ABSTRACT NOMENCLATURE
139
1 INTRODUCTION 2 THEORY
140
in which
(18)
FPP = ‘@$ [+f)T > (6)
I I and
5
[Cl = [ [$I [g] I (7) { F ’ ) = pp 2
I (19)
and
[ff,l= [ 1%) [g] ,
1 (8)
1 G J
for each structure to give Here [ZA] and [Z,g] are the dynamic stiffness matrices
for structures A and B respectively, given by
[-u” [r] + iw [C-I + i[H’] + [W] + [A*]] {pot) = {F*}
(13) [ZAI = [Rd-‘, (24)
141
and can be well approximated by using a highly truncated
modal series; aud hence only a few modal coordinates,
instead of the full n degrees of freedom are used. For
general nonliiear systems, on the other hand; this ap-
proach may not always be applicable since the vibration
denotes the forcing vector on the coupled structure C. modes may not be easy to define, or may not even ex-
Furthermore, ist at all. For structures that can be analysed by the
above mentioned quasi-linear approach however, modal
truncation may be used, provided that sufficient care is
taken. In the following, the term “modes” refers to the
modes of the linear undamped part of the overall cow
pled system. If only m modes are kept in the analysis,
m being much less than n, the total degree of freedom.
gives the complex amplitude vector of displacements the response may be approximated as
for the coupled structure C. If, initially, the receptance
matrices for the structures A and B are available, the
pseudo receptance matrix for the coupled structure C W)“,, - Pd%x, hL1 (32)
can be obtained from
where the subscript t refers to the modes kept. If Equa-
tion (32) is used for each structure instead of Equations
PC1 = k-’ (30)
(3) and (4) for the transformation into the modal do-
main, the order of the matrices to be dealt, with reduces
to m.
2.2 Numerical Solution It has been demonstrated in Ref. [8] that such a trun-
cation mayprove to be very effective when incorporated
with the quasi-linear approach. It is well worth noting,
The pseudo receptance matrix [@I given by Equations however, that the accuracy of modal superposition is
(20) and (30) may be used to determine the response dependent on the natural frequencies of each structure,
{X} of the coupled structure from the frequency of excitation, the nature and distribution
of the nonlinearities as well as the configuration and
{Xl = WC1 IFI (31)
amplitude of forcing. Hence it may not be possible to
where, through Equations (21) and (27), [e] depends on determine beforehand the minimum number of modes to
the nonlinearity matrix [A], which in turn depends on be included for an acceptable approximation of the re
the unknown response vector {X} [4]. The coupled set sponse. The value of m determined by a linear analysis
of nonlinear algebraic equations given by Equation (31) may give a good starting point, though.
can be solved iteratively at discrete frequency points
of interest by updating the nonlinearity matrix [A] at
each iteration. The inversions required at each itera-
tion, giveft by Equations (21) and (30) can be bypassed 3 CASE STUDY
by using Ozgiiven’s method (71, thereby obtaining sav-
ings iu computational time and avoiding ill-conditioned
matrices, especially for large structures. In this section, the performance of the method dc
veloped is evaluated by a case study. The accuracy
of the results obtained has been verified by using the
Newmark-Beta method in solving the simple demon-
2.3 Modal Truncation
strative problem considered. Comparison of the results
has demonstrated that the method developed has a high
The response of a linear system can be expressed as a accuracy. A detailed justification of the accuracy of the
summation of its natural modes, and this property can method, excluded here due to space limitations, can be
be exploited in order to obtain drastic reductions in the found in Ref. [9].
computational work involved in the analysis of a Lin-
ear structure, since only a few vibration modes are ex- Consider the two linear SDOF systems shown in Fig-
cited for harmonic forcing. In many cases, the response ure 1. The system parameters are as follows:
142
kz z k3 z ks = kri = lOOOON/m In t,hh study, two semi-analytical frequency domain
r)* = l,” = 0.02 techniques for forced harmonic response analysis of cou-
pled nonlinear MDOF structures have been developed.
