Journal of Sound and Vibration (1995) 187(3), 403420

ON IMPACT DAMPERSFOR NON-LINEAR

VIBRATING SYSTEMS
S. Cn:11rrrr, A. K. M:iiii :Nb A. Gnosn
Department of Mechanical Engineering, Indian Institute of Technology,
Kanpur 208016, India
(Received 17 May 1994, and in nal form 24 August 1994)
The performance of an impact damper for controlling the vibration of a harmonically
excited, hard Dungs oscillator is investigated. Both elastic and inelastic collisions are
considered. The optimum design is based around a low amplitude, stable solution predicted
by the harmonic balance method. The nal design is obtained and veried by numerical
integration. Also, the use of an impact damper as an ono migration controller for a guided
transition from the resonance to the non-resonance branch of the solution is proposed. This
scheme obviates the need of continuous, noisy operation of an impact damper.
7 1995 Academic Press Limited
1. INTRODUCTION
Impact dampers have been successfully used for controlling the high-amplitude vibration
of various systems, such as cutting tools [1], turbine blades [2] and tall exible structures like
chimneys [3]. The vibration of the primary systemis controlled by the transfer of momentum
to a secondary (loose) mass through repeated impacts. Several research works have been
reported on the design and performance of an impact damper for controlling the vibration
of a harmonically excited linear system. For example, Grubin [4] and Warburton [5]
considered a symmetric, two collisions per cycle motion of such a system and derived the
optimum free space for the loose mass to move between collisions. Masri and Caughey [6]
discussed the exact solution and its asymptotic stability for a similar situation. In another
paper, Masri [7] obtained all possible types of steady state motions and delineated the zones
in the parameter space where these dierent types of motion occur. It was concluded that
an impact damper is most eective with two symmetric collisions per cycle. Bapat et al. [8, 9]
have consideredbothsingle- andmulti-unit impact dampers andprovidedeasy-to-use design
charts for the former. Yasuda and Toyoda [10] have provided an experimentally obtained,
empirical, equivalent viscous damping coecient to account for the inelastic collisions. The
use of bean-bags as the secondary mass, instead of metallic shots, has been proposed by
Popplewell and Semercigil [11] for noiseless operation of the impact damper. Masri and
Ibrahim [12] have considered the performance of an impact damper where the system is
subject to a random excitation. Despite several advantages, such as simple constructional
features, the use of a passive impact damper is limiteddue toits ineectiveness for broadband
excitations. Moreover, the optimal design of a passive impact damper is highly sensitive to
the model of the primary system. To overcome this limitation, an active impact damper with
on-line control algorithm has also been proposed [13].
In this paper, the performance and design of an impact damper for controlling the high
amplitude vibration of a non-linear oscillator is discussed. Specically, the primary system
is taken as a harmonically excited, viscously damped, Dung oscillator. Two dierent
403
0022460X/95/430403+18 $12.00/0 7 1995 Academic Press Limited
s. cn:11rrrr r1 :i. 404
approaches are considered. First, the secondary mass is used in a passive manner to generate
a stable low amplitude solution for the primary system. Towards this end, an initial design
obtained by the single term harmonic balance method is improved upon by using the
numerically integrated results. In the second approach, the multi-stability feature of the
non-linear oscillator is utilized. A guided ono control strategy is followed to use the
secondary mass to change the initial state of the primary oscillator. This change of state
brings the primary oscillator within the basin of attraction of the non-resonance (low
amplitude) branch from the resonance branch. This procedure has been referred to as the
guided branch transition. Numerical results are presented to verify the solutions, their
stability and thereby the feasibility of the proposed approaches.
2. THEORETICAL ANALYSIS
2.1. n:rxoNic n:i:Ncr xr1nob
The model under consideration is shown in Figure 1. For a linear primary system, it has
been shown in previous works [7, 9] that an impact damper undergoing elastic collisions
performs better than the one having inelastic collisions. Accordingly, here also we rst
consider elastic impacts between the primary mass m
1
and the secondary mass m
2
. The elastic
impacts are modelled by two linear contact springs, each of stiness k
2
. Later on, we shall
include the impact damping to simulate inelastic collisions having a coecient of restitution
xQ1. The value of x is known to control the frequency bandwidth of the eectiveness of
the impact damper [7]. However, the friction force between m
1
and m
2
is neglected.
The equations of motion for m
1
and m
2
are
m
1
x0
1
+c
1
x'
1
+k
l
x
1
+k
1
x
3
1
+k
2
[(yd/2)U(yd/2)+(y+d/2)U(yd/2)]=F cos vt
(1)
and
m
2
y0+k
2
[(yd/2)U(yd/2)+(y+d/2)U(yd/2)]=m
2
x0
1
, (2)
where the prime denotes dierentiation with respect to time t, and y=x
1
x
2
. All other
symbols are explained in Figure 1. In equations (1) and (2), U( ) is the Heaviside step
function dened as
U(r)=
6
0
1
if rQ0
if rq0
7
.
