Journal of Sotmd and Vibration (I 974) 36(1), 105-117

OPTIMIZATION OF MECHANICAL VIBRATION ISOLATION

SYSTEMS WITH MULTI-DEGREES OF FREEDOM
C. NG
U.S. Naval Ship Research and Development Center,
Bethesda, ~Iaryland 20034, U.S.A.

AND

P. F. CUNNIFF
Department of ~lechanical Enghteering,
Unirersity of ~daryland,
College Park, Maryland 20742, U.S.A.

(Received I l December 1973, and ht revisedform 14 glarch 1974)

An optimization technique of vibration isolation systems with multi-degrees of freedom

is developed in this study. The primary goal is to seek a min.-max, point so that the objective
function of the system is maximized with respect to a single state variable, the frequency,
and at the same time is minimized with respect to multi-control variables, the damping
constants. The Newton-Raphson method, which utilizes the second-order derivatives of
the objective function, is used for maximization. The Davidon-Fletcher-Powell method is
selected for minimization. An iterative technique is developed to improve the solution until
the gradients of the objective function are within a prescribed tolerance. A mechanical isola-
tion system with three degrees of freedom is taken as a sample problem for experimental
verification. The optimized driving frequency and damping constants are determined by
a computer program based on the above-mentioned numerical methods. The experimental
test results agree favorably with predicted analytical values.

I. INTRODUCTION
In recent years, optimization techniques have received increasingly widespread interest in
the engineering sciences. These techniques were initiated, primarily, by the advent of high
speed digital computers which give a practical, economical and accurate means of obtaining
solutions to meaningful problems in the optimization of static and dynamic systems. The
optimization of a simple dashpot-spring-mass system was studied by Ormondroyd and
Den Hartog [1 ] and, in recent years, considerable work has been accomplished in optimizing
dynamic systems with one or two degrees of freedom subject to harmonic or impulsive loads
[2, 3, 4, 5]. More recently, parametric optimization studies were conducted [6] for various
combinations of dashpots and springs.
The main objective ofthe work reported in this paper was to optimize the steady response of
lumped parameter vibration systems with multi-degrees of freedom subject to one or several
harmonic excitations. Since the steady-state response varies with the resonant frequencies of
the system, the dampers introduced in the system are optimized in such a way that the most
critical response is reduced to a minimum. The paper includes a discussion of the techniques
of optimization and a report of the results of tests performed for supporting the technique
adopted.
105
106 C. NG AND P. F. CUNNIFF

2. FORMULATION OF THE PROBLEM

In this study, the model system is linear and consists of discretized elements of masses,
springs and dampers. The excitation is harmonic and can be applied at one or more points.
If the excitations are applied at more than one point, they must be synchronous.
The equation of motion for such a vibration system can be represented by the following
matrix form:
M.~ + C)~ + K X = Qe I~ (1)

The inertia matrix M is a positive definite matrix, and the damping matrix C and stiffness
matrix K are either positive definite or positive semi-definite. (A list of symbols is given in
Appendix I.)
For steady-state response, the displacement vector can be expressed as
X = x e I~'' (2)

and equation (1) can be written in the following form:

P X = Q, (3)
where
P = -o9 2 M + iogC + K. (4)
Thus,
X=p-1Q. (5)

In general, the system is initially assigned with certain damping values and the response
can be evaluated from equation (5). In order to avoid the imaginary terms, the matrix on the
right s~de of equation (5) is postmultiplied by its conjugate, so that

lXl 2 = p - 1 a ( P - 1 a ) , . (6)
The superscript * denotes the conjugate of the original form. The above equation is called
the primitive function. This function is first maximized with respect to the driving frequency.
Having found the driving frequency o9t which makes equation (6) a maximum, one can write
the result as follows:

max IX'l 2 = e-' a ( p - x Q),, (7)

where the superscript I denotes the first iteration. This function is now considered as the
objective function of the system and will be minimized with respect to the damping parameters
cl, c2, ..., c,. It can be written as

F(w, c) = min max [XI]2. (8)

C1 t~ 1

The original primitive function is no longer a maximum with respect to the argument 091
since the original damping values have been changed. Hence it must now be maximized to
form a new primitive function with respect to a new argument 092.
The iterations will be repeated until an optimum value of the objective function is deter-
mined, which may be expressed as

IXo~,l" = min max I x . r .

