction. Efcacy of the proposed method is analysed us-
ing a single-degree-of-freedom model of the elastic systems.
The basic control scheme is to modulate the normal load in
an onoff fashion depending upon the state of the system. The
Lyapunovs second method is employed to derive the primary
form of the control law that is nally modied based on some
practical considerations. Analytical and numerical analysis
reveals the basic mechanism of the control action. An approx-
imate method for estimating the control parameters is also
proposed and the results of this approximate method are
substantiated by numerical simulation. It is also shown by
examples that the success of the proposed control law is not
conned to only a specic functional form of friction repre-
senting the velocity-weakening effect. Mainly three types of
frictional models, viz., algebraic models and the state-variable
LuGre model and the switch model with time-dependent stick-
ing friction are considered.
2. Mathematical modelling
A spring-mass system on a moving belt, depicted in Fig. 1, is
an archetypal model of friction-induced self-excited vibration.
Non-dimensional equation of motion of the system reads as

X + X = f (

X)N(X,

X), (1)
where X(=y/x
0
) is the normalised displacement of the os-
cillator mass. f (.) and N(.) represent friction force for unit
normal load and normalised normal load, respectively. The
mechanism of self-excitation is ascribed to the drooping char-
acteristics of the coefcient of friction with relative velocity of
sliding.
S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469 461
Fig. 1. Mathematical model of friction-induced self-excited vibration.
Fig. 2. Velocity dependence of friction force.
A typical velocity dependence of friction is shown in Fig. 2.
For the algebraic models of friction, like the one shown in
Fig. 2, f (.) is nothing but the coefcient of friction. However,
this is not the case for dynamic friction models similar to the
LuGre models [9]. The normal load N is more generally a func-
tion of the states of the system and is determined from the con-
trol law of normal load modulation. In Eq. (1), all displacement
and velocity quantities are normalised by a reference displace-
ment x
0
and reference velocity c
0
x
0
, respectively. c
0
is taken
as the natural frequency (

K/M) of the spring-mass system


and x
0
is chosen latter in the article. All forces are normalised
by Mc
2
0
x
0
. In Eq. (1), dot denotes differentiation with respect
to non-dimensional time t = c
0
t, with t as the original time
variable.
3. Control law
In this section, a control law for modulating the normal load
to achieve global stability of the static equilibrium of the sys-
tem is discussed. Derivation of such a control law, based on
the Lyapunovs second method, is described below. A positive
denite Lyapunov function V is constructed from the potential
and kinetic energies of the system as follows:
V =
1
2
X
2
+
1
2

X
2
. (2)
The time derivative of V along the system ow is obtained as

V = f

XN. (3)
For algebraic passive models of friction f , topologically similar
to as given in Fig. 2, one can write
(v
0


X)f >0

X R. (4)
Thus, it can be concluded that for any positive value of N,

V <0
everywhere except in the range of v
0


X0. Therefore, with
the control law shown in Fig. 3(a),

V becomes negative semi-
denite (0). A more simplied law is shown in Fig. 3(b),
where normal load is switched to zero when

X>0. Both the
control laws shown in Fig. 3(a) and (b) render

V 0 and hence
according to the LaSalles theorem the system will converge to
the largest invariant set contained in the set satisfying

V = 0,
which is nothing but

X=0 in the present case. Thus, the system
is globally asymptotically stable with the control law shown in
Fig. 3(a) and (b).
The control law derived above produces zero normal load
condition during certain period. However, zero normal load
conditions may not be always practically viable. Therefore,
more general control laws, where the occurrences of zero nor-
mal load conditions may be avoided, are explored below. Such
a control law is mathematically recast as
N = N
0
N sgn(

X), (5)
where sgn(.) is the standard signum function returning the sign
of the argument.
It may be noted that the control function described in
Fig. 3(b) can be obtained from (5) for N = N
0
. Even though
the control law described in (5) is practically more encourag-
ing, one cannot conclude anything about the global stability
of the equilibrium (steady sliding) from the stability analysis
based on the Lyapunov function except for N =N
0
. The con-
trol law given in (5) is discontinuous at

X = 0, which makes
the system dynamics prone to chattering. A standard way of
avoiding chatter is to provide a thin boundary layer around
the discontinuity. Thus, the modied control law, as shown in
Fig. 4, is mathematically written as
N =
_
N
0
+N,