Figure 1. A model of an elastic impact damper.
ixi:c1 b:xirrs ror NoN-iiNr:r s.s1rxs 405
Substituting the non-dimensional parameters Z
1
=x
1
/x
0
, Z
2
=y/x
0
, v
0
t=T,
v
0
=z(k
1
x
2
0
/m
1
), r
1
=k
2
/m
1
v
2
0
, r
m
=m
2
/m
1
, d
0
=d/2x
0
, h=c
1
/m
1
v
0
, V=v/v
0
,
v
2
l
=k
l
/(m
1
v
2
0
) and x
0
=(F/k
1
)
1/3
in equations (1) and (2) yields
Z
1
+hZ
1
+v
2
l
Z
1
+Z
3
1
+r
1
f(Z
2
)=cos VT, Z
2
+(r
1
/r
m
)f(Z
2
)=Z
1
, (3, 4)
where f(Z
2
)=(Z
2
d
0
)U(Z
2
d
0
)+(Z
2
+d
0
)U(Z
2
d
0
) and the dot denotes dierentiation
with respect to T.
After substituting into equations (3) and (4) solutions of the form
Z
1
=a cos (VT+u
1
), Z
2
=b cos (VT+u
2
) (5)
one can expand f(Z
2
) in the Fourier series, retaining only the rst harmonic. Thereafter,
equating, respectively, the coecients of sin (VT+u
1
), sin (VT+u
2
), cos (VT+u
1
) and
cos (VT+u
2
) fromboth sides one obtains the following set of non-linear algebraic equations:
av
2
l
aV
2
+
3
4
a
3
+r
1
B(b) cos (u
2
u
1
)=cos u
1
, (6a)
haVr
1
B(b) sin (u
2
u
1
)=sin u
1
, (6b)
bV
2
+(r
1
/r
m
)B(b)=aV
2
cos (u
1
u
2
), sin (u
1
u
2
)=0, (6c, d)
where
B(b)=
6
b[1(2/p){(d
0
/b)z1(d
0
/b)
2
+sin
1
(d
0
/b)}]
0
if =d
0
/b=Q1
if =d
0
/b=e1
7
, (6e)
Equation (6d) implies u
1
u
2
=np, n=0, 1, 2, . . .. If u
1
u
2
=0, 2p, 4p, . . . , then equations
(6a), (6b) and (6c) yield
av
2
l
aV
2
+
3
4
a
3
+r
1
B(b)=cos u
1
, haV=sin u
1
(7a, b)
bV
2
+(r
1
/r
m
)B(b)=aV
2
. (7c)
Similarly, with u
1
u
2
=p, 3p, 5p, . . . , one obtains
av
2
l
aV
2
+
3
4
a
3
r
1
B(b)=cos u
2
, haV=sin u
2
, bV
2
+(r
1
/r
m
)B(b)=aV
2
,
(8ac)
which, with a=a*, can be rewritten as
a*v
2
l
a*V
2
+
3
4
a*
3
+r
1
B(b)=cos u
2
, ha*V=sin u
2
,
bV
2
+(r
1
/r
m
)B(b)=a*V
2
. (9ac)
Thus, one can see that the solution set (a, b, u
1
) obtained from equations (7) is the same as
the solution set (a*, b, u
2
) obtained from equations (9). Therefore, one can use only
equations (7) to obtain all the solutions.
In all real life situations, the value of r
1
is very high, implying
b=d
0
+e, where 0QeW1. (10)
Using equation (10) in equation (6e) yields
B(b)=(2/p)z2d
0
e+o(e) for d
0
q=b=, (11)
where e is obtained from equation (7c) as
ze=r
m
p(ad
0
)V
2
/2z2d
0
r
1
. (12)
s. cn:11rrrr r1 :i. 406
Using relations (11) and (12) in equations (7a) and (7b), one nally obtains
av
2
l
aV
2
+
3
4
a
3
r
m
V
2
(ad
0
)=cos u
1
, haV=sin u
1
. (13a, b)
By substituting D=tan (u
1
/2) in equations (13a) and (13b) and eliminating the variable a,
the following equation in D is obtained:
D
6
(n+1)+aD
5
+(3n+1)D
4
+(b+2a)D
3
+(3n1)D
2
+aD+(n1)=0. (14)
Here a=(2/hV)(v
2
l
V
2
r
m
V
2
), b=6/h
3
V
3
and n=r
m
V
2
d
0
. After solving the
polynomial equation (14) for D, one can nd the values of the amplitudes (a) from equation
(13b). Thus, all the symmetric, two impacts per cycle motions are obtained.
Before proceeding on to the stability analysis of these solutions (presented in the next
section), it may be worthwhile to note that equations (13a) and (13b) represent the single
term harmonic balance solution of the following dierential equation:
x+hx +v
2
l
x+x
3
=cos VTr
m
xr
m
V
2
d
0
cos (VT+u
1
). (15)
Equation (15) implies that the addition of an impact damper to the primary oscillator is
equivalent tothe additionof accelerationfeedbackandforce-modicationmechanisms. This
conrms the use of the term acceleration damper [4, 5] as a synonym for an impact
damper.
