Professional Documents
Culture Documents
Artigo-1995 JSV SubstructureSynthesisFRF Ren&Beards 1-s2.0-S0022460X8570421X-Main
Artigo-1995 JSV SubstructureSynthesisFRF Ren&Beards 1-s2.0-S0022460X8570421X-Main
C. F. B
Department of Mechanical Engineering, Imperial College of Science,
Technology and Medicine, Prince Consort Road, London, England
1. INTRODUCTION
The ability to predict the dynamic properties of an assembly from the properties of its
components, or to predict the effects of a modification on a structure when some points on
the structure are clamped together is often desirable. The techniques to deal with this type
of problems are usually referred to as substructure synthesis or coupling techniques. The
coupling methods can be divided into two groups, in one modal data is used (see, e.g.,
references [1–3]), while in the other the measured response data (FRF) is used directly (see,
e.g., references [1, 4, 5]). In this paper, coupling methods in which FRF data is used are
studied. First, commonly used coupling methods are reviewed, and some severe
shortcomings of these methods are identified. To overcome these problems, a new FRF
coupling method, which also possesses the prime advantages of the available coupling
methods, is developed.
The accuracy of substructure synthesis is significantly affected by levels of noise in the
measured data. It is demonstrated that commonly used modal analysis techniques alone may
not be enough to yield accurate results. It is very important to detect any linearly dependent
coordinate pairs in the substructure synthesis process. The application of the new method
in detecting linearly dependent joint co-ordinate pairs is demonstrated.
845
0022–460X/95/350845+22 $12.00/0 7 1995 Academic Press Limited
846 . . .
2. BASIC CONDITIONS FOR FRF COUPLING
The basic conditions in the coupling process are compatibility and equilibrium. As shown
in Figure 1, the conditions of compatibility and equilibrium are such that if co-ordinate A on
the assembled structure corresponds to n co-ordinates (i=1, 2, . . . , n) on the substructure,
the displacements and forces on the assembly and the substructure must have the relation
n
xA=x1=x2=· · ·=xn and f A = s fi , (1, 2)
i=1
where x and f represent displacement and force respectively, subscript A refers to the
co-ordinate on the assembly and subscripts 1, 2, . . . , n refer to the co-ordinates on the
substructure (a list of nomenclature is given in Appendix C) the conditions of compatibility
and equilibrium can also be written in their matrix forms as
{xS }=[T]{xA } and { fA }=[T]T{ fS } (3, 4)
where subscript s refers to the co-ordinates on the substructure.
The purpose of coupling is to find the relationship between displacement {xA } and force
{ fA } of the assembly: i.e., to find the receptance matrix [HA ] in the equation
{xA }=[HA ]{ fA }, (5)
from the properties of the substructure system
{xS }=[HS ]{ fS } (6)
and the conditions of compatibility and equilibrium (equations (3) and (4)).
2 n–1
1 n
Substuctures
a
Assembly
Figure 1. The substructures and assembly.
847
methods are also different in terms of the formulation, computation speed and occupied
computation memory; there can also be restrictions on some FRF coupling methods in their
applications. The coupling methods may be evaluated by considering the following:
(1) Accuracy: the predicted characteristics of the coupled structure should not be sensitive
to computation errors (e.g., rounding off errors).
(2) Efficiency; which has two aspects: (i) the amount of computation memory and
computation time required to implement the coupling process should be small; and (ii) the
amount of time spent by the user in preparing the computing code and data should be small.
(3) Simplicity: all kinds of problem can be solved by using the same formulae so that once
a computation code is made it can be used to solve different coupling problems without large
modifications.
(4) Generality: there should be no severe restriction on the applications of the coupling
methods; in other words, there should be no severe restrictions on the substructure
receptance matrix [HS ]. The generality can have two aspects. One is physical generality,
which means that the method can predict the response of an assembled structure generated
from (i) grounding some co-ordinates of a structure, or (ii) coupling several coordinates
together on one structure, or (iii) coupling several substructures. The other aspect is the
mathematical generality, which means that application of the method should not be
restricted due to numerical properties of the receptance matrix (e.g., singularity), if the
substructures can be coupled physically.
