You are on page 1of 14

Journal of Sound and Vibration (1988) 123(1), 157-170

OPTIMAL DESIGN OF LINEAR AND N O N - L I N E A R DYNAMIC


VIBRATION ABSORBERS
I. N. JORDANOV

Department of Programming and Computing Systems Application


AND

B. I. CHESHANKOV

Centre of Applied Mathematics, Higher Institute of Mechanical and Electrical Engineering,


Sofia 1156, Bulgaria
(Received 28 May 1987, and in revisedform 25 August 1987)
An efficient numerical method is applied to obtain optimal parameters for both linear
and non-linear damped dynamic vibration absorbers. The minimization of the vibration
response has been carried out for damped as well as undamped force excited primary
systems with linear and non-linear spring characteristics. Comparison is made with the
optimum absorber parameters that are determined by using Den Hartog's classical results
in the linear case. Six optimization criteria by which the response is minimized over narrow
and broad frequency bands are examined. Pareto optimal solutions of the multi-objective
decision making problem are obtained.

1. INTRODUCTION
The dynamic vibration absorber is a classic engineering device, consisting of a mass, a
spring and perhaps a damper, which is attached to a vibrating main system so as to
attenuate its undesirable forced vibratory response.
The first mathematical treatment of the passive dynamic vibration absorber, which is
attached to an u n d a m p e d force excited primary system, was given by Ormondroyd and
Den Hartog [1]. A detailed investigation and a discussion of optimal tuning and damping
parameters is to be found in the well known book by Den Hartog [2].
When there is no damping in the main mass and the mass ratio and spring constants
are fixed, the family o f amplitude frequency response curves passes through two invariant
points P and Q at the amplitude frequency plane irrespective of the value of the damping
constant. The classical Den Hartog procedure is to tune the absorber so that points P
and Q represent amplitudes with equal height, and then, finding the damping, to make
the corresponding response curve pass through either invariant point with a horizontal
tangent. The average of two possible damping values is taken to be the optimum.
Warburton [3] extended this optimization procedure and determined optimum parameters for absorbers which are attached to a main u n d a m p e d system with two degrees
of freedom, and minimized the m a x i m u m response. For narrowband optimization (over
a limited frequency range) his attention was focused on the first and second resonant
peaks and he has made a comparison with the classical results of Den Hartog. For
b r o a d b a n d optimization he minimized the three resonant peaks and showed that for large
values o f the mass ratio the optimum parameters are the same as for a narrow band, but
for small values of mass ratio the optimum parameters are larger than for a narrow band.
157
0022-460x/88/100157+ 14 s03.00/0
9 1988 Academic Press Limited

158

i. N. J O R D A N O V

AND

B. 1. C H E S H A N K O V

Thompson [4] considered a force excited and damped single degree of freedom primary
system to which the vibration absorber is attached. Using the frequency locus method he
determined graphical criteria for the optimization of the spring and damper rates.
Kitis et al. [5] demonstrated an efficient algorithm for minimizing the vibratory response
for a range of excitation frequencies by applying it to the problem of designing two
dynamic vibration absorbers. They were simultaneously applied to a 22 degree of freedom
undamped system with four design variables. The authors proved the applicability of the
method to large scale dynamic systems. The effect of a viscously damped dynamic absorber
on an undamped multi-degree of freedom vibrating system was investigated also by
Vakakis and Paipetis [6].
Roberson, in his work [7], included non-linearity and considered the impractical case
of a linear plus cubic spring. This was followed by that of Pipes [8], who considered a
hyperbolic sine spring with a hardening characteristic, and Arnold l'9] who confirmed
Roberson's results. In all these works the authors investigated undamped systems.
Soom and Lee [10] used non-linear programming techniques to obtain optimal parameters of linear and non-linear vibration absorbers for damped primary systems. They
examined optimization criteria other than traditional ones and obtained small improvements in steady state response by using non-linear springs.
Hunt and Nissen [11] described a means by which the suppression bandwidth (the
frequency range in which the modulus of the displacement amplitude ratio is not greater
than unity) can be appreciably broadened by use of a Beileville non-linear softening
spring connection in parallel with a viscous damper between the damped primary system
and the absorber. They demonstrated that the suppression bandwidth can be doubled
and that the use of such springs will greatly increase the possibilities of the passive
dynamic vibration absorber to reduce excessive vibration amplitudes to acceptable levels.
Some technical aspects with respect to realization were taken into account by Nissen et
aL [12]. Non-linearity has also been treated by Kojima and Saito [13].
The aim of the work described in this paper has been to optimize linear and non-linear
dynamic vibration absorbers for undamped and damped primary systems by applying
the method of sounding, proposed by Sobol and Statnikov [14], to examine the objective
functions, rather than the classical approach, and to find the Pareto optimal solution of
multi-criteria optimization.
2. SYSTEM DESCRIPTION
A model of the system under consideration is shown in Figure 1. The main mass (Mr)
is supported by a non-linear suspension (spring force ft(Xt), with a viscous damper
Y//////////////~