X< c,
N
0
N(

X/c), c

Xc,
N
0
N,

X>c.
(6)
The following continuous description of the control law de-
scribed above is mathematically (and often practically) more
convenient:
N = N
0
N tanh(z

X), (7)
where z?1 and NN
0
. At this point, the reference displace-
ment quantity x
0
is conveniently chosen to render N
0
= 1 and
dene the normal load modulation factor : =N/N
0
1. The
non-linear control law (7) physically switches the normal load
462 S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469
Fig. 3. Semi-active control function of normal load modulation.
Fig. 4. Modied control function with boundary layer.
between two predetermined levels depending upon the state of
the system. z controls the width of the boundary layer where
the switching of the normal load from one level to the other is
smoothened to avoid chatter. The numerical value of z should
be such that the sampling error (of the digital controllers) and
the sensor discretisation error remain considerably lower than
z
1
. Moreover, the value of z determines the effective viscous
damping coefcient during switching and hence the dynamics
inside the boundary layer.
4. Bifurcation and stability
Global stability of the system with the control law (7), as
mentioned earlier, cannot be ascertained by the Lyapunov anal-
ysis. Local stability of the steady sliding condition (

X=0) can
indeed be judged from the eigenvalues of the linearised ow of
the system. However, the basin of attraction of the steady slid-
ing condition cannot be determined from the local analysis. In
this section, global stability of the system is studied using the
local bifurcation analysis of the system.
Eq. (1) may be rewritten in the following state-variable form:

y =
_
0 1
1 0
_ _
y
1
y
2
_
+
_
h(y
1
, y
2
)
g(y
1
, y
2
)
_
, (8)
where y
1
=XX

,

X=y
2
, X

=f (0)N(0) and y ={y


1
y
2
}
T
.
For Eq. (1) one writes h = 0, g = X
s
+ f (y
2
)N(y
2
) and
h(0, 0) = g(0, 0) = Dh(0, 0) = Dg(0, 0) = 0.
Local stability of the steady sliding state ((0,0) state of
Eq. (8)) is analysed from the linearised dynamics around the
equilibrium as given below

y = J y =
_
0 1
1
_ _
y
1
y
2
_
, (9)
where
=
_
N
df
dy
2
+ f
dN
dy
2
_
y
2
=0
.
Eigenvalues of the Jacobian J are obtained as

1,2
=
1
2

1
2
_
4
2
. (10)
Hopf bifurcation occurs when the generalised bifurcation pa-
rameter = 0. This bifurcation is identied with two purely
imaginary eigenvalues crossing the imaginary axis. With a view
to determine the type of the bifurcation, i.e., whether the bifur-
cation is supercritical or subcritical, the following normal form
(in polar coordinates) of the Hopf bifurcation [12] is consid-
ered:
r = dr + er
3
,

0 = c + a + br
3
, (11)
where for the present problem,
e =
1
16
_
j
3
g
jy
3
2
_
y
2
=0
=
1
16
_
N
d
3
f
dy
3
2
+ f
d
3
N
dy
3
2
+ 3
d
2
f
dy
2
2
dN
dy
2
+ 3
d
2
N
dy
2
2
df
dy
2
_
y
2
=0
,
S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469 463
d =
d
d
(Re(
1,2
)) =
1
2
= 0
and
c = 1.
Using the expression of normal load given by Eq. (7), one
obtains
e =
1
16

d
3
f
dy
3
2

y
2
=0
+2z
2
df
dy
2

y
2
=0

3
f (0)
d
2
f
dy
2
2

y
2
=0
df
dy
2

y
2
=0

(12)
for = (Ndf/dy
2
+ f dN/dy
2
)
y
2
=0
= 0.
It is not difcult to see that for z?1, the second term of
the RHS of Eq. (12) numerically dominates over the other
terms. Further, when self-excitation is ascribed to the velocity-
weakening characteristics of friction, one expects df/dy
2
>0.
Therefore, from Eq. (12) it can be concluded that e >0. This
signies a subcritical Hopf bifurcation at =0. Therefore, the
bifurcation stabilises the unstable equilibrium when under-
goes a zero crossing from positive to negative value. But be-
cause of the subcriticality of the bifurcation, this gives rise to an
unstable limit cycle. This unstable limit cycle marks the bound-
ary of the basin of attraction of the equilibrium in the phase
plane. Beyond this bifurcation, the stable equilibrium is not
globally attractive. By virtue of the subcriticality of the Hopf
bifurcation, the stable equilibrium co-exists with another sta-
ble limit cycle, which physically represents oscillation (mostly
stickslip) of the spring-mass system.
As an example, the following smooth functional form of the
function f (