Accuracy, efficiency, simplicity and generality are the four criteria which can be used to
evaluate a coupling method. With respect to the problem to be solved and the computational
resources available, different weightings should be given to these criteria. The mathematical
generality should be considered for all the coupling problems. The physical generality is also
of particular importance. For example, it is often necessary to know the properties of the
structure when two points on it are clamped together.
Multiplying both sides of equation (7) by the inverse of the matrix [HS ] yields
Substituting equations (3) and (4) into equation (8) leads to { fA }=[T]T[HS ]−1[T]{xA } and
hence
Therefore
If the substructure system has several components which are independent of each other, the
inversion of the matrix [HS ] can be completed by inverting the substructure receptance
848 . . .
matrices separately and then assembling these inverted matrices. The column number matrix
[T] corresponds to the assembly co-ordinates and the row number corresponds to the
substructure co-ordinates. If the ith co-ordinate in the substructure corresponds to the jth
co-ordinate of the assembly, then T(i, j)=1; otherwise, T(i, j)=0.
The IC method satisfies the criterion of physical generality. It is also very simple to
implement. However, this method is computationally inefficient, since two inversions of the
full size matrices are required. Because the full size matrix inversion is required, the numerical
error levels can be significant (especially when the full size receptance matrix is rank
deficient). When the full size receptance matrix is singular, the IC method will collapse; in
other words, the method is not mathematically generalized.
In summery, the IC method is a physically generalized method, but not a mathematically
generalized one.
$ % $ %
aHmm a Hmi Hmm
b Hmi
b
[aH]= , [b H]= . (11, 12)
aHim Hii
a Him
b Hii
b
where subscripts m and i represent the coordinates to be coupled (master co-ordinates) and
not to be coupled (internal co-ordinates) on the substructures. Subscripts j and n represent
the joint co-ordinates (corresponding to master co-ordinates in the substructures) and
non-joint co-ordinates (corresponding to internal co-ordinates in the substructures)
on the assembly respectively. aH, bH and abH represent the receptance sub-matrices
corresponding to the co-ordinates on a, b and between co-ordinates on substructures a and
b respectively.
This method requires only one matrix inversion. The size of the matrix to be inverted (i.e.,
[aHmm+bHmm ]) is determined by the number of the joint co-ordinates on the assembly and
is usually much smaller than the size of the full FRF matrix. Therefore, it is computationally
more efficient and accurate than the IC method. In addition, because the matrix to be
inverted is a summation of two receptance sub-matrices and is much smaller than the whole
receptance matrix in size, it is more likely to be a well conditioned matrix. In other words,
the RC method is mathematically more general than the IC method.
The major limitation of the RC method is that it can only be applied to couple two (groups
of) substructures. In the formula derivation, it is assumed that, before coupling, the two
substructures are independent of each other. The RC method is not applicable to cases in
which two co-ordinates to be coupled are located on the same structure as shown in Figure 2.
849
a b
c
Substructure
n j
Assembly
Figure 2. The substructure and assembly systems. Q, Rigid connection.
89& '8 9
xa Haa Hab Hac fa
xb = Hba Hbb Hbc fb . (14)
xc Hca Hcb Hcc fc
For the assembled system, the relationships between response and excitation are
67$ %6 7
xn H Hnj fn
= nn , (15)
xj Hjn Hjj fj
where subscripts a and n represent the internal co-ordinates on the substructure and
assembly, and subscripts b, c and j represent the joint co-ordinates. Substructure joint
co-ordinates xb and xc are so arranged that conditions of equilibrium and compatability
yield
{ fb }+{ fc }={ fj } and {xb }={xc }={xj }. (16, 17)
Non-joint co-ordinates are not changed after the coupling process, and hence
{ fa }0{ fn } and {xa }0{xn }. (18, 19)
From equations (14) and (17),
{xb }=[Hba ]{ fa }+[Hbb ]{ fb }+[Hbc ]{ fc }=[Hca ]{ fa }+[Hcb ]{ fb }+[Hcc ]{ fc }. (20)
850 . . .