tx t
Primory
moss

fz( z.)
Absorber
moss

c2

x3J

Figure I. Systemwith dynamicvibrationabsorber,

OPTIMAL

DESIGN

OF ABSORBERS

159

(damping constant CI), and is excited by a harmonic force H sin (cot). The absorber mass
M2 is attached by means of a non-linear spring (spring force f2(X2)) and viscous damper
(damping constant C2). The governing equations are
(M, + M2).,Y, + M2X2+ C,X, +f~(X,) = H sin (tot),
M2X, + M2-~2 + C2,~'2 -}-A(X2) = 0.

( 1)

The non-linear spring forces acting on the masses have a hardening characteristic and
are expressed as
f ~ C X 3 = KiXi + aiX~,
3
i = 1, 2.
(2)
A steady state response is assumed,

X,=[A, lsin(cot+e,),

i=1,2,

(3)

and the method of straight linearization [15] is used (as opposed to that of Kojima and
Saito [13], who solved the non-linear simultaneous equations by applying the harmonic
balance method and using the Newton-Raphson technique, and Soom and Lee [10] who
used the simple 9
and non-linear dynamic programming). The non-linear
spring characteristics (2) thus are represented as linear functions K * X , i = 1, 2, which
are obtained by minimizing the functional
I(K~f) =

I/

[f~(X,)- g ~*x , ] 2X , 2 dX~,

i = 1,2,

(4)

Ai

and the expressions for K* are


K* = (5/A~)

Io'

f ~ ( X , ) X 3 dX,,

i = 1, 2.

(5)

By substituting equations (2) into equations (5), equations (3) into equations (1), and
rearranging, the main mass and relative dimensionless displacement amplitud e ratios can
be expressed as follows:
~k2~_.[t~4(lq.. 0:2~2Xst_
3 2
.02/82)2..1_

2 2

~:2r/ ] / ( A , + r / A2) ,

,~2

4
77
--(A~+rfA])"

(6)

Here
A , = ( I + oqh 2, X s2, _ r/2)82(1 + a 2 ; t 2 X _2~ 1 2 1 8 2 ) _ ~ 2 [ i . t ~ 5 2 ( l + a 2 , ~ X 2 ) + ~ , f 2 ] ,
A2=82~:|(1

2 2
q. a l A i X2~ 2 _ ~ 2 . _ ] d , T / 2 ) ,
q_t~2~2Xsl_172/82)q_~2(1

to 2 = K , / M,,
8 = todcoz,

ix = M 2 / M , ,

~,, = G / ( M i c o t ) ,
X~, = H / K , ,

a, = 5 a , / ( 7 K , ) ,

A, = A , / X , , ,

n = w/co,,

i = 1, 2.

(A list of notation is given in the Appendix.)

3. THE OPTIMIZATION PROBLEM


The general form for optimization problem is minimize F(~) with respect to ~. The
components of the vector (~) are the n unknown design variables Yi, 1 <~ i ~ n, which are

160

I.N.

JORDANOV

AND

B. I. C H E S H A N K O V

bounded by physical realizability criteria so that the minimization of F(y) is subject to


the constraints
y~<~y,<~y~,

l<~i<~n,

(7)