‘The natural frequencies of each substructure are eval- The first one, a Modal Coupling method. can be used for
uat,ed a s w,,, ~17.8, Irish =lu2.5, WA, =173.7;
the analysis of large complex structures for which time
yIBl =12.6. tin1 ~72.5, us, ~122.8, all in rad/s. domain numerical integration procedures are either too
costly or infeasible. The method also presents an ad-
As a case study, a softening cubic stiffness with force vantage over existing frequency domain techniques that
displacement relationship do not involve substructuring because it reduces the to-
tal time needed for the eigenvalue problem (EVP) solu-
nj* = 500~~~ - 2.5 x 10G$, (33) tion of a large structure drastically by concentrating on
the individual eigenvalue problems of its substructures.
Although many substructuring methods based on the
was oscd t,o couple the two systems via coordinates 3
linearity assumption are available, there are only a lim-
and 6. Harmonic external forcing whs applied on coordi-
zted number of proposed methods for nonlinear coupling
natp 3. Preliminary linear runs over a broad frequency
analysis; most of which have been developed for specific
range demonstrat,cd t~hat, the method was able to cap-
types of nonlinearities. The proposed m&hod can deal
fwe t,he natural frequencies of t,he linear undamped part
with any odd, single valued, symmetrical nonlinearity’.
of t,he overall system given by
Hence most of the common types of nonlinearities in-
herent in structural joints such as polynomial or piece-
(u,,)T = (13.9,23.2,73.3,103.7.123.0,174.0] (34) wise linear types of stiffnesses, clearances and Coulomb
friction can be analysed. The second method has been
start,ing from t,he individual natural frequencies of the developed for predicting the quasilinear receptance ma-
two subsystems. Not? the shifty in the lowest two natural trix of a nonlinear structuw cons&&g of several lin-
frequencies and that the eet given by Equation (34) was ear subsystems coupled by nonlinear elements of the
not calculated directly by any means; the figures can be above cited form. The subsystems are to be described
deduced from the frequency response information cal- by their individual FRF matrices which may either be
culated with the nonlinearity excluded. This presents experimentally obtained or computed using the mathe-
another powerful aspect of the method presented. matical modelof each substructure. Hence, the method
offers the important advantage of combining dynamic
Figure 2 shows a comparison of the linear and nonlinear descriptions of various kinds.
response X3 of the bhird coordinate to forcing of magni-
t,ode FB=lN. As can be seen, the second resonance peak The applicability of the methods developed has been
of the nonlinear syst.em is at about 21.2 Tad/s, 2 rad/s investigated by developing computer programs for each
below that of the linear system. Note the substantial de- formulation, and performing several case studies by us-
crease in amplitude around both resonances, especially ing the programs. It has been verified that the pre
the second one. where the nonlinearity is more effective. gr a m s yield results of high accuracy; by comparing their
Consider Figure 3, in which ze as well a~ Fs is plot results to “accurate” time domain results obtained by
ted for F3=lN. As can be seen, little increase occnrs in numerical integration. It has been shown that typical
xe around the second resonance, hence the nonlinear frequency response behaviour pertinent to certain types
spring between coordinates 3 and 6 is activated over a of nonlinearities, such as jumps and m&iple solutions
relatively broad frequency interval in this range. The in the frequency response for cubic stiffness, can be PI+
jump in TS at 21.2 rad/s, a typical behaviour for cubic dieted by the methods developed.
stiffness type of nonlinearity, can be clearly observed.
The increasing effect of nonlinearity can be observed in The computational time requirements for the methods
Figure 4, in which FS is plotted for Fsz0.5, 1, 2 and have also been investigated. For the first method, apart,
4N. The large peak in the curve for F3=4N corresponds from the gross reduction in EVP solution time explained
to a case of snap-through; i.e., to the case where the above, an extra reduction in the computational time
nonlinear force njb in Equation (33) becomes negative 'Tobe p~edse.~~oddnonlinearityth~t doesmtproduoe a
(Y36 > 1.4 x lO+n). bias term in the describing function codEdents.