Substituting equations (16) and (18) into equation (20) yields
([[Hbb ]+[Hcc ]−[Hcb ]−[Hbc ]){ fb }={([Hca ]−{Hba ]){ fn }+([Hcc ]−[Hbc ]){ fj }}. (21)
If matrix ([[Hbb ]+[Hcc ]−[Hcb ]−[Hbc ]) is non-singular, then
{ fb }=([[Hbb ]+[Hcc ]−[Hcb ]−[Hbc ])−1{([Hca ]−{Hba ]){ fn }
+([Hcc ]−[Hbc ]){ fj }} and { fc }={ fj }−{ fb }. (22, 23)
Substituting equations (22) and (23) into equation (14) and comparing with equation (15)
yields
[Hnn ]=[Haa ]−([Hab ]−[Hac ])([Hbb ]+[Hcc ]−[Hbc ]−[Hcb ])−1([Hba ]−[Hca ]), (24)
[Hnj ]=[Hac ]−([Hab ]−[Hac ])([Hbb ]+[Hcc ]−[Hbc ]−[Hcb ])−1([Hbc ]−[Hcc ]), (25a)
−1
=[Hab ]−([Hac ]−[Hab ])([Hbb ]+[Hcc ]−[Hbc ]−[Hcb ]) ([Hcb ]−[Hbb ]), (25b)
[Hjn ]=[Hnj ]T, (26)
−1
[Hjj ]=[Hcc ]−([Hcb ]−[Hcc ])([Hbb ]+[Hcc ]−[Hbc ]−[Hcb ]) ([Hbc ]−[Hcc ]) (27a)
=[Hbb ]−([Hbc ])([Hbb ])[Hbb ]+[Hcc ]−[Hbc ]−[Hcb ])−1([Hcb ]−[Hbb ]). (27b)
Equations (24)–(27) can be arranged in a more concise form as
$ %$ %$ % $ %
T
Hnn Hnj H Hac (Hab−Hac ) (Hab−Hac )
= aa − [Hbb+Hcc−Hbc−Hcb ]−1 ,
Hjn Hjj Hca Hcc (Hcb−Hcc ) (Hcb−Hcc )
(28a)
$ %$ % $ %
T
Haa Hab (Hac−Hab ) (Hac−Hab )
= − [Hbb+Hcc−Hbc−Hcb ]−1 .
Hba Hbb (Hbc−Hbb ) (Hbc−Hbb )
(28b)
The GRC method is very similar to the RC method in its formulation. Actually, the RC
method is a special case of the GRC method [6, 7]. The GRC method retains all the
advantages of the RC method, but it is physically more generalized. The GRC method can
also be used to couple some co-ordinates on one structure or to couple several substructures
together in a single step without any difficulty. If some co-ordinates are to be grounded, one
can consider the ground as a structure with infinite mass. To ground a structure at some
co-ordinates is the same as coupling the structure at these co-ordinates to another structure
with infinite mass. The FRFs of the infinite mass structure are zero. The formulae for
grounding co-ordinates are obtained simply by setting sub-matrices [Hac ], [Hbc ], [Hcc ], [Hcb ]
and [Hca ] in equation (28) to zero.
2
1
3
α
2 2
1 1
3 β 3
Figure 3. One example of using two-step coupling and pseudo co-ordinate coupling.
co-ordinates a and b which are located an infinitestimal distance from point 1 (hence they
have exactly the same response as point 1). The coupling is completed in a single step by
coupling point 2 with point a, and point 3 with point b.