where the superscripts ' and " denote lower and upper bounds respectively.
In the linear case one considers the primary damping ratio ~j, the absorber damping
ratio ~:2 and the tuning ratio 8 as the design variables y ,
In the non.linear case, the non-linear spring ratios cq and a2 are considered to be the
fourth and the fifth design variables.
If there are two or more objectives the probe m is extended to the multi-criteria or
multi-objective optimization problem. Pareto optimization is a methodology for solving
multi-criteria decision problems. Pareto optimality is defined formally as follows.
A feasible solution to a multi-criteria optimization problem is Pareto optimal (or noninferior) if there exists no other feasible solution that will yield an improvement in the
performance of one criterion without causing a decrease in performance of at least one other
criterion.
A detailed description of multi-objective problems and methods for their solution can
be found in Cohon's book [16].
The following objective functions are considered and examined numerically.
The first is F~ = IAlp-A ~QI-This function minimizes the differences between the values
of amplitudes in points P and Q (in a linear case for an undamped primary system only).
It gives the optimum value of tuning ratio 8 in Den Hartog's sense.
The second objective function is F2 = maximum (IA~l). This function is the classical
one, used by Den Hartog in his optimizing procedure. It minimizes the maximum value
of the primary system steady state displacement response in the frequency domain.
The third is F3=maximum(IA~m~ -A'~,,,oJ). This function minimizes the difference
between the values of first and second main mass amplitude peaks. (The two resonant
peaks are assumed to be approximately of equal height.)
These first three functions for an undamped primary system in a linear case are
minimized numerically and comparison with Den Hartog's analytical optimizing procedure is made in what follows.
The fourth objective function is F4=minimum (IA~l). This function maximizes (or
minimizes the maximum of minus IA~l) the minimum of the amplitude in the frequency
range 7/~ <~~ <~7/~', where ~/~ and T/~'are the frequency ratios of the maximum values" of
the main mass amplitude.
The fifth function is F s = I T / L - ~ o I. This function maximizes the frequency range
T/L<~7/~< .qV, in which the main mass displacement amplitude ratio A~ is not greater than
unity (IA~l<~1). It is the so-called "suppression band": i.e., the fifth objective function
maximizes the suppression bandwidth. This function is optimized in a linear case.
Finally the sixth objective function is F6 = #. This function minimizes the value of mass
ratio. It is of great importance that the absorber mass should be small.
Here the optimization problem is considered as a two objective problem. By using a
special searching method, namely the method of sounding, proposed by Sobol and
Statnikov [14], the LPT sequence of testing points is obtained in the space of design
variables, with the constraints (7), by means of a simple subroutine written by these
authors. The LPT sequence is a uniformly distributed sequence of points in an ndimensional parallelepiped. In our opinion it is the best uniformly conversent sequence
known so far. A comprehensive description is to be found in reference [14].
The optimization algorithm may be summarized by the following steps: (1) give the
intervals for the parameters (the lower and upper bound for each parameter), which

OPTIMAL

DESIGN

OF ABSORBERS

161

determine the parameter space; (2) give the number of testing points; (3) compute a value
for the design vector ~ by the LP, sequence of testing points; (4) compute the displacement
response for all frequencies 77 of interest; (5) for this vector compute the values of the
functions F~(~), 1 ~<i<~ 6; (6) repeat steps 3-5; (7) make the comparisons between the
values of the function and choose the better of then for all points of the LP, sequence;
(8) construct the Pareto optimal solution.
The optimization has been carried out by using a number of subroutines, written by
ourselves in FORTRAN 77, to compute the steady state response curves and objective
functions for both linear and non-linear abs6rbers..Some subroutines from IMSL and
MINPACK packages also have been used.
4. LINEAR CASE
If there is no damping in the primary system, i.e., C, = 0 (~:l= 0) and f(Xi) = KiXi,
(al = 0), i = 1,2, the expressions for the dimensionless displacement ratios are

~,%2+(n2- ~2)2

A2 - ~'r/2('r/2- 1 -'r"p.'q2)2 + [p.6 2r / 2 - ( 2 _

1)(.q2_ 62)]2,

774
A~ = ~2n2(n2 _ 1 + ~ z ) : + [#62v/2_ ( 2 _ 1)(n2 _ 6~)]2-

(8, 9)

The design ratios are bounded by physical realizability criteria in the intervals 0.04~/.t
0.2. 0-7 ~ 6 ~ 1-0, 0.05 ~ sez~ 0.25, and the optimization is carried out in the parameter
space, which is confined to these intervals, and 256 testing points from the LP, sequence
are examined.
For the given value of the mass ratio/z, the objective function Ft = IAip-A tol, which
determines the optimum tuning ratio 6", is minimized first. Dimensionless frequency
ratios for points P and Q are given by the well known expression
n2e'~-

2+/.t

2 +/.t"

(10)

The objective functions F2, F3 and'F4 with regard to the damping ratio ~r (the optimum
tuning ratio already having been determined) are optimized subsequently.
Pareto optimal solutions of th6 objectives F~ and F4 with respect to objective F6 are
given in Figure 2. The F2 optima obtained in agreement with those of Den Hartog [2]
are also shown. It is clear from Figure 2 that our optimal curve for Fz is lower than that
I

\
10

~6

9o

~
%..
o..
".....