2.2. s1:niii1. :N:i.sis
To analyze the stability of the solutions obtained in the previous section, one considers
Z
1
=a(T) cos (VT+u
1
(T)), Z
2
=b(T) cos (VT+u
2
(T)) (16a, b)
with a(T), b(T), u
1
(T) and u
2
(T) as slowly varying functions of time. Substituting equations
(16a) and (16b) in equations (3) and (4) and carrying out the harmonic balance after
neglecting small terms such as a(T), b(T), u
1
(T), u
2
(T), ha , hb , etc., one obtains the following
four autonomous dierential equations:
a =(1/2V){haV+r
1
B(b) sin (u
2
u
1
)+sin u
1
}, (17a)
u
1
=(1/2aV){cos u
1
r
1
B(b) cos (u
2
u
1
)av
2
l

3
4
a
3
+aV
2
}, (17b)
b=(1/2V){sin u
2
+haV cos (u
1
u
2
)+(av
2
l
+
3
4
a
3
) sin (u
1
u
2
)}, (17c)
u
2
=(1/2Vb){bV
2
r
1
B(b)(1+1/r
m
)+haV sin (u
1
u
2
)
(av
2
l

3
4
a
3
) cos (u
2
u
1
)+cos u
2
}. (17d)
While deriving equations (17a)(17d), the Fourier series of the function f(Z
2
), with
Z
2
=b(T) cos (VT+u
2
(T)), has to be worked out. However, the Fourier coecients of f(Z
2
)
cannot be obtained in a closed form. Hence, the following approximation was made. Since
r
1
is very large, the contact time during an impact is very small. Consequently, f(Z
2
) is zero
for most of the time during the period 2p/V and assumes a non-zero value only for a very
small interval of time during which the slowly varying functions b(T) and u
2
(T) can
eectively be considered as constants. This approximation makes the closed formevaluation
of the Fourier coecients possible. The stability of the solutions can thus be judged from
the signs of the real part of the eigenvalues of the Jacobian (see Appendix A) (evaluated at
(a, b, u
1
, u
2
)) of the right side of equations (17). If the real parts of all the eigenvalues are
negative, the solution is stable and otherwise it is unstable.
ixi:c1 b:xirrs ror NoN-iiNr:r s.s1rxs 407
3. DESIGN OF AN ELASTIC IMPACT ABSORBER
By design of the impact absorber, we mean the determination of the values of the
parameters r
m
and d
0
for which the response of the primary system, i.e., the value of the
amplitude (a), is rendered small. Before going into the details of this design, the following
points should be mentioned.
(i) The variation of the amplitude of the primary system with the forcing frequency, in
the absence of the impact absorber, has the general trend shown in Figure 2 (for not too
high values of h and v
l
). The frequency corresponding to the maximum value of the
amplitude is dened as the peak frequency V
p
.
(ii) The relative amount of non-linear contribution in the restoring force of the primary
system can be controlled by the parameter v
l
. A higher value of v
l
implies a lower relative
non-linearity and vice versa.
(iii) To integrate numerically the equations of motion, the fourth order RungeKutta
Merson algorithm of NAG with adaptive step-size control was used. We have randomly
selected a large number of initial conditions within a ball of radius 2 around the origin of
the phase space. All these initial conditions resulted in motions where collisions occurred.
There may exist certain special combinations of initial conditions (depending on the value
of other parameters) where no collision occurs. In this work, no further attention has been
given to such non-colliding motions.
(iv) The steady state of the motion has been ascertained by considering the Poincare
section (stroboscopic map with 2p/V as the strobe period) of the phase trajectories. When
the iterates of the Poincare map reach a xed point, the corresponding motion obviously
has a period 2p/V. Similarly, the existence of n xed points in the iterates implies a motion
having a period 2pn/V, which is used to locate the subharmonic solutions.
From the approximate solution given by equation (14), it is easy to see that if r
m
and d
0
are chosen to satisfy the relation
r
m
V
2
d
0
=1 (18)
then one of the values for the amplitude (a) turns out to be zero. Of course under such a
situation, i.e., when equation (18) holds good, the solution of equations (7) does not give
the amplitude as exactly zero, but does reveal a very small amplitude of the primary system.
By following the stability analysis presented in section 2.2, this small amplitude solution is
found to be stable. The existence of such a stable small amplitude solution is also conrmed
by obtaining it through direct numerical integration. It was also observed that the other
impacting solutions, dierent from this small amplitude solution, obtained from equations
(14) and (7) match very closely.
Figure 2. A typical frequency response curve of a sinusoidally driven Dung oscillator. ----, Stable branch;
, unstable branch; r, one-third subharmonic. vl=0, h=0215.
s. cn:11rrrr r1 :i. 408
With the design frequency selected as V
p
, the parameters r
m
and d
0
are chosen to satisfy
the relation
r
m
d
0
=1/V
2
p
. (19)
The frequency responses obtained from equation (7) for a xed value of r
m
with d
0
given by
equation (19) are shown in Figures 3(a)3(c) for various value of v
l
. With the impact damper
in operation, only the stable solutions are shown. It is seen that, irrespective of the value
of v
l
, a stable small amplitude solution is always present around the peak frequency V
p
.