8 9
1
, aqe
[Ha ]=[HS ]−b({hsj }−{hsi })({hsj }−{hsi })T, where b= hii+hjj−2hij .
0 , aWe
(32, 33)
The linear dependency between the co-ordinates is automatically checked by using the MTC
method; no additional checking is required.
There is, however, a small problem with this method. Experience indicates that the a value
can be very large at the anti-resonance of hii , hjj or hij , even if the co-ordinate pair is linearly
dependent. If the a value is set to be large enough so that the rigid joint pair is deleted at
all the frequencies, some other joint co-ordinate pairs can be deleted too, which results in
an erroneous prediction.
A better way is to divide the coupling process into two steps. First, all the co-ordinate
pairs are coupled with a values calculated over the frequency range of interest. These a values
are checked over the frequency range to determine if some pairs of the co-ordinates should
be deleted. The coupling process is carried out again with the exclusion of the linearly
dependent joint co-ordinate pairs.
Using MTC to detect the linearly dependent coordinate pairs is physically more significant
than any other available methods.
For the PCC method, the following five co-ordinate pairs are assumed, (x3 , x6 )(x3 , x4 ),
(x4 , x7 ), (x5 , x6 ) and (x1 , x0 ), where the co-ordinate x0 refers to a co-ordinate on the ground.
The RC method is not applicable directly, but it can be used to couple co-ordinate pairs
(x3 , x6 ) and (x4 , x7 ). The remaining co-ordinate pairs are coupled by using the MTC method.
The coupled results are shown in Figure 6(a); two curves are from noise-free substructure
FRFs, and from FRFs polluted by 5% random errors (an FRF is polluted by 5% random
–80
(a)
H(2,2) (dB)
–180
–80
(b)
–180
10 300
Frequency (Hz)
Figure 6. The predicted and the exact assembly point receptance (Case A). (a) Effects of random errors;
—w—, exact; —Q—, 5% noise. (b) Effects of consistent errors; —w—, exact; —Q—, consistent errors.
855
x6
x1 x2 x3 x4 x5
k1 k2 k3 k4 k5
m1 m2 m3 m4 m5 k6
x11 x10 x9 x8 x7 m6
k11 k10 k9 k8 k7
m11 m10 m9 m8 m7
Substructure system
x6
x1 x2 x3 x4 x5
k1 k2 k3 k4 k5
m1 m2 m3 m4 m5 k6
k11 k10 k9 k8 k7 m6
m11 m10 m9 m8 m7
Figure 7. The substructures and assembly of Case B. Q, Rigid coupling. mi values (kg): m1 , 1; m2 , 2; m3 , 3; m4 ,
4; m5 , 5; m6=m7=m8=m9=m10=m11=1. Ki values (N/m): k2=k3=k5=k6=k7=k9=k10=1.E 6;
k1=k4=k8=k11=1.E9; [D]=0·001[k].
noise if its real and imaginary parts are multiplied by two independent random numbers from
0·95 to 1·05). When the FRFs are noise-free, exact results are obtained from all the coupling
methods. When the FRFs are polluted, although the results from all of the coupling methods
are not exact, they are the same. This demonstrates that the final results of substructure
synthesis are independent of the coupling methods if numerical errors are insignificant.
The deterioration of the accuracy of the substructure synthesis clearly indicates the effects
of measurement errors.
Some peaks which do not represent real resonance frequencies can be found in the coupled
receptance data. Further investigation indicates that the frequencies of these peaks
correspond to the natural frequencies of the substructures. The reasons for these peaks are
–80
H(6,6) (dB)
–170
10 200
Frequency (Hz)
Figure 8. The assembly receptance by coupling five and three co-ordinate pairs (error free). —w—, Five pairs;
—Q—, three pairs (+1 dB).
856 . . .