2
I

0.@4

0'08

0.12

0-16

0.2

F~
F i g u r e 2. Pareto o p t i m a l i t y curves for objective functions.
p a r a m e t e r s ) ; 9 9 F4; p l o t t e d a g a i n s t /=6.

, F2; - = =, F 2 (with Den H a r t o g ' s o p t i m a l

162

!. N. J O R D A N O V A N D 13. I. C t l E S H A N K O V
I

of Den Hartog for the whole range. For the value of # = 0.04 our optimum for maximum
amplitudes is about 30% lower than Den Hartog's optimum, for/.t =0.1 it is about 18%
lower, and for/~ =0.2 the difference is no more than 10%. If the objective F6 (mass ratio)
is greater than 0.14, a slight further reduction in F2 and F4 could be effected.
The optimization of F5 has been done simultaneously as regards parameters 8 and ~ .
In order to find the suppression band numerically, i.e., the frequency bandwidth between
the intersections of the unity ordinate and the response A~, a simple subroutine in
FORTRAN 77, based on the regulafalsi method is used.
As objective F6 is increased, Pareto optimality requires objectives F3 and F~ also to be
increased as shown in Figure 3. The increase of F~ is approximately linear and it is slower
than the increase of objective Fa.
The optimal tuning ratio is one of the most important design parameters for a linear
vibrating system with a dynamic absorber. Its curve, obtained as a result of optimization
of objective F~, is plotted against the mass ratio/~ in Figure 4. The change of the optimal
tuning ratio with the increase of the mass ratio is linear for F, and it is the same as the
analytically determined curve, representing Den Hartog's expression
(ll)

8 = 1/(1 + / ~ ) .

Figure 5 shows the response curves themselves, corresponding to optima at a mass ratio
of/~ = 0-1. The values of optimal parameters, simultaneously satisfying the objectives F2,
F3 and F4, are ~:2= 0.249 and 8 = 0.909. The frequency response that would be obtained

'

'

06

.1,7 0 '~

O-2

0 08

004

0.12
F6

0.16

Figure 3. Pareto optimality curves for objective functions. - -

~.0{

'

'

0.2

-, F3; - - - ,

Fs; plotted against F 6.

.I
i

0"8/~"

004

|
008

{
0-12

I
0"16

Figure 4. Optimal tuning ratio (:uP'as, plotted against the mass ratio.
d a m p e d p r i m a r y system,
i

I
0-2
, Undamped primary system; - - -

OPTIMAL
I

DESIGN
I

'

0.7

/\
I_~

/I

163

OF ABSORBERS

0-9

1.1

13

1.5

~7
Figure 5. Optimum linear response curves for the system with /x = 0.1.
, Optima (by F=, Fz, F~, F4),
~2=0.2491, 8 =0.909; - - - , with Den Hartog's optimum Parameters ~2=0.1678; ~ =0.909.

if the Den Hartog optimal values sc2=0.168 (~2=~/3/.t/[8(1 +/z)3]) and 8 =0.909 were
used instead of ours using dashed lines is plotted. It is evident from this figure that the
height of the maximum resonance peak is about 18% lower than that of Den Hartog,
and the differencebetween the two resonance peaks is also smaller.
The frequency response curve with maximum suppression bandwidth at a mass ratio
o f / x =0.1 and optimal parameters g~=0.0184, 8=0.9556, obtained after optimization
of objective Fs, is shown in Figure 6. It is seen that the response in the frequency range
0.9 ~< 7/<~ 1-05 is appreciably lower than that in Figure 5. For a vibrating system with a
dynamic absorber, adjusted to work in this range, it is advisable to design with these
optimal parameters. The lower amplitudes in this range are at the expense of the greater
heights o f the resonance peaks.
More often than not, this bandwidth is insufficient to accommodate changes in speed
due to load or supply variations, and that is the reason why means have to be found for
broadening it, as in reference [11].
Most vibrating systems, however, do have a certain amount of inherent damping, which
may be treated as viscous for small motions. In this case points P and Q do not exist at
the amplitude frequency plane and the classical Den Hartog procedure is impracticable.
For this reason objective Ft is not given any further consideration.

1
I

0-7

0.9

I
1.t

I
1-5

I
15

1'/

Figure 6. Optimum frequency response with maximum suppression bandwidth for a system with V---0.l,
~2= 0-0184, 8 = 0-9556.

164

t.N.