Apart from this low amplitude solution, a high amplitude stable solution is also present.
However, there is always a region around V
p
in which only the lowamplitude solution exists.
We dene the frequency range, around V
p
where the impact damper introduces a single low
amplitude solution, in place of the resonance branch of the primary system, as the
suppression band (DV
s
). These suppression bands, indicated in Figures 3(a)3(c), widen with
increasing v
l
(i.e., decreasing relative non-linearity). However, within the suppression band,
the percentage reduction in the amplitude is relatively insensitive to the degree of
non-linearity. A decrease in v
l
reduces the value of V
p
. Consequently, for a given value of
r
m
, the optimumvalue of d
0
obviously increases (see equation (19)) as can also be noted from
Figures 3(a)3(c). It is known from previous works [9] on linear systems, that near the
resonance frequency, the loose mass undergoes two symmetric collisions per cycle. To
examine what happens in the presence of a non-linear restoring force, the phase diagrams
of the primary system were also studied. A representative diagram is shown in Figure 4,
which clearly reveals the existence of two symmetric impacts/cycle.
From now onwards, the primary system will be considered to have only a cubic restoring
force: i.e., v
l
=0. The frequency responses of the system obtained from equation (7) for
dierent combinations of r
m
and d
0
satisfying equation (19) are shown in Figures 5(a)5(c).
The results obtained by the direct numerical integration are also shown in the same gures.
The small amplitude motion is vindicated by the direct numerical integration (revealing two
Figure 3. The frequency response of a primary system with and without the impact damper. ----, Stable branch
(harmonic balance) with damper; , stable branch without damper; , unstable branch without damper.
h=0215, r1=2000, rm=012. (a) vl=15, d0=069; (b) vl=1, d0=12; (c) vl=0, d0=2.
ixi:c1 b:xirrs ror NoN-iiNr:r s.s1rxs 409
Figure 4. Aphase plot of the lowamplitude motionof the primary system. rm=012, d0=2, V=2, h=0215, vl=0,
r1=2000. Initial conditions (1, 05, 0).
symmetric impacts per cycle). However, the high amplitude solution, obtained by integrating
the equation of motion (although numerically matched by those obtained by the harmonic
balance method), revealed more than two impacts per cycle. The frequency regions having
dierent number of impacts per cycle were also seen to be separated by narrow chaotic
regimes.
The number of impacts per cycle is not an important issue in the method of harmonic
balance. For a single-term approximation to yield meaningful results, only the energy
content in the fundamental frequency is of primary importance. Of course, the stability
analysis carried out in this work is approximate, and therefore is not able to capture all the
delicate instability regions. However, both the amplitude and the width of the suppression
Figure 5. The frequency response of the primary systemwith and without the impact damper. ----, Stable branch
(harmonic balance); w, numerical integration; , stable branch without damper; , unstable branch
without damper; r, one-third subharmonic without damper; , one-third subharmonic with damper. h=0215,
vl=0, r1=2000. (a) rm=012, d0=2; (b) rm=025, d0=1; (c) rm=04, d0=0625.
s. cn:11rrrr r1 :i. 410
band are estimated accurately by the harmonic balance method, which goes a long way to
providing an initial design. Both the harmonic balance and numerically integrated results
showed that the motion of the primary and the loose mass are almost in anti-phase for VQV
p
and almost in phase for VqV
p
.
One can dene the set of values of the pair (r
m
, d
0
) satisfying equation (19) as the
optimum set. In the following discussion we vary (r
m
, d
0
) only in that set. A comparison
of Figure 5(a) with Figure 5(b) reveals that the suppression band widens with increasing
values of r
m
. Thus, a high value of r
m
is to be preferred for a wider suppression
band. Moreover, the higher the value of r
m
, the lower is the value of the oscillation
amplitude in the suppression band. However, too high a value of r
m
makes the system
more prone to subharmonic oscillations (within the suppression band) as indicated in
Figure 5(c).
To determine the optimum values of r
m
and d
0
, we compute the response of the primary
system at the design frequency for three dierent values of r
m
with varying values of d
0
. The
maximum displacement of the primary mass is obtained both from equations (7) and
numerical integration. These results are shown in Figure 6. The numerically integrated
results are seentoyielda slightly higher value of the optimumd
0
as comparedtothat obtained
from the harmonic balance method. However, the motion of the primary mass at V
p
is seen
to be rather insensitive to the value of r
m
if the optimum value of d
0
is used. Thus, depending
on the required suppression bandwidth and available space, a value of r
m
can be chosen and
the rst estimate of d
0
should be obtained from equation (19). Around these values of r
m
and
d
0
, the exact optimum values can then be obtained from the numerically integrated results.
The insensitivity of the primary response near the optimum design ensures satisfactory
performance of the impact damper, even if the actual parameters are a little away from the
theoretical optimum values. These observations remain valid for dierent values of the
parameter h, as can be seen from Figure 7.