Cn ~Pi
(9)
 OPTIMIZATION OF VIBRATION ISOLATION 107

3. MAXIMIZATION OF A FUNCTION WITH A SINGLE VARIABLE

It was decided to use the Newton-Raphson method for a single parameter maximization
due to its simplicity and effectiveness. In this context, the right side of equation (6) can be
regarded as a function of some variable, u (actually u corresponds to to in the vibration
system):
ix12= ( e - , Q)(p-I Q), = F(u). (10)

When F(tt) becomes a maximum, the derivatives o f F ( t 0 with respect to u will vanish. The
problem is now to find a solution o f it = Umaxwhich makes X = 2", .... or at least an approxima2
tion ofttm~x which is sufficiently accurate for some given purpose. One approach is to linearize
F'(u), if it is differentiable. Thus,

F'(u) = F'(u ~ + F"(u ~ (u - u ~ + R(u, u~ (11)

where R ( u , u ~ = higher order derivatives. I f u ~ u . . . . F'(u)--+ 0; then by neglecting the

presumably extremely small quantity R(u, u~ the linear equation

F'(u ~ + F"(u ~ (u - u ~ = 0 (12)

is obtained. This equation will have a unique solution, td, ifF"(u~ -1 exists. It can be written as

u I = uo - [F-(uO)]-1 F,(uO). (13)

The concept of iteration proceeds from the idea that ifu ~ is close to u . . . . then u I is possibly
closer and the foregoing procedure may be repeated with u ~ replaced by u ~to obtain a succes-
sive approximation u 2 to um~xand so on. If F"(uJ) -~ exists f o r j = 0, 1, 2 . . . . j, then

td +1 = uS - [F"(us)] -1F'(uJ). (14)

The iterative process represented by equation (14) is called the Newton-Raphson method.
It is noted that the variable u is a real valued number and thus the first derivative, F'(tt),
must also be a real value. By mathematical manipulation, the first and second derivatives of
equation (10) become

~' 0 + 2p ~ t , - ~ P - Q (p-, Q)* +

(
+ p _ OP
g~u P -
iQIp_~_g__p_~Q
/ ou
o, ).]. (16)

3.1. CONVERGENCE
The convergence in the iterative process for equation (14) is ensured by introducing a
scalar parameter, 2, in the following form:
uJ +l = uJ - ;.[F"(uO] -l F'(uO = u J - 2h, (17)
where 0 < 2. ~ 1. This step size, 2, is determined by an iterative method such that either
010, u0 -- 0, (I 8a)
108 C. NG AND P. F. CUNNIFF

or

02(2, u0 >/0, (18b)

where
0,0., uO = F(uJ + 2h) - F ( u O - 2(1 - a) (hF'),
0 2 0 , uO = F(u1 + ).h) - F ( u O - 2 a ( h F ' ) , (19)
in which
a = scalar quantity = 0.4 as suggested in reference [7],
h = [F"(uO] - l F'(uO.

A first trial value of 2 = I is assumed in equations (19). The resulting values of 01 and 0.
are tested according to equations (I 8). If the test fails, a smaller value of 2 is tried, say 2 = 0.9,
etc.