J O R D A N O V A N D B. I. C H E S H A N K O V

N o w the main mass damping ratio ~:, is considered to be a third design variable, and
the dimensonless amplitude ratios are expressed as

= { 172 2 + (172_ 82)2}/{ 172[ ,( 172 _ 82) + 2(172 _ 1 +

172)]2

+ [172(#82+ ~,~2)- (172- I)(172- 82)]2},

(t2)

A 2 = 174/{172[~:1(172' 82)'~- ~2(172-- l +/,/.172)] 2

+ [.r/2(/x82 + ~, se2)_ (7/2 _ 1)(172 _ 82)]2}.

(13)

The parameter space is bounded by the intervals 0.04 <~/x <~ 0.2, 0.7 <~ 8 <~ 1.0, 0.05 <~ r <~
0-25, i = l, 2. The optimization procedure is carried out again with 256 testing points from
the LP, sequence.
Figures 7 and 8 show Pareto optimal solutions for objectives /72, F3, F4, Fs and F6
when there is damping in the primary system. The change in objective F2 is very small,
it is about 10% for the whole range of the change in F6, and the maximum amplitudes
are reduced by a factor of two in comparison with those of the undamped primary system,
as shown in Figure 7. When the main mass is damped, the result is that this objective is
less sensitive to changes of mass ratio. Analogous to this is the behaviour of objectives
F3, as shown in Figure 8. The Pareto optimal curves for objective F4 (Figure 7) and
objective Fs (Figure 8) are hardly different from the corresponding curves for the system

""....
4

%.
"-...
%.
%.
~
~

%..

9. .....

0.08

0.04

0.1;'

0.16

0-2

F6
9 Objective i=2; "", objective F4.

Figure 7. Pareto optimality curves for a damped system.


i

//
j.

0-14

0-10

t,P
006
f

.i

0.02
004

I
0-08

I
0.12

0-16

0.2

F6
Figure 8. Pareto optimality curves for a damped system.

, Objective F~; - - -, objective Fs.

165

OPTIMAL D E S I G N OF ABSORBERS

without main mass damping. The most critical effect on the maximum suppression
bandwidth (objective Fs) is that of the absorber damping ratio, and it must be as small
as possible. These results show good agreement with the results in reference [12].
The optimal tuning ratio curve obtained when using objective F2 is plotted in Figure
4. In the first half, the change of 8 is very slight from that for an undamped primary
system, but in the second half hardly any changes can be observed.

4L

.-"~

/N /

I
0-7

0-5

I
0 9

/.~%1\

I
1"4

I
1"3

1-5

Figure 9. O p t i m u m frequency response curves f o r a d a m p e d system w i t h # = 0 - 1 . - - ,


F2, ~:t=0.247,
~'z = 0. ! 69, $ = 0-868; - - - , F3, ~= = 0.191, ~2 = 0.229, 5 = 0.869; - 9 - , F4, ~:= = 0.006, scz= 0-249, ~ = 0-874; 99 - ,
Fs, '~t =0"206, ~2 =0"03, ~5=0.959.

The optimum frequency response curves for the system with main mass damping and
/.t = 0.1 are shown in Figure 9. In agreement with reference 1"10] the effect of all objectives
is the increase in the height o f the first resonance peak and the decrease in the height o f
the second (with the exception of F4). It is clear from the optimum response curve
corresponding to objective F4 that this function is inapplicable to a system with main
mass damping and it has not been given further consideration.
The optimum response curve corresponding to the maximum suppression bandwidth
(objective Fs) has the smallest amplitudes in the antiresonance domain and the highest
first resonance peak. This objective is employed for a system with an absorber, adjusted
to operate in this antiresonance range. But for a system with a broadband dynamic
absorber, objectives F2 and F3 are most successful.
5. NON-LINEAR CASE
Making the substitutions and rearranging expressions (6) with respect to dimensionless
amplitude ratios A~ and A2, one obtains the following non-linear simultaneous equations:
G,

(A,, A~) =

G2(A,,

A~S,(A~) + 2A",S2(A2) + A~,S~(A~) - S,(A2) = O,

A2) = A 6 Q , ( A , )

+ 2 A ~ Q 2 ( A , ) + A~ Q 3 ( A , )

- Q 4 = O,

Here
S,(,X2)

4 2
2
= A2R3+2A2R3R4+

R~o,

$2(A2) = A~R3R5+ A22R,,+ R,2,


$3(A2) = A4R,3+ A2R,4+ R,5,
$4(A2) = A4R, + A~R2+ Ro,

Q4= TI4,

Q I (,~.1) = A 4IR32 + 2A 2IR3Rs + Rt3,

Q2(A,) = A 4I R s R 4 + A I2R l l + R , 4 / 2 ,
Q3(At) = A4iRto+2AtRi2+
2
R~5,

X~,= 1,

Ro = ( 8 2 - r/2)z + ~:22"02,

(14)

166

i.N.

