Journal of Sound and Vibration (1987) 112(1), 23-30

DYNAMICS OF MACHINE F O U N D A T I O N S WITH

R A N D O M PARAMETERS

W. MIRONOWICZ AND P. SNIADY

Institute of Civil Engineering, Technical University of Wroclaw, 50-370 Wroclaw Poland

( Receired 16 November 1985)

Solution of the problem of free and forced vibration of a discrete system with random
parameters, as a model of a machine foundation, is considered. Expressions for the
expected values and correlation functions of the solution are formulated, in terms of the
probabilistic characteristics of the structure, initial conditions and excitation, which are
assumed to be known. The stochastic linearization method is used. The solution is illustrated
by an example of a block foundation with random parameters.

1. INTRODUCTION
Vibration problems o f structures in which random factors are involved have been exten-
sively studied, with the randomness being most often assumed to be introduced by the
Ioadings [1-4]. However, in a few papers the random character of structural parameters
has been taken into account [5, 6]. The sensitivity o f the system dynamic response to
changes in its parameters depends on the kind of load. It is less in the case of random
wide band excitation and greater for deterministic harmonic excitation, as it is then
possible for resonance to occur [7].
The parameters o f a machine foundation are often subject to random variations, of
technological, operational and structural causes. Foundation dynamic analysis thus can
be improved by taking them into account.
In what follows the vibration of a discrete system with random parameters is considered
as a model of the machine foundation. General solutions are given for the eigenvalue
problem, and the free and forced vibration problems. Expressions are derived for the
expected values and correlation functions, it being assumed that the appropriate prob-
abilistic characteristics of the system as well as the initial conditions and excitation are
known. The first order probabilistic method is used [8].

2. PROBLEM FORMULATION
The problem to be considered is that o f the vibration of a discrete system of n degrees
of freedom, governed by the equation
B?j(t) + C~t(t) + Kq(t) = f ( t ) , (2.1)
where q(t) is the vector of generalized co-ordinates, f ( t ) is the random excitation vector,
and B, C and K are random matrices of inertia, stiffness and damping. It is assumed that
C =/.tB + KK, (2.2)
with ~ and ,r non-negative real constants. The probabilistic characteristics, such as the
expected values and the covariance functions of the matrix constituents B, K, C and the
23
0022-460x/87/010023+05 s03.00/0 (~ 1987 Academic Press Inc. (London) Limited
24 w. M I R O N O W I C Z AND P. S N I A D Y

vectorf(t) are assumed to be known. The random elements of equation (2.1) are expressed
as sums of their expected values and the random fluctuations:
B=/~+B, C = C'+ C', K =/~+/~,
f(t) =J~(t) + f ( t ) , q(t) = 4(t) + @(t). (2.3)
Here the expected values are denoted as follows:
E[B] =/~, E[C] = C, E[K] = / ( , E[f(t)] =.f(t), E[q(t)] = t~(t). (2.4)
The random fluctuations of the constituent parts of the matrices B, C, K and the vector
q(t) are assumed to be small enough relative to their expected values for the first order
probabilistic method [8] to be a good approximation of the solution. In this case the
product of the random values X and Y has the form

XY = 9(~'+ X Y + 9~9. (2.5)

Formula (2.5) can be considered as an expansion of the p r o d u c t X Y in a double Taylor's
series, about the expected values 9(, 'r', in the fluctuations X, Y, with only the first two
terms of the expansion being retained. This is similar to the immediate linearization
method [4] applied to non-linear problems.
After expression (2.3) have been substituted in equation (2.1) and the approximation
(2.5) has been used one obtains

(2.6)
For the eigenvalue problem equation (2.1) is transformed to the form

( K - t a 2 B ) q = 0. (2.7)

Hence, after approximation (2.5) has been used, one obtains

(/~ - A/~)c~+ (/~ - A/}- ~/~)4 + (/~ - A/~)t~ =0, (2.8)

where to2 = Z = A + %,, E[~] = 0.

Equations (2.6) and (2.8) can be solved to give expressions for the probabilistic
characteristics of the displacements and frequencies of the system.