4. MINIMIZATION OF A FUNCTION WITH SEVERAL PARAMETERS

The gradient methods seem more suitable for minimization of multi-variables than the
differentiation methods. Their rate of convergence is faster as compared to the direct search
method and they require only the first partial derivatives, while the Newton-Raphson
method requires the second derivatives.
The first gradient method was introduced by Cauchy in 1847 as the so-called "steepest
descent method" [8]. A particularly clear and attractive exposition of the method was given
by Courant [9] in a 1941 address to the American Mathematical Society. Since then a number
of techniques have been proposed which include the "differential gradiefit method" by
Arrow [10], the "conjugate gradient method" by Hestness and Stiefel [11], the "gradient
projection method" by Rosen [12], the "optimum gradient method" by Bromberg [13] and
the "variable metric method" by Davidon [14]. There is one disadvantage common to the
methods: namely the problem ofconvergence in iteration. For instance, it is stated in reference
[I 5] that it took thousands of steps to give a satisfactory approximation in one example. By
carefully studying the rate of convergence among these methods, the variable metric method
proved to be superior to the other methods.
This method was first proposed by W. C. Davidon in 1959. Since then, there have been
quite a few papers published [16-20] either to suggest some modifications about the original
technique or to inform others of their experiences in using the technique. Among those
papers, the modification made by Fletcher and Powell has proved to be very efficient in
practical experiences. The Davidon-Fletcher-Powell method is briefly described below [21].
Consider equation (10) as a function of the control variable v (actually the damping terms
in the vibrating system). This function can be represented in the form of a Taylor series
expansion for an n-dimensional function as follows:
F(v) = Fo + b r v + v r G v / 2 , (20)
in which
Fo = initial value of F,
b= constant vector,
v= control vector variables,
G= positive definite matrix of elements equal to a 2 F[(avt avj).
The derivative of equation (20) is the gradient, g, so that
g = F'(v) = b + Go. (21)
 OPTIMIZATIONOF VIBRATIONISOLATION 109
AS seen from equation (14), the corresponding increment o f v for iteration is - G - I g , which
is difficult to evaluate numerically with multi-variables. In the present method, the matrix G -1
is not to be evaluated directly. Instead, an arbitrary positive definite matrix, H ~ which is the
so-called variable metric, is chosen, usually as an identity matrix. Let v~ be the initial point and
s t = - H l g I, (22)

a I = )'s l, (23)

v l * l = v I + a l, i = 0 , I, 2 . . . . . (24)

where ? is chosen to minimize F(v) along the search direction s t at current v I. I f H I is positive
definite, ? must be greater than zero unless v I is a m i n i m u m ofF(v). I f F LDrepresents the lower
bound of an estimated o p t i m u m function o f F(v), ~, should be chosen between the following
two values, whichever is less:
f 2(F LB - F(v))]
~= min [1, (-gl~ "J" (25)
L e t f l = gl+l _ gl. T h e new approximation of the inverse of the matrix of second derivatives
of F(v) is updated recursively as suggested in reference [17]:

tri(tri) r H i ( y i ) ( y l ) r I-1i
H TM = Hi+ - - (26)
(al)ryi (yt)r H l y i

It is shown that H I+~ is positive definite if H t is positive definite. It ensures that F(v) is
decreased on each move. By Taylor's expansion, it can also be shown that t h e vector
a ~ a 1. . . . . tr n is orthogonal to the gradient g~+L Hence for a non-zero vector tr, the gradient g
must be identically zero when b is at its m i n i m u m . At that point, the generated matrix, H,
will eventually be identical to G -1.

5. E X P E R I M E N T A L TEST
Figure 1 is a model which was tested experimentally to verify the preceding analytical
treatment o f the optimization process. A foundation is modeled as a rigid b e a m of mass m2
and m o m e n t of inertia J2 supported by two massless damped elastic supports at its two ends.
A hypothetical engine is modeled as a lumped mass m: located at the center o f the beam.

/ / / / ,i! r lilI/I/tl I

& ki

9 12 -F II

Figure l. A three degrees of freedom model for test.

I I0 C. NG AND P. F. CUNNIFF
A periodic exciting force is applied at the same point to produce the motion which would be
generated by the engine over a wide range of frequencies. A lumped mass block m3 represents
an instrument which is mounted on a massless isolator consisting of a spring and a dashpot
which, in turn, is attached to the beam. The problem is to minimize the vibration of the
instrument resulting from the operation of the engine.
The equations ofrnotion in non-dimensional form for the system in Figure I are as follows:

M,~ + C 2 + K 2 = Q. (27)

(See Appendix II for the definitions of the M, K, and C matrices.)

The mass ratio, p = m2[ml, was taken from a design example in reference [22] where
p ~ 5/19. Due to the limitation of available commercial dampers, it was decided to seek an
optimal solution with two identical dampers at both supports. With the aid of the computer
program, several schemes were tried by varying the values offl, tl, 4, and v. Finally, a solution
was obtained with the values
fl = 0.5, t / = 1-25, ~ = 0.25, v = 0. l 0, i t = 1.45
and with the following optimized damping constants for the three dampers:

(z = (2 = 0.283; (3 = 1.928.