X) is chosen:
f = (c|v
0


X| + j
0
+je
(v
0


X)
2
/v
2
s
) tanh((v
0


X)),
(13)
where j
0
and j represent minimum dynamic coefcient of
friction and the difference between the static and the minimum
dynamic coefcients of friction, respectively. v
s
is the Stribeck
velocity and c is the viscous damping coefcient. is a large
positive quantity. Fig. 5(a) shows the variation of e with dif-
ferent parameters. It is observed from Fig. 5(a) that except in
a small region (the white encircled region) of parameters, e is
positive. This signies that the Hopf bifurcation is uncondi-
tionally subcritical for large values of z. For small values of
v
0
and z (the white encircled region), e is negative signifying
supercritical bifurcation. The critical values of the bifurcation
parameter : = :
HB
(normal load modulation factor) are plot-
ted in Fig. 5(b). From Fig. 5(b), it can be observed that the
critical values associated with the supercritical bifurcation are
relatively high. Supercritical bifurcation can indeed give rise
to globally attracting equilibrium. However, this is possible
only in a very restricted range of velocities and is not practi-
cally encouraging either because of the involvement of higher
Fig. 5. Variation of (a) e and (b) critical value of :=:
HB
for Hopf bifurcation
with z and v
0
. j
0
= 0.3, j = 0.2, v
s
= 0.1.
magnitude of normal load modulation. Moreover, the region of
supercritical bifurcation is very much sensitive to the choice
of the exact model of friction. Acknowledging all these facts,
the major emphasis is laid on the region of subcritical bifurca-
tion that is shown to be generic and insensitive to the model
(algebraic) of friction for sufciently larger values of z.
4.1. Numerical results
Before studying the efcacy of the proposed control law, it
is worthwhile to draw the bifurcation diagram of the uncon-
trolled system. The co-dimension one, bifurcation diagram of
the system with varying v
0
is numerically constructed using
AUTO [13] and is shown in Fig. 6. It can be observed from
Fig. 6 that the steady sliding condition (the xed point with

X = 0) is unstable below a critical belt velocity. A stable


limit cycle, corresponding to stickslip motion of the oscilla-
tor, coexists with the unstable steady sliding condition. Above
the critical belt velocity, the xed point undergoes subcritical
Hopf bifurcation and becomes only locally stable. Beyond this
velocity, a stable limit cycle co-exists with an unstable limit
464 S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469
Fig. 6. Bifurcation diagram of the uncontrolled system.
Fig. 7. Bifurcation diagram of the controlled system.
cycle and the locally stable steady sliding condition. At some
higher critical velocity, the unstable and stable limit cycles col-
lide and disappear at the cyclic-fold bifurcation point. Thus,
above this higher critical velocity, steady sliding becomes glob-
ally stable. The bifurcation structure shown in Fig. 6 is typical
of a frictionally excited system and has been experimentally
veried [14].
Bifurcations of the system dynamics with the varying level
of normal load modulation : are shown in Fig. 7. It can be
observed from Fig. 7 that due to normal load modulation, an
unstable limit cycle is generated through subcritical Hopf bi-
furcation. With the increasing level of modulation, the unsta-
ble limit cycle grows and the stable limit cycle shrinks. Ulti-
mately, at some critical value of :=:
c
, the stable limit cycle is
annihilated upon colliding with the unstable limit cycle. This
is identied with the cyclic-fold bifurcation. Therefore, steady
sliding becomes globally stable above a critical value of normal
load modulation. Stabilisation is possible even for lower val-
ues of : <1. This fact is practically very encouraging, as men-
tioned earlier, because global stability is achieved even without
switching to zero normal load condition.
Fig. 8 shows the evolution of the stable and unstable limit
cycles in the phase plane with the increasing value of :. It can be
clearly observed from the phase-plane plots that the amplitude
as well as the duration of sticking of the stable stickslip limit
cycle decreases with the increasing value of :, whereas the
amplitude of the unstable limit cycle, representing oscillatory
sliding of the oscillator, increases with :. Above the critical
value of :, the stable and the unstable limit cycles touch and
annihilate each other.
Fig. 9 shows the efcacy of the proposed control action in
the time-history plots of the displacement and velocity of the
oscillator along with the normal load modulation. The above
plots are generated by numerically integrating the equation of
motion (1) using the stiff solver ode23s, an inbuilt subroutine
of the MATLAB software. The integration is run initially for
certain duration of time with : = 0 and then the control is
switched on by changing : to 0.21. One clearly observes that
the steady state velocity of the oscillator is zero, whereas a static
deection of the spring is caused due to the dynamic friction
force. Normal load settles down to its initial value (here unity)
at the steady state. From Fig. 9, it can be observed that the rate
of approach to the equilibrium state is rather sluggish. This is
imputed to the high level of damping inside the boundary layer
because of the very high value of the parameter z (1000 for
the present example). However, selecting lower values of z can
make the rate of approach faster.
4.2. Approximate estimation of critical control parameter for
global stability
From the foregoing, it is clear that the steady sliding condi-
tion is globally stable for the normal load modulation factor :
above a critical value :
c
. Thus, from the design point of view,
a method for quick calculation of :
c
has a great value. In what
follows, an approximate method to calculate :
c
is presented.
Knowing an approximate solution X
LC
= X
LC
(t + T ) of the
limit cycle oscillation, one promptly calculates the energy dis-
sipation around the cycle. At the point of the cyclic-fold bifur-
cation one then writes
_
T
0
N(