3. SYSTEM NATURAL FREQUENCIES, FREE VIBRATION

Performing the expected value operation on equation (2.8) gives a deterministic relation,

(/( - A,~)~ = 0, (3.1)

from which eigenvalues Ai ( i = 1 , 2 , . . . , n) and the normal mode vector t~= ~,~ can be
determined by using well known methods. These vectors can be used to construct a normal
mode matrix. Taking into account expression (3.1) in equation (2.8) where the unknown
factors ~, t~ occur, and after left-hand premultiplication by the vectors ~T (i = 1, 2 , . . . , n)
in order to eliminate the vector q, one obtains the vector of random fluctuations of the
eigenvalues,
= {/~-'}( l~/T/~IV - I~VT/~IV{Ai}), (3.2)

where {/;,} = l~Vr/~l~/. Symbols (')T and {. } denote the transposed vector or matrix and
the diagonal matrix, respectively.
 RANDOM PARAMETER MACIIINE FOUNDATIONS 25

By making use of equation (3.2) the correlation matrix of the eigenvalues A, ( i =

1 , 2 , . . . , n) is obtained in the form
K ~ = E[AA"r] = {/~-'} I~'TE[(/(~V--/~IV{A,})( I~T/~T__ {~j} i~,'r/~T)] l~{/~-t}. (3.3)
In the particular case when/~ = a/~,/( = ),/(, where the parameters a and Y are random
variables, one obtains, from formula (3.3),
o2
Kaa = {A,}E[(a - 7)2]. (3.4)
In this case the matrix Kxx is diagonal, from which it follows that correlation between
different eigenvalues does not exist.
The expected value and the variance of the natural frequencies are approximately
governed by the formulae [8]
2
o~, ~ ff-~i, o',o = (114,~,)o']. (3.5)
The variance cry, is an element on the diagonal in the ith line of the matrix Kax.
Given the fluctuations of the eigenvalues (3.2), the corresponding fluctuations of the
normal mode vectors can be calculated from formula (2.8).
Next the problem of free vibrations of the system can be considered, with damping
taken into account. It is assumed that for t = 0 the initial random conditions are
q(O)=d=t*l+d, 4(0) = e = ~+,~. (3.6)
Assuming](,) = f ( , ) = 0 and performing the expected value operation one obtains, from
equation (2.6),
/~c~'(t) + C'c~(t) +/(c~(t) = 0. (3.7)

This is a deterministic, homogeneous set of differential equations satisfying the deter-

ministic initial conditions
4(0) = d, d(0) = ~. (3.8)

Random fluctuations of the vector of generalized co-ordinates q(t) satisfy the equation

/~'(t) + C'~(t) +/(t~(t) = - Bt~'(t) - C'q(t) - / ( 4 (t), (3.9)

and the random initial conditions

,~(0) = d, ,~(0) = ~. (3.10)

It is convenient to solve equations (3.7) and (3.9) with the initial conditions (3.8) and
(3.10) being transferred to the expected values of the normal co-ordinates:

~(t)= I~')*(t), t~(t)= l~,'f(t), d= l,~'y(0)= lh,, e= i~,')(0)= ;~'v. (3.11)

One thus obtains
f t3, clio

2(t) = - ~'T~li'
I2 /t~,(t- ~')};"(T) d~ - ~'Tdl~'
Io {t~,(t -- ~')}~(~') d~

fo f 3, &~,l (3.12)
26 W. M I R O N O W I C Z A N D P. .SNIADY

where
f,~(t) = e -a,' sin .f'/,t, f~(t) = e -'~,' cos .0,t,
.0,~=,~-a, ~, a,=~,/2~,, :,~=:~,1:,,,

{h,(t)} = {(1/8,.O,) e -a,' sin .0:}. (3.13)

As can be seen from equation (3.12) the random fluctuations of the normal co-ordinates
are the sums of the fluctuations due to the randomness of the system parameters and the
randomness of the initial conditions.
Making use of equation (3.12) one can determine the correlation function of the normal
co-ordinates from the formula
K~r(tl, t2) = E[37(t1))~T(t2)], (3.14)
and then the vector of expected values and the correlation matrix of the normal co-
ordinates,
,i(t) = W~,(t),
~ Kqq(t,, t2) = E[q(tl)#T(t2)] = l:VKyy(t,, t2) fiT. (3.15)

4. SYSTEM FORCED VIBRATION

For forced vibration the system to be considered is that described by equation (2.1)
with the initial conditions
q(0) = 0, 8(0) = 0. (4.I)
From equation (2.6) one obtains expressions for the expected value and the random
fluctuations of the vector q(t) in the forms
~:i(t)+ C:1(t)+ :c:1(t)=](t),
/~c~(t)+ (~c~(t)+/~c~(t)=:(t) -/~'- 6"c~(t)-/~c](t). (4.2)
Applying the normal mode approach (3.11)to equations (4.2)gives

:(t) =
Io {1;,(t-~-)}~:.:0) tiT,

.~(0 =
Io{~,(t--r)}I~'T](,)dr
-
Io{h',(t--~')I(~'TN~'}(T)+ I~TcI'v~(7)+ I,VT/~I,V~(,))dr. (4.3)

By using the second of equations (4.3) in formula (3.14) one can determine the correlation
function. With the excitation v e c t o r f ( t ) assumed to be independent of the system random
characteristics B, C, K one obtains

K,.r(t,, t2) = io'io'o{h,(q- r,)} l'~'TK/r(r,, z2) 14:{i*l)(tz- rz)} dr, dr2

- fo" fo2{h~(t,-r,)}fV-~E[R(~',)R'r(~'2)]fVIt~(t2-~',)}
d~',dr2, (4.4)

where R(t) =/~g'.~'(t) + CI:V}(t)+ ~I:Vf~(t).