–80
(a)
H(6,6) (dB)
–180
–80
(b)
–180
10 200
Frequency (Hz)
Figure 9. The assembly receptance by coupling three co-ordinate pairs (with consistent error). (a) —w—, Exact;
—Q—, five pairs; (b) —w—; —Q—, three pairs.
–80
(a)
–180
–80
(b)
H(6,6) (dB)
–180
–80
(c)
–180
10 200
Frequency (Hz)
Figure 10. The assembly receptance from SVD related coupling. —w—, exact —Q—, SVD threshold.
(a) threshold 0·001; (b) threshold 0·01; (c) threshold 0·05.
858 . . .
–70
(a)
–170
–70
(b)
H(6,6) (dB)
–170
–70
(c)
–170
10 200
Frequency (Hz)
Figure 11. The assembly receptance from MTC coupling with threshold. —w—, Exact; —Q—, MTC theshold
(a) threshold 0·05; (b) threshold 0·1; (c) threshold 0·3.
In Figure 8 is shown the predicted receptance from coupling three pairs of co-ordinates
((1, 11), (3, 9) and (5, 7)), from coupling five pairs of co-ordinates ((1, 11), (2, 10), (3, 9), (4, 8)
and (5, 7)). It is noted that the differences between the resuslts are negligible. Therefore, it
is only necessary to couple three pairs of co-ordinates.
In Figure 9 is shown the predicated assembly data obtained by using simulated modal
analysis data (with 5%, 1% and 1% errors in mode shape, natural frequency and modal
damping). It is noted that the predicted results with five pair coupling form a smooth curve,
but they are quite different from the real assembly response. The second resonance frequency
is significantly shifted. However, the results from coupling three co-ordinate pairs ((1, 11),
(3, 9) and (4, 8)) yield very high accuracy.
The results obtained with the commonly used threshold SVD algorithm are shown in
Figure 10. If e=0·01, very good results are found; however, if e is too large (0·05) or too
small (0·001), the results at some frequencies are very poor. Because the proper value of e
depends on the size of the matrix to be inverted and the frequency increment, it is very difficult
to determine the threshold of the SVD. Indeed, if a very fine frequency increment is chosen,
it is possible that no threshold is available to yield the accurate results at all the frequencies.
859
–60
(a)
Receptance (dB)
–175
1
(b)
log10(α)
–4
10 200
Frequency (Hz)
Figure 12. Receptance and a values (error-free). (a) Receptance at co-ordinates 2 and 5 before they are
coupled (b) a values for different co-ordinate pairs; —w— (1, 11); —Q—, (2, 10); —q—, (3, 9) —r—, (4, 8);
—W—, (5, 7).
–60
(a)
Receptance (dB)
–170
1
(b)
log10(α)
–4
10 200
Frequency (Hz)
Figure 13. Receptance and a values (with consistent error). (a) Receptance at co-ordinates 2 and 5 before they
are coupled. (b) a values for different co-ordinate pairs: —w—, (1, 11) —Q—, (2, 10); —q—, (3, 9); —r—, (4, 8);
—W—, (5, 7).
860 . . .
To avoid the dependency of the threshold on the size of the coupling, the threshold MTC
method is used. The results obtained by using the threshold MTC method are shown in
Fig. 11. It is noted that when e=0·3 the accuracy of the results is improved significantly;
however, at some frequency points, the results are poor (which coincides with the results
of four or five pair coupling).
In order to obtain a more accurate assembly response, the problem of linearly dependent
coordinate pairs should be dealt with systematically. The a values over the frequency
range are shown in Figure 12(b) for the error-free case in Figure 13(b) for the case with
errors. It can be clearly noted that the overall a values for co-ordinate pairs (2, 10)
and (5, 7) are smaller than the a values for the other pairs, which clearly indicates
that co-ordinate pairs (2, 10) and (5, 7) tend to be dependent on other joint co-ordinate
pairs.