J O R D A N O V A N D B. l. C H E S H A N K O V

R8
g~ = ct,284,

a282~:1"tT,

R9 = ~:~r/(82" ,/2) + ~:2r/(I - -r/2-/,r/2),

2
2
RIo=R4+RT,

R2 = 2a2~2(82- r/2),

RII=R3R6+R4R~+RTRs,
R4 = c t l ( ~ 2 - "r]2),
Rl: = R4R6+ RTR9,
R, = a282(1 - r/2- p.r/2),
R,3=R~+R~, R6= (1 - r/2)(82- ~7:) ' r/2(/.82 + ~c,~:2),
R3 = a l a 2 t $ 2,

R~4= 2RsR6+ 2RaR9.

RT= atsc2"q,

R~5=R~+ R~.

The frequency response is determined by the system (14), non-linear with respect to
3.~ and A2. It is impossible to obtain the complete response curves by applying the standard
routines for solving non-linear systems directly to equations (14), because there exist
intervals of 77 where the solution of equations (14) is multi-valued. One can introduce
as a parameter dl, the curvilinear co-ordinate along A~(r/) and, expanding (14) into Taylor
series, and ignoring the non-linear terms, obtain
Gi(X ~ Ao, o).~ ~G'(A~ X~ r/~ dA,+ ~G'(x~ X~ 7~ dA24 ~G'(A~ A~ ~2) do
0AI

0A2

d r / : + d A ~ - d / 2 = 0,

~0~

Or}

i = 1,2,

(15)

where (A ~ A~ 7/~ is a point satisfying equations (14). A simple subroutine has been
written to compute equations (15) by using a modification of the Newton-Raphson
method. The calculations of the functions A~(r/) and A2(r/), accol:ding to equations (15),
start with a point obtained from the linear system (8). As an illustration, the frequency
response of a system with parameters/.t = 0.1, a~ = 0"005, a 2 = 0 " 0 0 5 , ~1 = 0 " 0 8 , ~2 = 0 " 0 5 ,
8 = 0"9091 is shown in Figure 10. The simultaneous equations (15) have also been solved
by using the standard routine HIBRJ1 from the MINPACK,package. The purpose Of
this routine is to find a zero of a system of N non-linear functions in N variables by a
modification of the Powell hybrid method. The user must provide a subroutine which
calculates the functions and the Jacobian. Both frequency responses showed excellent
agreement. The dashed lines in Figure 10 show a frequency response, computed by direct
integration of equations (1), with non-linearity of form (2) by using the DGEAR routine
from the IMSL package (differential equation solver--variable order Adams predictor,
corrector method or Gears method). It is seen that the responses ai-e in good agreement
and unstable ranges are observed. A similar system has been investigated in reference
[12], the difference being that the main mass-spring characteristic was linear and a
non-linear softening spring was used in the absorber. Similar unstable regions in the
frequency response are shown in reference [12].
10

,,<

2
0'6

1,0

"q

1"4

1-8

Figure I0. Frequency response for a non-linear system with /* = 0 . l , 8 =0.909, cq = az=O'O05, ~t =0"08,

6=0.05.

167

OPTIMAL D E S I G N OF ABSORBERS

If the spring characteristic o f the primary system is linear, i.e., at = 0 equations (14)
are expressed as
Z(Xt,

)k2) = ,~ 2S3(,~2) = - $ 4 ( X 2 )

= 0,

Z()k2' )

6
4
2
=A2Rts+A2Rt4+A2Rts-Q4=O.

(16)

- (2omputing the frequency response in this case is easier. The second of equations (16)
is first solved by means of Z R P O L Y routine from the IMSL package. Substituting the
real roots of this solution in the first o f equations (16), one can solve it for 3.~ and obtain
the frequency response o f the primary system.
Now optimization is carried out with 128 testing points from the LP~ sequence and
the parameters are bounded in the intervals 0.04 ~ p. ~ 0.2, 0.7 <~ ~ ~< 1.0, 0.05 <~sri <~0.27,
0.002<~ a2<~ 0.1. Pareto optimal Solutions for objectives F2, F3 and F6, when there is
non-linearity in the absorber, are shown in Figure 11. Comparing the solutions in Figure
7 and Figure l l ( a ) one can see that some improvement is achieved in this case. For
example, the optimal parameters obtained for lhe system with ,u. =0.1 are ~ =0.757,
~:, = 0.269, ~:: = 0.241 and a2 = 0.0048. The differences in the solutions for objective F3
(Figures 8 and l l ( b ) ) were found to be negligible, except for a very small mass ratio.
Optimal parameters relating to this objective f o r / z =0.1 are 8 = 0 . 7 7 3 , ~:t =0.162, s%=
0.226 and a2 = 0"045.