Four viscous dampers (two respectively identical pairs) and one hydraulic damper were
obtained commercially. One set of viscous dampers had a value of ( = 0.283 while the other
set had a value of ( = 0.24 for the test configuration. Both sets of identical viscous dampers
were calibrated and it was found that the force varied linearly with velocity between 0.2 inches
per second and 2.5 inches per second, the range planned to be used during the test. The damping
constants of the hydraulic damper can be varied by turning a knob with a special wrench.
The optimized damping was determined and the particular oriented position of the knob was
marked.

5.1. TESTSET-UP AND INSTRUMENTATION

Figure 2 shows the general arrangement of the test set-up while the photograph in Plate l
shows the actual configuration. A 16 in x 281 in x 88in aluminum beam of mass m2, shown
in Plate l, served as the foundation with one steel tape of ,10.83 in x 88in and one viscous
damper attached to each end as the elastic supports. The upper end of each steel tape was
connected to an 18 in deep steel I-beam, and the lower was bolted to the concrete floor. The
tapes were slightly tensioned in order to maintain straightness. At the mid-height of the tapes,
bolts were used to connect the tapes with the aluminum beam. Since the tape was flexible, it
could resist effectively only in tension.
A steel rod of 8 ~ in • 121 in x 189 in, which acted as the mass of a diesel engine, rex, was
bolted to the beam at its midspan. The lower end of the steel rod was bolted to the top of the
shaker. The threaded bolt at the upper end was adjusted so that there would be no initial
stresses induced in the suspended armature flexures (see Detail A in Figure 2) [23]. The weight
of the impedance head and the moving element of the shaker were included as part of ml.
A 4 in • 2 in x a4 in aluminum block of mass m3 served as the mock instrument to b e
isolated. It was connected to the beam by means of a sandwiched spring and a hydraulic
damper. The sandwiched spring was made of two 88in • 0.004 in steel tapes as facing with a
88in thick styrofoam core and was 9.07 inches in length. The modulus of elasticity of the core
is only 3000 lb/in 2, which is negligible in comparison with that of the steel tape, 29.1 • 106
lb/in 2. The distances of the sandwiched spring and the damper from the centerline of ma
were arranged in such a way that at peak response of the optimized frequency, the moments
 OPTIMIZATION OF VIBRATION ISOLATION 11 ]

t
Lock I~ate ~ . . ~ _ . ~
a.cormature ~
coil ,~"-.'-.'-.'-.'-.~Gn~l~11 \
.
d.c. armature ~
I'---~lllluIII,

Armature
flexure
Detail A-
moving element
inside 1he shaker

Figure 2. Test set tip. Legend: I/Viscous damper, ~i; 2, viscous damper, ~2; 3, hydraulic damper, ~'3;
4, sandwich spring, ks; 5, J in x 0.012 in steel tape, kl; 6, 88in • 00125 in steel tape, k2; al, a2, as, accelero-
meters; F, force gage.

Phase
Zero drive angle Switch Volt
amplifier Band filters meter box meter Oscilloscope

9 L i i n e ~ Oscillograph

Shaker- ~ Power supply

Pre- Frequency Variable
amplifier counter osdllofor
Figure 3. Schematic diagram of instrumentation.
Plate 1. Photograph of actual set up. (a) With two viscous dampers; (b) with four viscous dampers.

(facingp. 1 I0)
112 c. NG AND P. F. CUNNIFF
resulted from the forces of the spring and the damper about the centerline would approxi-
mately cancel each other. However, there was some small error induced to the accelerometer
labeled aa in Figure 2 due to the rotation caused by the unbalanced moment for other
frequencies. For example, at 600 H z the maximum error was estimated to be less than 6%.
Since the responses of the system at high frequencies were far less critical, the error may be
considered as insignificant. The system arranged in this way gives coo equal to 100 Hz.
The instrumentation is depicted schematically in Figure 3.