Another important feature of the impact-damper for a non-linear primary system is the
suppression of the one-third subharmonic response as discussed below. Figure 2 reveals that
the primary systemwithout the damper can have one-third subharmonic oscillation near the
design frequency. In Figures 5(a) and 5(b) it is clearly shown that the one-third subharmonic
motion can even be completely suppressed by a properly designed impact damper. With
Figure 6. A comparative study of analytical and numerical optimum design values, h=0215, V=2. rm=025:
, analytical, e, numerical. rm=018: , analytical, r, numerical. rm=012: ----, analytical, w, numerical.
ixi:c1 b:xirrs ror NoN-iiNr:r s.s1rxs 411
Figure 7. A comparative study of analytical and numerical optimum design for various values of the parameter
h. h=0107, V=Vp (=28): rm=018, , analytical, W, numerical; rm=012, , analytical, q, numerical.
h=015, V=Vp (=237): rm=018, , analytical, w numerical; rm=012, ----, analytical, r, numerical.
higher values of r
m
, the one-third subharmonic motion may appear but with considerably
lower amplitude. Such control of the one-third subharmonic response is prominently
displayed with lower values of h as indicated in Figure 8.
Figure 8. The frequency response of the primary systemwith and without the impact damper. h=0107, r1=2000,
rm=012. ----. Stable solution without damper; , unstable solution without damper; , analytical stable
solution with damper; w, numerical low amplitude solution with damper; W, numerical high amplitude solution
with damper; r, one-third subharmonic solution without damper; q, one-third subharmonic solution with
damper.
s. cn:11rrrr r1 :i. 412
Figure 9. A model of an inelastic impact damper.
4. INELASTIC IMPACT DAMPER
A mathematical model of an impact damper undergoing inelastic collisions is shown in
Figure 9, where the impact is modelled by a linear spring with stiness k
2
and a dashpot with
viscous damping coecient c
2
. The non-dimensionalized equations of motion can be written
as
Z
1
+hZ
1
+Z
3
1
+c
1
(Z
2
, Z
2
)=cos VT, Z
2
+c
2
(Z
2
, Z
2
)=Z
1
, (20, 21)
with
c
1
(Z
2
, Z
2
)=r
1
f(Z
2
)+2j
c
zr
1
r
m
M(Z
2
)Z
2
, (22)
c
2
(Z
2
, Z
2
)=(r
1
/r
m
)f(Z
2
)+2j
c
zr
1
/r
m
M(Z
2
)Z
2
, (23)
M(Z
2
)=U(Z
2
d
0
)+U(Z
2
d
0
). (24)
As shown in Appendix B, the damping ratio j
c
is obtained in terms of the coecient of
restitution (x) as
j
c
=ln (x)/z(1+r
m
)[p
2
+(ln (x))
2
]. (25)
To study the eects of the coecient of restitution on the performance of the impact
damper, equations (20) and (21) were numerically integrated to obtain the primary response.
The frequency response of the primary system is plotted in Figure 10 for x=075. One can
observe from this gure that the suppression band with xQ1 is wider than that with x=1.
However, the response amplitude with inelastic collisions is marginally higher than that with
Figure 10. The frequency response of an inelastic impact damper. x=075, rm=025, d0=1, h=0215. ,
Stable branch without damper; , unstable branch without damper; ----, stable branch with elastic damper
[analytical]; w, numerical with elastic damper, r, one-third subharmonic without damper; +, numerical with
inelastic damper.
ixi:c1 b:xirrs ror NoN-iiNr:r s.s1rxs 413
Figure 11. Numerical optimum design values of inelastic damper. , rm=012; + , rm=018; ----,
rm=025; all for x=075. , rm=012; , rm=018 all for x=09.
elastic collisions. In general, the higher the impact damping (i.e., the lower the value of x)
the wider is the suppression band, but the larger is the response amplitude (in the suppression
band). A similar behaviour was also seen for a linear primary system [7]. Thus, a high value
of x, signifying almost elastic collisions, is preferable for optimum performance in the case
of a synchronous operation.
The variations of the amplitude of the primary system with d
0
for various values of r
m
and
x are shown in Figure 11. From this gure it may be concluded that the higher the impact
damping, the lower is the required value of d
0
for optimum performance. Moreover, as the
impact damping increases, the primary response becomes more insensitive to the variation
in d
0
(for a near optimal design).