(a)

0"1

(b)

008

0-12

2.4
0"06

u~
2.2

2.0
004

I
0 08

i
0"12

Fe

i
0-16

0"2

0"04

T
0'16

0.2

#6

Figure 11. Pareto optimality curves for a system with a non-linear absorber.

The system with two non-linear spring characteristics of form 9 was optimized in a
different manner from that in the previous case. The values o f the damping ratio ~t and
st2 are fixed (srt=sr2=0.1), the tuning ratio is determined from equation (11), only
non-linear ratios are changed in the intervals 0.002~< ai<~0"l, i = 1, 2, and the objective
functions F2, F3 and F6 are optimized first.
After that the non-linear ratios are fixed at a~ = a2 = 0.05, and the damping ratios are
changed in the intervals 0.05<~ ~:i<~0-27. In both cases the number of testing points from
the LP~ sequence is 128. The results of the optimization are given in Figures 12 and 13.
From Figure 12 it is seen that the change of objective F2 is about 28% for the whole
range of F6 when the damping is fixed (e.g., for a system with p . = 0 . 1 , 8=0.9091,
~t = st2 =0"1, the optimal p a r a m e t e r s a r e a, =0.0807 and a 2 = 0"0065), while the change
is about 17% when the non-linearity is fixed (e.g., for a system with p. =0.1, 8 =0.9091,
a~ = a2 = 0.05, the optimal damping ratios are ~:~= 0.241 and st2 = 0.222). It is clear that
the damping influence on objective F2 is weaker than that o f the non-linearity (in the
parameter range considered here). As regard s objective F3, when the damping ratios are
fixed, the improvement of the Pareto solution is insignificant for F6 greater than 0.1,
while for F6 less than 0.1 it is considerable, as shown in Figure 13. For example, for a
system with/~ = 0 - I , 8 =0.9091 and srt = st2 =0.1 the optimal parameters are a~ =0'0681

168

I.

N.

JORDANOV

AND

B.

\,,

2.6

u~

I. C H E S H A N K O V

2.2

1.4

004

0"08

0"12
F6

0-16

Figure 12. Pareto optimality curves for a non-linear system.

0"2

- - - , F i x e d ~1 = ~r2 = 0 . 1 ;

,fixed a I

= a2 = 0.05.

1-O

'~ 0.6

.,,,,

0-2

%%
I

0"0

0"08

"

"I~ . . . .

0-12

0-16

0-2

F6
Figure

13.

Pareto optimality curves for a non-linear system.

- - - , F i x e d ~:t = 'f2 = 0 . 1 ;

fixed a t =

a 2 = 0'05.

and a~ = 0"0057. By comparison with the F3 curve for fixed non-linear ratios, it is obvious
that then the objective is influenced more substantially by non-linearity than by damping.
Finally, optimization is carried out with five design parameters bounded in the following
intervals 0.7 ~< 8 ~< 1.0, 0.005 ~< ai <~ 0.1 and 0.05 ~< ~:~<~ 0.25, i = 1, 2. Pareto solutions for
objectives F2, F3 and F6 are given in Figure 14. It is evident that the Pareto solutions for
both objectives are not so sensitive to parameter changes for F6 greater than 0.12 than
they are for F6 less than 0-12. In other words, the greater absorber mass makes Pareto
optimality less sensitive to parameter changes.
The frequency response curves, with the obtained optimal parameters for the system
w i t h / z = 0 . 1 are shown in Figure 15.

(o)

(b)

3"2

0-02

2"8

0'01

2.4
0"04

0"08

012
F6

0"16

0"2

0'0
0-04

008

0-12

0'16

F6

Figure 14. Pareto optimality curves for a non-linear system.