X
LC
)f (

X
LC
)

X
LC
dt = 0. (14)
Developing an approximate solution is no way a simple task.
However, from the observation of the fact that the stable limit
cycle, at the fold bifurcation, just touches the line

X=v
0
. One
may also observe that the period of the limit cycle oscillation is
not much different from 2 at the fold bifurcation point. This
prompts one to write an approximate symmetric solution in the
following form:

X
LC
= v
0
sin(t ). (15)
S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469 465
Fig. 8. Evolutions of the stable and unstable limit cycles with :. Parameter values are the same as in Fig. 6. Dotted curves and solid curves represent unstable
and stable limit cycles, respectively.
Fig. 9. Time history plot. Parameter values are the same as in Fig. 7.
Using (15) in (14), one quickly calculates :
c
. Results
obtained from the above calculations are shown in
Fig. 10. The results, as shown in Fig. 11, are matching
very closely with that obtained from the numerical sim-
ulations. It is also interesting to observe that (results are
not presented) the nal estimation of :
c
does not depend
strongly on the numerical value of the period T of the limit
cycle.
466 S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469
Fig. 10. Variation of the critical normal load modulation factor :
c
.
Fig. 11. Variation of the critical normal load modulation factor :
c
with belt
velocity. Solid line represents the results obtained from Eqs. (9) and (10),
whereas square markers represent the results of numerical simulation.
5. Structural stability issues
This section addresses a very important question regarding
the scope of the proposed technique in reality, where the fric-
tion is not necessarily governed by any nice algebraic func-
tion. As friction is a more complex process, it is worthwhile
to explore whether the above bifurcation structure (which is
the back-bone of the proposed method) still survives for some
complex realistic friction characteristics. In addressing the is-
sue, rst the state variable model of friction, the well-known
LuGre model is considered for the bifurcation analysis. Then a
more complex model incorporating time-dependent static fric-
tion is also considered.
5.1. LuGre model of friction
In recent times, a group of sophisticated models of friction
has been developed considering the microscopic degrees of
freedoms of the friction interface. Such models are known as
the bristle models, where asperities of the friction interface
are considered as elastic spring-like bristles. The bristles deect
like springs under tangential force, and the friction force is
represented as the average force of deection of the spring-like
bristles. When the deections are sufciently large, the bristles
start slipping. The relative velocity of sliding determines the
average deection during steady slipping. The LuGre model
[9] is the most widely used form of the microscopic models
of friction those rely on bristle interpretation. According to the
LuGre model, one writes
f (v) = o
0
z + o
1
dz
dt
+ o
2
v,
dz
dt
= v o
0
z
|v|
s(v)
, (16)
where v is the relative velocity between the matting surfaces
and z is the average deection of the bristles. s(v) models
the Stribeck effect (velocity-weakening characteristics) and the
most common form of g(v) is as given below
s(v) = j
0
+je
(v/v
s
)
2
. (17)
o
0
and o
1
represent bristle stiffness and damping, respectively.
o
2
is viscous damping coefcient and v
s
is the characteristic
Stribeck velocity. To render the model dissipative, the bristle
damping term o
1
is taken to be a function of velocity. The most
common form of this function is
o
1
(v) = o