5.2. TEST PROCEDURE

The purpose of this test was to seek experimental support of the optimized damping
constants obtained analytically. Table 1 lists the testing sequence with various combinations
of viscous dampers attached to the beam ends. The hydraulic damper connected to block ma
can be adjusted such that the response of the system becomes a minimum.

TABLE 1
Test series for various damph~gfactors
Damping constants
Test 9 A
No. (, (2 (3
I Without damping Without damping 2"27
2 ((o)1 ((0)2 1.928
3 ((,)~ ((,)2 1"98
4 ((, + (o)1 ((, + (o)z 1"78 "

(~'o), = ((o), = 0-283, ((,)t = ((,)2 = 0-24, ((, + (o) = 0.283 + 0.24 = 0.523.
During each test, the damping constant G is adjusted such that the response becomes a saddle point
(min.-max.) with respect to frequency.

The procedure of each test required that the force amplitude be constant throughout.
Initially, the hydraulic damper was set equal to 1-928. F r o m an initial frequency of 145 Hz,
the frequency was varied to find a maximum response of the accelerometer aa. At this frequency
the hydraulic damper was varied to minimize the response. With this new hydraulic damping
value, the frequency was again varied to seek a new maximum response. At this frequency, a
new hydraulic damper setting was found to minimize the response. This procedure was
repeated until a min.-max, response was found. Other than test 2, it normally required five
or six trials.
Once the value of (a (see Table 1) had been found, the system response was found by
sweeping over a frequency range from 40 Hz to 650 Hz.

6. COMPARISON BETWEEN PREDICTED AND TEST RESULTS

The normalized displacement of ma can be evaluated analytically by
I.(3[2=(~ ParQr)(~ Pa, Qr)*, r = 1,2,3, (28)
where Par = elements in matrix P as defined in equation (3), and Q, = Q~ = applied
excitation.
Since the elements of masses, damping and spring constants contained in the matrix P and
the driving force, Q, are known, the normalized displacement, X3, at various frequencies can
be computed readily. A computer program was written to calculate these normalized displace-
ments at various frequencies. The computer time required was approximately 16 seconds on
 OPTIMIZATION OF VIBRATION ISOLATION l 13

i I J I i I ~ I i i ' I II
50

Test I ~ i c t ' i o n
25

"~ 2-0

~ I'0

Test 2 Ixedlction
~ (optimized)
05

i I i I i I ! I : ~ ~ I
04 0"6 08 I 2 :5 4 .5 6

Frequency rotio, ~ / w o

Figure 4. Response of m3 versus frequency for tests 1 and 2. zx, Test I result; e, test 2 result.

0 . 6 1~ ~ | ' I ' I i I ' I ' I ' I 'I

Test 3 prediction J3~
| \ ~-_~.. _ ~ - T t ' s t 4 prediction

- o-5~(" J ~ r ' - * - ' F -~ ~ |i

~----~c 0 ' 4 \~ ./Test 2 predictiod

",~ (optimiZed) ,.
o

~ 0.2

o., \'~
, , , , l l , , 1 , ,~,~
04 0"6 08 I 2 :3 4 5 6
Frequency rote, u/w o

Figure 5. Response ofma versus frequency for tests 2, 3 and 4. e , Test 2 result; 1:3, test 3 result; ~, test 4
result.

the CDC Computer 6600 for each set of results. These predictions are plotted in Figures 4
and 5 for the normalized displacement of m3: i.e., ~a v e r s u s frequency ratio oJ/(oo. The test
results are also marked in the figures along with the corresponding predictions. It is seen from
these figures that the maximum response value was a minimum in test number 2 which was
the predicted optimized result.
The evaluation of the overall experimental results for each test based on the prediction
can be indicated by the so-called root-mean-square deviation, which is defined as

/ y (A'g - A'~')~
' (29)
114 C. N G A N D P. F. CUNN1FF

I i I i I t i * I
5"0 ~ Test I

,,~ 2.0 -

I.O Test 3
0.9

"~ 0-7
~ 0.6
~ 0.5
Test 4

I 1 I T I I I I I
0 0.1 0-2 0"3 0.4 0-5

Dampingc o n s t a n l s , ~l=~Z

Figure 6. Comparisonbetweenpredicted and measured response of m3 for varyingdampingconstants.