0-2

OPTIMAL

3-0

DESIGN

OF

169

ABSORBERS

018

12

2.0

1-0

04

16

Figure 15. Optimum frequency response curves for a non-linear system with t.t = 0. I : - - -, (by F2) a~ = 0.039,
az = 0.004, ~l = 0-236, ~z = 0-172, ~5= 0.783;
,(byFa) al=O.OO7, az=O.OI6,~j=O.189,~,_=O.184,8=0.754.
6. CONCLUSIONS
A n efficient algorithm, based on the Sobol a n d Statnikov t e c h n i q u e , has b e e n d e v e l o p e d
a n d a p p l i e d to the o p t i m a l design o f linear a n d n o n - l i n e a r d y n a m i c v i b r a t i o n absorbers.
Design o p t i m a l p a r a m e t e r s are easily generated for cases where analytical solutions have
not b e e n f o u n d .
Pareto o p t i m a l s o l u t i o n s for objective f u n c t i o n s , other t h a n the traditional one, lead
to different optimal design parameters.
In some cases, results better t h a n those k n o w n until recently are o b t a i n e d .

REFERENCES
1. J. ORMONDOYD and J. P. DEN HARTOG 1928 Transactions of the American Society of
Mechanical Engineers 50, 9-22. Theory of the dynamic vibration absorber.
2. J. P. DEN HARTOG 1956 Mechanical Vibrations. New York: McGraw-Hill, fourth edition.
3. G. B. WARBURTON 1981 Earth-quake Engineering and Stn~ctural Dynamics 9, 251-262.
Optimum absorber parameters for minimizing vibration response.
4. A. G. THOMPSON 1981 Journal of Sound and Vibration 77, 403-415. Optimum tuning and
damping of a dynamic vibration absorber applied to a force excited and damped primary system.
5. L. KITIS, B. P. WANG and W. D. PILKEY'1983 Journal of Sound and Vibration 89, 559-569.
Vibration reduction over a frequency range.
6. A. F. VAKAKIS and S. A. PAIPETIS 1986 Journal of Sound and Vibration 105, 45-60. The effect
of a viscously damped dynamic absorber on a linear multi degree of freedom system.
7. R. E. ROBERSON 1952 Journal of the Franklin Institute 254, 205-220. Synthesis of a non-linear
dynamic vibration absorber.
8. L.A. PIPES 1953 Transactions of the American Society of Mechanical Engineers Journal of Applied
Mechanics 20, 515-518. Analysis of a non-linear dynamic vibration absorber.
9. F. R. ARNOLD 1955 Transactions of the American Society of Mechanical Engineers Journal of
Applied Mechanics 22, 487-492. Steady-state behaviour of systems provided with non-linear
dynamic vibration absorbers.
10. A. SOOM and M. LEE 1983 Transactions of the American Society of Mechanical Engineers Journal
of Vibration, Acoustics, Stress, and Reliability in Design 105, 112-119. Optimal design of linear
and non-linear vibration absorbers for damped systems.
11. J. B. HUNT and J.-C. NlSSEN 1982 Journal of Sound and Vibration 83, 573-578. The broadband
dynamic vibration absorber.
12. J.-C. NISSEN, K. POPP and B. SCttMALtlORST 1984 International Conference on Non-linear
Oscillations X Papers 2, 706-711. Optimization of a non-linear dynamic vibration absorber.
13. M. KOJIMA and H. SAIiO 1983 Journal of Sound and Vibration 88, 559-568. Forced vibrations
of a beam with a non-linear dynamic vibration absorber.

170

|. N. J O R D A N O V

AND

B. I. C H E S H A N K O V

14. 1. M. SO BOL and R. B. STATN 1KOV 1981 Optimal Parameters in the Multicriteria Design Problems.
Moscow: Science (in Russian).
15. Y.G. PANOVKO 1953 Transactions o f , b e A N USSR 13, 72-78. Method ofstraight linearization
in non-linear vibration problems. (In Russian.)
16. J. F. COHON 1978 Muhiobjective Programming and Planning. New York: Academic Press.
APPENDIX: NOTATION
Mr, M 2
C,, C:
Kt, K2
XI, X3

x==x3-x,
H
Xs, = H / K ,
to

t
CO1
0)2

'~1 ~ E2

~2
A , , A2

w,

Oli

primary and absorber masses


viscous damping constants
primary and absorber spring constants
primary and absorber mass displacements
relative displacement
excitation force amplitude
static deflection of the primary mass
excitation frequency (rad/s)
time
= vC-K~/M,, primary system natural frequency
=~/k2/M2, absorber natural frequency
phase angles
= M 2 / M r , mass ratio
=to2/to,, tuning frequency ratio
=w/to,, frequency ratio
= C t / ( M , tol) primary damping ratio
= Cz/(M2tol) absorber damping ratio
non-linear spring coefficients
amplitudes
differentiation with respect to t
=5ciJ(7Ki), non-linear spring ratios
=A~/X~,, i = 1, 2, displacement amplitude ratios

You might also like