5. USE OF IMPACT DAMPER FOR GUIDED BRANCH TRANSITION
The concept of a guided branch transition, or migration control as it is called, is recent
in the literature [14, 15]. Non-linear systems generally have multiple, steady state, periodic
or strange (chaotic) attractors for xed parameter values and all such attractors coexist with
their respective basins of attraction. The role of a migration controller is to lead the system
to a pre-assigned attractor from any initial condition. If the initial condition falls within the
basin of an undesirable attractor, then the migration controller operates and guides the
systemto the basin of attraction of the desired attractor. Once this is achieved, the controller
is released. The whole concept can be recast mathematically as follows:
A forced non-linear oscillator can be described by an autonomous rst-order vector
dierential equation as
Z =f(Z, p(u)), u=a ,
where Z $ R
n
and u $ S
1
, f : R
n
S
1
:R
n
, and p: S
1
:R
l
with lEn. Here the dot represents
dierentiation with respect to time T and a is a constant. With control variables Z
c
and the
control force f
c
(Z
c
), the system equations take the form
Z =f(Z, p(u))+f
c
(Z
c
), Z
c
=f *
c
(u, Z, Z
c
), u=a , (27)
where Z
c
$ R
m
, f
c
: R
m
:R
n
and f *
c
: R
n
R
m
S
1
:R
m
. We represent the basin of attraction of
the undesirable attractor A
ud
of equations (26) by the set B
A
ud
and that of the attractor
s. cn:11rrrr r1 :i. 414
Figure 12. The migration control scheme.
A
c
of equations (27) by B
A
c
. Let the basin of attraction of the goal attractor A
d
of equations
(26) be represented by B
A
d
. Thus, B
A
ud
UR
n
S
1
, B
A
d
UR
n
S
1
and B
A
c
UR
n+m
S
1
. We dene
the ow represented by equations (26) and (27) by G
T
wc
and G
T
c
, respectively. Let the initial
conditions be represented by a point P
r
1
(Z
0
, u
0
) $ B
A
ud
such that lim
T:a
G
T
wc
(P
r
1
(Z
0
, u
0
)):A
ud
.
However, if the control force is added, P
r
1
gets extended to P
1
(Z
0
, u
0
, Z
c0
) $ B
A
c
and
lim
T:a
G
T
c
(P
1
(Z
0
, u
0
, Z
c0
)):A
c
UB
A
c
. If the control is released when the motion reaches a
point P
2
$ [A
c
+B
A
d
], then P
2
is restricted to P
r
2
(Z
1
, u
1
) [$ B
A
d
] when
lim
T:a
G
T
wc
((P
r
2
(Z
1
, u
1
)):A
d
[UB
A
d
]. The whole control action is schematically shown in
Figure 12.
In sections 2 and 4, we have seen that an impact damper can reduce the response of
a harmonically driven, Dung oscillator, especially in a frequency band corresponding
to the high amplitude resonance branch. Using the multiplicity features of the steady
state motion of Dung oscillator, here we use an elastic impact damper as a migration
controller. It will be seen how, after starting from the basin of attraction of the resonance
solution, the system goes to the non-resonance branch when an impact damper is used
as a controller. It is generally seen that, near the high frequency part of the resonance
branch, the probability of nding an initial condition for the non-resonance motion is
higher near the trivial (zero) initial conditions. Around these trivial initial conditions, a
wide zone exists which forms the subset of the basin of attraction of the non-resonance
motion [16]. It has already been seen in sections 2 and 3 that an impact damper brings the
primary system to a near-trivial motion. Thus, the points on the steady state attractor of
the system (including the impact damper) might be completely embedded within the
basin of attraction of the non-resonance motion of the primary system (without the impact
damper). At the steady state, the primary motion might choose the initial point suitable
for the non-resonance oscillation when the impact damper can be switched o. However,
as the system is non-autonomous, the phase of the excitation at the time of withdrawal
of the control is very important. The phase should correspond to the desired goal. Thus,
using the migration control one can avoid the continuous noisy operation of an impact
damper.
A numerical example showing the feasibility of the above scheme is as follows: say
Z=
6
Z
1
Z
17
, Z
1
=Z
1
and f=
6
Z
1
f
1
(Z
1
, Z
1
, u)
7
, a=V,
ixi:c1 b:xirrs ror NoN-iiNr:r s.s1rxs 415
Figure 13. The oscillator signature of the primary system with migration control. V=195, h=0215, rm=012,
d=2, r1=2000.
where f
1
(Z
1
, Z
1
, u)=Z
3
1
hZ
1
+cos u. If an impact damper is used as a controller
Z
c
=
6
Z
2
Z
27
, Z
2
=Z
2
, f
c
(Z
c
)=
6
0
f
c
1
(Z
2
)
7
and
f*
c
(u, Z, Z
c
)=
6
z
2
(1+1/r
m
)f
c
1
(Z
2
)f
1
(Z
1
, Z
1
, u)
7
,
where f
c
1
(Z
2
)=r
1
{(Z
2
d
0
)U(Z
2
d
0
)+(Z
2
+d
0
)U(Z
2
d
0
)}. Let us take r
1
=2000,
h=0215, d
0
=2 and r
m
=012. We consider the operation of the impact damper as a
migration controller at three dierent values of the excitation frequency.
(i) V=195. With the above-mentioned functional formand prescribed parameter values,
we numerically integrate equations (26) starting from an initial condition. Let the initial
condition be within the basin of the resonance motion. After the motion has reached the
second state, we turn the controller (i.e., the impact damper) on and again numerically
integrate equations (27) uptothe steady state. Whenthe control is switchedoat some point,
the subsequent motionleads tothe non-resonance branch. We repeat the procedure toensure
that even if the control is released at any arbitrary forcing phase, the motion still migrates
to the non-resonance branch. One example showing the primary response (Z
1
) plotted
against the non-dimensional time (T) is shown in Figure 13. The instants of switching on
and o the impact damper are indicated in this gure.