, Predicted; I--),measured.

where the superscripts p and m denote the predicted and measured values, respectively.
The data reveal that the root-mean-square deviation of each test is relatively small, the
largest deviation being only 2-8 ~o, which indicates that the test results are in good agreement
with the predicted results.
The curve in Figure 6, which is plotted for the predicted maximum normalized displace-
ment of m3, )?3 v e r s u s (1 = (2, was obtained with the aid of the computer program. Approxi-
mately 43 seconds was required on the CDC Computer 6600. Corresponding measured
values from Figures 4 and 5 are also plotted in the same figure for comparison. They are fairly
close to the predicted curve and both show the response as being a m!nimum at the optimized
damping values.
Similar tests have also been performed under the same procedure with other combinations
of the four viscous dampers. Those results are not presented here. However, maximum
responses of those tests exceeded the optimized one.

7. CONCLUDING REMARKS
A numerical iterative method has been developed to obtain the optimized response and the
corresponding damping values for vibration isolation systems with multi-degrees of freedom.
The method incorporated the modified Newton-Raphson method for the maximization of
functions with a single variable and the Davidon-Fletcher-Powell method for the minimiza-
tion of several variables.
An experimental investigation of a mechanical vibration isolation system has been
performed which provided close agreement between measured and predicted response. This
experiment demonstrated that the proposed optimization technique indeed predicts the
smallest response ofthe critical frequency. This was performed experimentally by comparing
the system responses at their respective critical frequencies for various combinations of the
damping devices.
According to the authors' experiences, less computer ~ime was taken if the assumed initial
values of damping were smaller than the evaluated optimum damping constant instead of
larger. It may also be worth noting that, in another ease, it took approximately 96 seconds
in computing time for a system of 7 degrees of freedom on computer UNWAC1108.
 OPTIMIZATIONOF VIBRATIONISOLATION I 15

ACKNOWLEDGMENTS
This paper is part of a thesis submitted to the Department of Mechanical Engineering,
University of Maryland, in partial fulfilment of the requirements for the Degree o f Doctor of
Philosophy,
The computer time for the project was supported by the National Aeronautical and Space
Administration, Grant NsG-398 to the Computer Science Center of the University of
Maryland.
Assistance provided by the Naval Ship Research and Development Center, Carderock,-
Maryland, for the experimental program is recognized.