 RANI)OM PARAMETER IqACilINE FOUNDATIONS 27
Similarly, for the generalized co-ordinates one obtains, respectively,

4(t) = l~':(t) =
fo~(t-~-)/(,) d~-,

q(t) =
Io /4(t- r)f(r) dr-
Io I?t(t- r)R(r) dr, (4.5)

and
Kqq(ll ' 12) = l~Ksy(ll, t2) ~T

=Io"Io'*l?f(ti-ri)Klr(ti,t2)fflX(t,-h)dridh

-Io"~2l~l(t,-'r,)E[R(r,)[((h)llZl(t2-h)dr, dh, (4.6)

where H(t) = IV{/~i(t)} IV r is the matrix of impulse response functions.

For the case of constant frequency due to harmonic excitation, which is important in
practical applications, the excitation vector can be taken to be of the form
f(t) =f(t)+f(t) = d sin pt + d sin pt. (4.7)
Assuming the solution in the form
~(t)=~ssinpt+~ccospt, ~(t)=~,sinpt+~ccospt, (4.8)
and substituting this into equation (4.2) gives a set of algebraic equations for the expected
values,
( Ko - / ~ ) q ,o *" = ~,
- pCqr pCq,
o
*" + ( l< - p : B)q<
* =0, (4.9)
and random fluctuations,

pCtl, + (/< -p2B)tT< = -pCti, - (I< - p2/~) ~e. (4.10)

After the equation sets (4.9) and (4.10) have been solved one obtains expected values of
the generalized co-ordinates and their random fluctuations from which the correlation
matrix can be determined.
The presented theoretical solutions are going to be illustrated by the exemplary analysis
of the block foundation of random parameters.

5. BLOCK FOUNDATION OF RANDOM PARAMETERS

An example of the analysis can be provided by the vibration of a block foundation
(see Figure 1) in the plane xz, for which

i ~ ~ ~k~A~ ~ A ~ k~J~ ~ (5.1)

~=tg'~l--L-~,= L= i,
Here rh is the mass of the block, -~x.- and S~.z are the products of inertia of the block
mass, iy is the moment of inertia of the block mass, ,4 is the area of the block base, Jr
is the moment of inertia of the block base, and/~,,/~x a n d / ~ are the base stitinesses.
28 W. M I R O N O W I C Z AND P. S N I A D Y

Z
2

Jr(t)

2p
t 2

I 4 I O,y 4 t 2

Figure 1. B l o c k f o u n d a t i o n .

The system random fluctuations are assumed to be given in the form of matrices B, K
of the same structure as those of equations (5.1). With the notation
0 = ~,~,T, 0 ' = ~'(a,} ~,T, 0" = ~'{a,}{a,} IP,
the correlation matrix (3.3) becomes

The elements r, s of the constituent matrices (5.2) are given by the formulae
3
E[(/(C//(T),] = /~,,E[/~,,/~,,], E[(/(CP/~T),,] = E O{,E[/~,,/~I,],
I=1

3 3 3
E[(/3OI/(T),] = E 01,E[/~,f,i], E[(/~/~/0/~T),~]= Y~ 2 0~,E[b,,bh,], (5.3)
I-I I=1 h = l

where the/~'s and /~'s are elements of the matrices /~,/~.