1
e
(v/v
d
)
2
, (18)
where v
d
is a characteristic velocity that controls the rate of
change of bristle damping with sliding velocity.
With the above model of friction and the parameters nor-
malised according to the previous procedure, bifurcation di-
agram of the controlled system is constructed and shown in
Fig. 12. From Fig. 12, it is observed that the Hopf bifurcation
is subcritical and the bifurcation structure is qualitatively very
similar to what has been observed for a purely algebraic model
of friction. Thus, the success of the proposed control scheme
is qualitatively independent of the complexities of the friction
model considered.
5.2. Switch model of friction
Another realistic model of friction, the switch model pro-
posed by Liene et al. [15], is considered below to further
consolidate the above claim. This model incorporates the depen-
dence of static friction coefcient on sticking time, an experi-
mentally observed phenomenon in friction-driven systems with
low stiffness. According to [15], time-dependent static friction
S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469 467
Fig. 12. Bifurcation diagram of the controlled system with the LuGre model
of friction.
is described as
j
stick
= j
s
{1 + ([ 1)(1 e
,t
stick
)}, (19)
where [ = j
stick
/j
s
|
t
stick

and j
s
is the usual static friction
coefcient at t
stick
= 0.
Extending the state vector of Eq. (1) to

X = [X

X t
stick
]
T
,
one mathematically recasts the non-dimensional model of the
system as

X =
_

X
X + f N
jt
stick
_
|v
0


X| >p: slip, (20a)

X =
_

X
X + j
stick
N sgn(X)
jt
stick
_
|v
0


X| <p, |X| >j
stick
N: stick to slip transition,
(20b)

X =
_
v
0

X v
0
1
_
|v
0


X| <p, |X| <j
stick
N: sticking,
(20c)
where p is a small quantity (>v
0
) and j is a model parameter
determining the rate of exponential decay of the non-trivial stick
time state to zero during stick to slip transition. Here, another
model of velocity-weakening characteristics, different from the
Gaussian model, is considered below
f (