143
can be obtained by employing modal truncation. Ta matrix will be calculated must not exceed (or, for, say,
this end, the linear undamped modes of the linear sub- Coulomb damping, go below) the limit at which non-
structures are utilized, the higher ones being discarded. linear effects become nonnegligible.
However, the number of modes kept should be selected
carefully in order not to impair the accuracy of the re- In the methods presented, no special treatment has been
sults. Since it is not straightforward to extend the con- given for localized nonlinearit,ies. Hence the same com-
cept of modes to nonliiear systems, the worthiness of putational effort is spent for distributed and localized
the approximation depends on several factors such as nonlinearities. For very large systems having only a
the point, level and frequency of excitation as well as small number of coupling related coordinates. the corn-
the type, strength and kxation of the nonlinearity. In- put&canal time can be further reduced by using au
vestigating the possibility of determining the minimum extension of the procedures mentioned Ii’] in order to
number of modes to be used for a prescribed accuracy avoid inversion. The details of this extension can be
may constitute an area for potential research. The see found in Ref. 141. It must be noted that ‘linear springs’
ond method, on the other hand, utilizes no EVP solu- in the coupling region are also considered as external
tion, and hence the computational time is determined forces on each substructure in the methods presented.
by the iteration time. Hence, they are considered in the nonlinearity matrix
[A] as well. When two substructures are connected in
The methods developed can be used in structural design this fashion, the undamped natural frequencies of the
modification/optimization problems, with large savings overall structure are different from those of the sub-
in computer space and computation time. Since the structures. Note that, in the methods presented, the
methods are compatible with standard modal analysis new natural frequencies of the coupled system are not
procedures; they may be used in conjunction with many calculated directly, nor are they used by any means.
well established routines such a4 commercial finite el- The convergence rate of the modal coupling formulation
ement analysis packages upon implementation of the could be further increased by using the nat,ural frequen-
procedures formulated herein. cies of the overall structure as a starting point. These
new natural frequencies must, be evaluated without solv-
Although the formulations were made for a case of lin- ing the overall EVP of the structure. Methods for de-
ear subsystems, actually the nonlinearity matrix may termining the new eigenvalues of a modified structure
not be restricted to the coupling coordinates, i.e., the by using the eigenvalues and eigenvectors of the original
matrix formulation enables the nonlinear forces internal structure are available in the lit.erature [ll]. Such tech-
to the substructures to be expressed in a similar way to niques may be incorporated with the present method;
the interface forces. In this case, the internal nonlinear although it is not clear whether the overall iteration
forces as well as the coupling forces will be viewed as ex- time will decrease in this case; since these methods also
ternal forces acting on the linear parts of each structure. require iterative solutions.
Thus, theoretically speaking, the methods are also ca-
pable of analysing structures formed by assemblingnon- A final remark on the practical applicability of the
linear systems. Although the first method is perfectly methods is worth noting. Although the accuracy of
suitable for purely analytical studies involving nonlin- the methods has been tested numerically, in order t,o
ear substructures; its application to real structures us- he used in practical applications, their results must be
ing discrete nonlinear elements is not straightforward compared to experimental ones. It is known that the
because the problem of detection, identification and 1~ available linear receptance coupling methods are quite
cation of nonlinearities becomes an important problem sensitive to measurement errors. Therefore. especially
in this case. Hovewer, the method can successfully be the second method needs to be tested for accuracy by
used to determine the frequency response of structures using experimental data before applying it to practical
modelled by the finite element method, where the non- problems.
linear effects due to large amplitude vibrations can be
taken into account as being ‘continuous nonlinearities’
[lo]. This will be discussed in a forthcoming paper [9].
As for the second method, it can be used at most for REFERENCES
slightly nonlinear substructures, provided that the non-
linear effects are not evoked while measuring the FRF [l] Craig Jr., R.R. A Review of Time Domain and Fw-
mat.rices of each structure. In this case, the forcing level qumcy Domain Componeni Mode Synthesis Melh-
on t,he overall~tructure for which the pseudo receptance ads, Combined Experimental Analytical Modeling
144
of Dynamic Structural Systems, AMD. Vo1.67, pp
1-30. 1985.
145