As an example, free vibration at the block foundation, as shown in Figure 1, of the
base random stiffness
g = Tx,i, (5.4)
has been considered. The correlation between the r, s elements of the matrix (5.4) is
assumed to be of the form

K,~ = E[ k,.k~,] = IA I ~Z,, exp (-~,x2- x,D cos , ( x 2 - x,) dxt dx2 , (5.5)

2
where or,, ~', and ,f are constants.
Figures 2(a) and (b) show the effect of the variations in the parameters ~', ~r on the
correlation function (5.5) and consequently on the correlation matrix (5.2).
In machine foundation design the ratio oJ'p is important because of the transitory
resonance problem. It is convenient to introduce the resonance index given by
/3 = (c3 -p)/,,/crL + cry, (5.6)
which is the analog of the reliability index for structural reliability problems. Here o~ and
o'~ are the natural frequency and its variance, and p and o-p2 are the frequency of excitation
and its variance, respectively. This index can be treated as the randomness measure at
the first-order level of the transitory resonance problem.
Figure 3 presents the variation of the index /3 =/3(~', ~:) for the foundation under
consideration; here o-~,=/~x =0, o'~, = o']3 and lack of correlation between/~ and/~y have
been assumed.
 RANDOM PARAMETER MACHINE FOUNDATIONS 29

(a)
IO0

8O
(=0

40

I I T 1 I
05 I 1.5 2 2'5

,oo~..~o

oo l- -\

20 "

0.2 04 06 08 1.0

Figure 2. Correlation function (a) K,, = K,,(~) and (b) K,, = K,,(,~).

250

, ~ 200
*3
150

6- l o o

50

015 10 1 5 2 0 2 5 310

Figure 3. Resonance i n d e x / 3 = fl(~').

The forced vibration of the foundation in question has also been considered. For C = 0,
p = 2500 kg m -3,/~ = 3 k , / ~ =2k,/~. = 7k, k = 2 x 107 Nm -3, p = 6.32 s -I, ~"= 0.5, ~: = 0.25,
2 --
or,, - o" = constants, ao = [ 1,0, 2]Va, and the standard deviation of random fluctuation (4.7)
equal to 0. I a, the results are
o
~, = [559"5 1.29 57"3]Tx 10-t2a, qc = 0,

E[qqT] =
[3,30.4 72 32,.2]
7-2 0.02 0.7[+
"2473"9
10"9
10"9
.3"0
108"01 \
0.5[0"2 / x 10-24a 2.
1 #
321.2 0.7 32.9_1 108.0 0.5 4.9J ]
30 w. M I R O N O W I C Z A N D P. S N I A D Y

6. CONCLUSIONS
The dynamic response o f a structure having parameters with small random fluctuations
has been considered. A probabilistic analysis of such structures has been presented,
providing estimates o f the response and the structural reliability.
The general problem, in view of the randomness of the system parameters, is of a
non-linear character. The solution presented here is based on the linearization method
(probabilistic method of the first order) and hence is approximate. It is the more precise
the smaller the fluctuations are in relation to their expected values. This solution can also
be treated as a first iteration step. In the second step formula (2.5) should be taken in
the form
X Y = ,~"~" + ) ~ f ' + ,~"Y"+ .~Y. (6.1)
Performing the expected value operation then gives
E[X (2) Y(')] = ..Y~"+ E[.~ (') ~'(')], (6.2)
where indices ('), (2) denote the first and second iteration steps, respectively.
As an example a block foundation problem has been considered. Analysis of the
eigenvalue problem has been carried out, with the stiffness matrix assumed to have small
random fluctuations with a correlation function of the form (5.5).
As can be seen from Figures 2(a) and (b) the structural correlation function is consider-
ably affected for small values of ~', ~, but the disturbance to it stabilizes and diminishes
as these parameters increase in value.
Figure 3 shows the effect of the randomness on the resonance index fl =fl(~') (5.6).
Significant effects of the parameters, especially when ~" has small values, are evident.
An example of forced vibration analysis of the foundation under investigation has been
given as well.

REFERENCES
I. V. V. BOLOTIN 1965 Statistical Methods in Structural Mechanics. Moscow: Stroiizdat.
2. J. D. ROBSON 1964 An Introduction to Random Vibration. Edinburgh: Edinburgh University
Press, Amsterdam: Elsevier.
3. Y. K. LIN 1967 Probabilistic Theory of Structural Dynamics. New York: McGraw-Hill.
4. B. SKALNIERSKI and A. TYLIKOWSKI 1982 Stochastic Processes in Dynamics. Warsaw: PWN--
Polish Scientific Publishers.
5. J. O. COLLINS and W. T. THOMSON 1969 American Institute of Aeronautics and Astronautics
Journal 7, 642-648. The eigenvalue problem for structural systems with statistical properties.
6. M. SHINOZUKA and C. J. ASTILL 1972 American Institute of Aeronautics and Astronautics
Journal 10, 456-462. Random eigenvalue problems in structural analysis.
7. H. GRUNDMANN 1977 Munich ICOSSAR'77, 81-96. On the reliability of structures under
periodic loading.
8. C. A. CORNELL 1970 Structural Reliability and Codified Design, SM, Study No. 3, University of
Waterloo. A first order reliability theory of structural design.