REFERENCES
I. J. ORMONDROYDand J. P. DEN HARTOG1928 Transactionsof the American Society of Mechanical
Enghwers 50, A9-22. Theory of the dynamic vibration absorber.
2. F. M. LEwis 1954 American Society of Mechanical Enghteers Paper 55-APM-3. The extended
theory of the viscous vibration damper.
3. T. LIBERand E. SEVIN1966 Bulletin of the Shock attd Vibration bz.formation Center 35, 203-215.
Optimal shock isolation synthesis.
4. J. A. BENESHOand J. G. BOLLINGER1966Proceedings of the 7th International MTDR Conference,
Un&ersity of Birmbtgham, 229-241. Theory and design of a self-optimizing damper.
5. E. SEv•N and W. D. PlLKEY•968 S•ciety •f Attt•m•tive Engineers• Aer•nautic and Space Eng•wer-
hzg andManufacturhzg Meeting, Paper 687049. Optimization of shock isolation systems.
6. D. E. ZEIDLERand D. A. FnOHnIB1972 Bulletbt ofthe Shock and Vibrationhtformation Center 42,
77-83. Optimization of a combined Ruzicka and Snowdon vibration isolation system.
7. E. POLAK1971 ComputationalMethods in Optimization. New York: Academic Press.
8. A. CAUCHY1847 Comptes Rendues de i'Acad~mie des Sciences 25, 536-538. M6thode g6ndrale
pour la resolution des syst~mes d'6quations simultan~es. "
9. R. COffnANT1943 Bulletin ofthe American Atathematical Society 49, 1-23. Variational methods
for the solution of problems of equilibrium and vibrations.
10. K. J. Annow 1951 RAND Corporation, Santa Monica, California, Paper P-223. A gradient
method for approximating saddle points and constrained maxima.
11. M. R. Hrs'rNrss and E. STIEFEL1952 U.S. NationalBureau of Standards Report 1649. Methods
of conjugate gradients for solving linear systems.
12. J. B. ROSEN 1960 Journal of the Society of hldastrial and Applied Mathematics 8, 181-217. The
gradient projection method of non-linear programming. Part I: Linear constraints.
13. N. S. BROMnERG1961 Transactions of the American htstitute of Electrical Enghwers 80, 725-730.
Maximization and minimization of complicated multivariable functions.
14. W. C. DAVIDON 1959 Argonne National Laboratory R and D Report ANL-5990 (Revised).
Variable metric method for minimization.
15. A. V. FIACCOand G. P. MCCORr,tlCK 1963 Research Analysis Corporation Technical Paper
RAC-TP-96. Programming under non-linear constraints by unconstrained minimization.
16. H.A. SPAN6III 1962 Society for lndustrial and Applied Mathematics Review 4, 343-365. A review
of minimization techniques for non-linear functions.
17. R. FLETCHERand M. J. D. POWELL1963 Computer Journal ofthe British Computer Society 6,
163-168. A rapid convergence descent method for minimization.
18. G. W. STEWART1967 Journal of the Associationfor Computing Machinery 14, 72-83. A modifica-
tion of Davidon minimization method to accept difference approximation of derivatives.
19. G. E. MEYERS 1968 Jottrnal of Optimization Theory and Applications 2, 209-219. Properties of
the conjugate gradient and Davidon methods.
20. D. GOLDFARB1969 Journal of Applied Mathematics, Society for htdustrial and Applied Mathe-
matics 17, 739-764. Extension of Davidon's variable metric method to maximization under
linear inequality constraints.
21. R. L. Fox 1971 Optimization Methods for Engineering Design. Reading, Massachusetts:
Addison-Wesley Publishing Company.
116 c. NG AND P. F. CUNNIFF
22. ~ A~ I ~ v ~ ~965J~urna~~f the Envir~nmenta~Sciences ~nstitute~ 3~-32. A meth~d ~fincreasing
vibration exciter capability.
23. L. YAMAKAWAand S. TAKEDA 1960 Journal of Japanese Society of Mechanical Engineering 63,
234-239. Experiments on elastically supported damper systems.

APPENDIX I
LIST OF SYMBOLS

a
scale parameter as defined in equation 0 9 )
B
constant matrix
b
constant vector
C
damping matrix
c
damping factor
F
function (vector)
F',F" first and second derivatives o f function F
G matrix of O'F/av~Ovj
g gradient
H Hessian matrix
h increment
i V'-I
K stiffness matrix
k spring constant
P matrix as defined in equation (4)
Q applied excitation vector
s search direction
u variable
V variable vector
X displacement vector or response function
.~', )(, derivatives o f X with respect to time
y difference of gradients
a slope as defined in equation (23)
~. step size parameter as defined in equation (18)
damping constant
09 driving frequency
/7~ decision variable as defined in equation (18)
Superscripts
i, j, k number of iteration
0 initial value
T transpose of a matrix
* conjugate
LB lower bound
Subscripts
ij ith row a n d j t h column of a matrix
i, j, k order of an array of elements

A P P E N D I X II
MATRIX NOTATION

T h e inertia m a t r i x in equation (27) is

1
0 0
l+v
M= o ,1 o
\- l

0 0 =
4
 OPTIMIZATIONOF VIBRATIONISOLATION 117
and the stiffness matrix is

For computer programming, the damping matrix can be expressed as

C = Bt ~t + B2 (2 + Be (3,
where

BI= ~ 2/= z ,

B2 = 2 fl 2~/fl 2 ,

0
-2~

Be = -2r 2r 24 .

-2r 2r 2r

The parameters concerned are defined as follows:

X, normalized displacement vector (dimensionless)
,i', 37, normalized derivatives of X with respect to time (dimensionless)
= = ldll
,8 = l z / l l
tl = k2/kl
= ka/kl
p = m2/ml
v = rod(m1 + m2)
= J2/(m217)
9/2, mass moment of inertia of beam m2
G = (COo/2ki)c,
p co/%
=

COo= V k d ( m l + m2 + me).