(ii) V=17. The steady state attractor of the system with the controller (equations (27))
passes through the basin of one-periodic non-resonance motion and three-periodic
motion of the system, described by equations (26), as shown in Figure 14. Therefore,
depending on the time of release of the control, the system might choose initial conditions
s. cn:11rrrr r1 :i. 416
Figure 14. The steady state attractor of the primary. V=17, rm=012, d0=2, h=0215. r, The points which
the motion without the impact damper goes to one-third subharmonic oscillation; , the points from which the
motion without the impact damper goes to the non-resonance motion.
(denitely on the above-mentioned attractor) leading to either of the two states. Thus,
one may need several onos to reach the non-resonance one-periodic motion, as shown
in Figure 15.
(iii) V=185. The response is plotted against time in Figure 16 for the case in which the
control is released before the steady state is reached. The motion is seen to reach the
three-periodic solution. Several switchings of the controller ultimately lead the system to the
non-resonance one-periodic motion.
For a number of parameter values tested, it was observed that several ono switchings
spanning, in total, not more than 350 cycles of operation could lead the response to the
non-resonance branch. For a safe application of an impact damper as a migration controller,
Figure 15. The oscillatory signature of the primary with migration control. V=17; other parameters as in
Figure 13.
ixi:c1 b:xirrs ror NoN-iiNr:r s.s1rxs 417
Figure 16. The oscillatory signature of the primary with migration control. V=185; other parameters as in
Figure 13.
one needs to design the impact damper for its optimum performance. Thereafter, a hit and
try control algorithm depicted in Figure 17 can be followed for migrating to the desired
motion. An experimental investigation of the proposed scheme is in progress.
Figure 17. An algorithm of migration control.
s. cn:11rrrr r1 :i. 418
6. CONCLUSIONS
The performance of an elastic impact damper attached to a harmonically excited Dung
oscillator has been presented. The existence and stability of a low amplitude motion near
the high frequency, large amplitude resonance branch of the primary system has been
established by using the harmonic balance method. The optimum design predicted by the
harmonic balance solution is rened by numerical integration. For a given mass ratio (r
m
),
the optimum free space for the loose mass (d
0
) is obtained. The primary response is seen to
be rather insensitive to the values of both r
m
and d
0
near this optimal design. With the
optimum design, two symmetric collisions per cycle are always seen to occur. The
suppression band of the impact damper narrows down with increasing non-linearity in the
restoring force, but the percentage reduction of amplitude at the design frequency is seen
to be insensitive to the degree of non-linearity. An impact damper is very eective for
reducing (or even eliminating) the one-third subharmonic response. An inelastic damper
widens the suppression band at the cost of higher primary response. An ono migration
control strategy, using an impact damper, for a guided transition from one solution branch
tothe other, is proposed. The feasibility of the proposedscheme is veriedthroughnumerical
examples. This scheme, applicable to other non-linear primary systems having multi-stable
solutions, obviates the need for continuous, noisy operation of an impact damper.
ACKNOWLEDGMENT
The authors gratefully acknowledge the suggestions of the reviewers (unknown) on an
earlier version of the paper.
REFERENCES
1. M. M. S:bri 1972 Machinery 120, 152161. Impact dampers for controlling vibration in machine
tools.
2. A. L. P:cr1 1937 Engineering 143, 305307. Vibration in steam turbine buckets and damping by
impacts.
3. H. H. Rrrb 1967 Wind Eects on Buildings and Structures, Proceedings of Internal Research
Seminar, Ottawa. Hanging chain impact dampersa simple method of damping tall, exible
structures.
4. C. GriniN 1956 Journal of Applied Mechanics 23, 373378. On the theory of acceleration damper.
5. G. B. W:rnir1oN 1957 Journal of Applied Mechanics 24, 322324. On the theory of acceleration
damper.
6. S. F. M:sri and T. K. C:icnr. 1966 Transactions of the American Society of Mechanical
Engineers, Journal of Applied Mechanics E33, 586592. On the stability of the impact damper.
7. S. F. M:sri 1970 Journal of the Acoustical Society of America 47, 229237. General motion of
impact dampers.
8. C. N. B:i:1, S. S:Ni:r and N. Poiiirvrii 1983 53rd Shock and Vibration Bulletin 4, 112.
Experimental investigation of controlling vibrations using multi-unit impact damper.
9. C. N. B:i:1 and S. S:Ni:r 1985 Journal of Sound and Vibration 99, 8594. Single unit impact
damper in free and forced vibration.
10. K. Y:sib: and M. To.ob: 1978 Bulletin of the Japan Society of Mechanical Engineers 21,
424430. The damping eect of an impact damper.
11. N. Poiiirvrii and S. E. Srxrrcicii 1989 Journal of Sound and Vibration 133, 193223.
Performance of the bean bag impact damper for a sinusoidal external force.