X) =
j
s
1 + o|v
0


X|
sgn(v
0


X), (21)
where o determines the strength of the velocity-weakening char-
acteristics of friction.
Evolutions of the limit cycles in the phase plane with the
increasing value of :, for the above model, are shown in
Fig. 13(a)(d). One clearly observes that an unstable limit cy-
cle grows with the control parameter and nally collides with
the stable limit cycle to annihilate it. This scenario reconrms
the suitability as well as the mechanism of the control action
proposed in the paper.
6. Discussions
In the foregoing, efcacy of the proposed technique in
controlling friction-induced self-excited vibration has been the-
oretically established. Analytical bifurcation analysis demon-
strates that the Hopf bifurcation is subcritical for large values
of z. In the analytical treatment, no specic functional form
representing the velocity-weakening characteristics of friction
is assumed. It is shown that the Hopf bifurcation is potentially
subcritical for any algebraic model of friction provided that z
is sufciently large. This claim is substantiated by the subse-
quent numerical bifurcation analysis for a specic functional
form of friction (the so-called Gaussian model as in Eq. (13)).
The proposed method utilises the subcriticality of the Hopf
bifurcation, which introduces an unstable limit cycle in the
phase plane. Size of the unstable limit cycle increases with the
control parameter : and nally collides with the stable limit
cycle to annihilate it. Thus, the equilibrium, which physically
corresponds to the steady sliding condition, becomes globally
stable above some critical value of :(>1). The proposed strat-
egy is shown to be structurally stable against various functional
models of friction. Thus, with the help of the approximate
method of estimating the critical control parameter to achieve
global stability of the equilibrium, the control system can be
designed to function in a robust way.
It is pertinent to say a few words about the basic advan-
tages of the proposed method over the other methods in exist-
ing literature. Various methods of controlling friction-induced
self-excited instability are discussed in the introduction section.
These methods are essentially linear in nature and hence focus
only on the local stability of the equilibrium. Moreover, struc-
tural stability remains an important issue in the control of non-
linear systems with friction on which existing literature remain
silent. However, the present method aims at the global stabil-
ity of the equilibrium. Robustness of the method under para-
metric and structural uncertainty of friction characteristics is
also addressed. The basic control action is onoff in nature and
essentially semi-active. As the proposed method controls the
normal force, power requirement is extremely low compared
to tangential force control. Moreover, in the proposed method,
no time-delay or phase shifting of the measured signal is nec-
essary, unlike that proposed by Elmer [2], Popp [4] and others
[1,3].
However, the theoretical predictions based on average fric-
tion models are to be nally veried by experimentations. An
experimental setup consisting of an oscillator on a moving belt
and an electromagnetic normal load actuator is being fabricated
in the laboratory. Experimental results may be published in fu-
ture.
7. Conclusions
The present paper proposes a new method of controlling
friction-induced self-excited vibration. Theoretical analysis
468 S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469
Fig. 13. Evolutions of the stable and unstable limit cycles with :. Dotted curves and solid curves represent unstable and stable limit cycles, respectively.
, = 0.1, j = 10, [ = 1.25, v
0
= 0.1, j
s
= 0.3, o = 3, p = 0.001, z = 1000.
shows that modulating the normal load according to the pre-
scribed control law can quench stickslip oscillation of a
single degree-of-freedom mechanical oscillator on a moving
belt. The primary control law is arrived at from the Lyapunov
based stability analysis and a more practicable control law is
deduced from an intuitive extension of the primary law. The
global dynamics of the system with and without control are in-
vestigated using analytical and numerical bifurcation analysis.
The bifurcation analysis reveals the basic mechanism behind
the quenching of limit cycle oscillations. An unstable limit cy-
cle is introduced in the phase plane and the amplitude of this
limit cycle grows with the control parameter. Above a critical
value of the normal load modulation factor (the control param-
eter), the unstable limit cycle collides with the stable stickslip
limit cycle and annihilates it. An approximate method is also
proposed to calculate the critical value of the normal load
modulation factor. It is encouraging that the qualitative nature
of the global bifurcation scenario of the system is structurally
stable for a large class of functions representing the velocity
dependence of friction that may depart considerably from the
specic examples considered in the present paper. Consider-
ations of various models of friction, like, Algebraic model,
LuGre model and switch model with time-dependent sticking
friction force in the discussion validate the robustness of the
bifurcation scenario and hence the suitability of the proposed
method.
References
[1] M.A. Heckl, D. Abrahams, Active control of friction driven oscillations,
J. Sound Vib. 193 (1) (1996) 417426.
[2] F.-J. Elmer, Controlling friction, Phys. Rev. E 57 (1998) 49034906.
[3] J. Das, A.K. Mallik, Control of friction driven oscillation by time-delayed
state feedback, J. Sound Vib. 297 (2006) 578594.
[4] K. Popp, M. Rudolph, Vibration control to avoid stickslip motion,
J. Vib. Control 10 (2004) 15851600.
[5] M.G. Rozman, M. Urbakh, J. Klafter, Controlling chaotic friction, Phys.
Rev. E 57 (6) (1998) 73407343.
[6] J.J. Thomsen, Using fast vibrations to quench friction-induced
oscillations, J. Sound Vib. 228 (5) (1999) 10791102.
[7] S. Chatterjee, T.K. Singha, S.K. Karmakar, Effect of high-frequency
excitation on a class of mechanical systems with dynamic friction,
J. Sound Vib. 269 (2004) 6189.
[8] B.F. Feeny, F.C. Moon, Quenching stickslip chaos with dither, J. Sound
Vib. 237 (1) (2000) 173180.
[9] B. Armstrong-Helouvry, P. Dupont, C. Canudas de Wit, A survey of
models, analysis tools, and compensation methods for the control of
machines with friction, Automatica 30 (7) (1994) 10831138.
[10] F. Altpeter, Friction modeling, Identication and compensation, Ph.D.
Thesis, cole polytechnique fdrale de lausanne dpartement de gnie
mcanique, 1999.
[11] D. Putra, Control of limit cycling in frictional mechanical systems, Ph.D.
Thesis, Technische Universiteit Eindhoven, 2004.
[12] J. Guckenheimer, P. Holmes, Nonlinear Oscillations, Dynamical Systems,
and Bifurcations of Vector Fields, Springer, New York, 1983.
[13] E.J. Doedel, AUTO97, Continuation and bifurcation software for ordinary
differential equations, 1998.
S. Chatterjee / International Journal of Non-Linear Mechanics 42 (2007) 459469 469
[14] R. Horvath, Experimental investigation of excited and self-excited
vibration, Masters Thesis, University of Technology and Economics,
Budapest, 2000 http://www.auburn.edu/horvaro/index2.html.
[15] R.I. Leine, D.H. Van Campen, A. De Kraker, L. Van Den Steen, Stickslip
vibrations induced by alternate friction models, Nonlinear Dyn. 16 (1998)
4154.

Você também pode gostar