12. S. F. M:sri and A. M. Inr:nix1973 Earthquake Engineering and Structural Dynamics 1, 337346.
Stochastic excitation of simple system with impact damper.
13. S. F. M:sri, R. K. Miiirr, T. J. Drncn:N.:r and T. K. C:icnr. 1989 Transactions of the
American Society of Mechanical Engineers, Journal of Applied Mechanics 56, 656666. Active
parameter control of nonlinear vibrating structures.
14. E. A. J:cisoN 1991 Physica D 50, 341366. On the control of complex dynamic systems.
ixi:c1 b:xirrs ror NoN-iiNr:r s.s1rxs 419
15. L. Koc:rrv, A. Sn:Nc and L. O. Cni: 1993 International Journal of Bifurcation and Chaos 3,
479483. Transitions in dynamical regimes by driving: a unied method of control and
synchronization of chaos.
16. T. F:Nc and E. H. Dovrii 1987 International Journal of Non-linear Mechanics 22, 267274.
Numerical simulations of jump phenomena in stable dung system.
APPENDIX A
The elements of the Jacobian J of the right side of equations (17) are as follows:
J=G
G
G
K
k
J
11
J
21
J
31
J
41
J
12
J
22
J
32
J
42
J
13
J
23
J
33
J
43
J
14
J
24
J
34
J
44
G
G
G
L
l
,
where
J
11
=
h
2
, J
12
=
r
1
B(b)
2V
cos(u
2
u
1
)
1
2V
cosu
1
, J
13
=
r
1
2V
sin(u
2
u
1
)
1B(b)
1b
,
J
14
=
r
1
B(b)
2V
cos(u
2
u
1
), J
21
=
cosu
1
r
1
B(b)cos(u
2
u
1
)
2a
2
V
+
3
4
a
V
,
J
22
=
1
2aV
{sinu
1
r
1
B(b)sin(u
2
u
1
)}, J
23
=
r
1
cos(u
2
u
1
)
2aV
1B(b)
1b
,
J
24
=
r
1
B(b)
2aV
sin(u
2
u
1
), J
31
=
h
2
cos(u
1
u
2
)
0
9a
2
8V
+
v
2
l
2V
1
sin(u
1
u
2
),
J
32
=
ha
2
sin(u
1
u
2
)
0
3a
3
8V
+
av
2
l
2V
1
cos(u
1
u
2
), J
33
=0,
J
34
=
1
2V
{cosu
2
+haVsin(u
1
u
2
)(av
2
l
+
3
4
a
3
)cos(u
1
u
2
)},
J
41
=
1
2bV
{hVsin(u
1
u
2
)(v
2
l
+
9
4
a
2
)cos(u
2
u
1
)},
J
42
=
1
2bV
{haVcos(u
1
u
2
)(av
2
l
+
3
4
a
3
)sin(u
2
u
1
)},
J
43
=
1
2b
2
V
6
b
0
V
2

0
1
r
m
+1
1
r
1
1B(b)
1b
1
f
5
7
,
J
44
=
ha
2b
cos(u
1
u
2
)
0
av
2
l
2Vb
+
3a
3
8bV
1
sin(u
2
u
1
)+
1
2bV
sinu
2
,
f
5
=bV
2
r
1
B(b)
0
1+
1
r
m1
haVsin(u
1
u
2
)(av
2
l
+
3
4
a
3
)cos(u
2
u
1
)+cosu
2
.
s. cn:11rrrr r1 :i. 420
Figure B1. A model of an impact pair.
APPENDIX B
The mathematical model of an impact pair is shown in Figure B1. The contact dynamics
of the two masses m
2
and m
1
are represented by the dierential equations
m
2
X
2
+c
2
(X
2
X
1
)+k
2
(X
2
X
1
)=0, (B1)
m
1
X
1
+c
2
(X
1
X
2
)+k
2
(X
1
X
2
)=0, (B2)
where the dot denotes dierentiation with respect to time t.
Substituting Y
c
=X
2
X
1
and r
m
=m
2
/m
1
in equations (B1) and (B2) yields
Y
c
+2jv
n
Y
c
+v
2
n
Y
c
=0, (B3)
where j=c
2
z1+r
m
/2zm
2
k
2
and v
n
=z(k
2
/m
2
)(1+r
m
). At the beginning of the contact
phase one can write that at t=0, Y
c
=0, Y
c
=V, and at the end of the contact phase, when
Y
c
is again zero, one obtains Y
c
=Ve
jv
n
p/v
d
where v
d
=v
n
z1j
2
. Thus, according to the
denition of the coecient of restitution (x), x=exp[pj/z1j
2
], which gives
j
c
=c
2
/2zk
2
m
2
=ln(x)/[p
2
+ln(x)
2
]
1/2
(1+r
m
)
1/2
. This model of an impact pair is seen to be
satisfactory only for high values of the coecient of restitution. The model seems to fail for
low values of x implying a high value of j
c
. A high value of damping enhances the time of
contact and thus violates the condition of zero contact time during impact.