The peaks of the a values for the co-ordinate pairs (2, 10) and (5, 7) in Figure 12(b) and
Figure 13(b) correspond to the anti-resonance of these co-ordinates when they are to be
coupled. The point receptance curves at co-ordinates 2 and 5 just before they are coupled
(i.e., the receptance at co-ordinate 2 after coupling the co-ordinate pair (1, 11) 5 after
coupling co-ordinate pairs (1, 11), (2, 10), (3, 9) and (4, 8)) for the error-free case and the
case with consistent errors are shown in Figures 12(a) and 13(a).
15,16 19,20
17,18 21,22 23,24 25,26
2.54
3 3 4 10 10 1.27
Substructure
9,10 13,14
1,2 3,4 5,6 7,8 11,12
Assembly
Figure 14. A diagram of the FE beam system. Q, Rigid connection; dimensions in cm. Beam width 2·54 cm;
Young’s modulus 6·55×1090 N/m2; density 2661 kg/m3.
much more difficult to identify, because the FRF curve in these frequency regions can be
smooth and the shape of a reading peak may look like a real resonance peak. In other words,
both inconsistent errors and random errors can cause reading peaks, and some reading peaks
can be difficult to detect. As a result, modal analysis becomes essential to remove the
inconsistent errors.
The synthesis results obtained by using FRFs regenerated from modal data are shown
in Figure 17. It is assumed that the FRFs are measured in a frequency range of 100–3000 Hz.
In this frequency range, there are six modes (out of a total of 26 modes of the substructure
system). The modal data of these six modes are assumed available. To compensate the effects
of the modes outside the measured frequency range, two residue terms are used. The lower
–50
–75
H(1,1) (dB)
–100
–125
–150
–175
0 500 1000 1500 2000 2500 3000
Frequency (Hz)
Figure 15. The assembly receptance by coupling two pairs and from coupling three pairs. —w—, Exact
(three pairs); —Q—, two pairs.
862 . . .
–50
(a)
–75
H(1,1) (dB)
–100
–125
–150
–175
–50
(b)
–75
H(1,1) (dB)
–100
–125
–150
–175
103
(c)
102
101
100
α
10–1
10–2
10–3
0 500 1000 1500 2000 2500 3000
Frequency (Hz)
Figure 16. Results from data with inconsistent measurement errors. (a) Results from coupling three pairs;
—w—, exact. (b) Results from coupling two pairs; —w—, exact. (c) The corresponding a values: —Q—, pair
(9, 15); —R—, pair (13, 19); —w—, pair (11, 17).
and higher residue frequencies are assumed to be at 1 Hz and 4000 Hz, and the magnitudes
of the residue terms are determined from the FRFs at 110 Hz and 2900 Hz respectively. To
simulate the remaining errors in the regenerated FRF data, the consistent error model is used
with mode shape, natural frequency damping being polluted by 10%, 1% and 10%
respectively.
It can be noted in Figure 17 that reading peaks are effectively removed by using consistent
FRF data. Even if the modal data are not exact and only a few modes are measured, accurate
prediction is achieved by coupling just two co-ordinate pairs. However, coupling additional
linearly dependent co-ordinate pairs leads to poor prediction. Therefore, they must not be
used in the coupling process. Fortunately, this linear dependency of the co-ordinate pair
(11, 17) on the other two pairs can be easily detected from the corresponding a values
(see Figure 17(c)) by using the MTC method. As a result, this co-ordinate pair can be
removed.
863
9. CONCLUSIONS
A new generalized coupling method (GRC) method is proposed, which is physically more
general than the RC method and is computationally more efficient than the IC method. The
GRC method includes the advantages from both the IC and RC methods and is superior
to both of these methods.
By using the GRC method, the coupling procedure can be implemented in either a single
step or in several steps by coupling two co-ordinates at each step. This leads to two modified
versions of GRC; i.e., PCC and MTC.
The MTC method is particularly attractive. Not only is it computationally more efficient
and easier to program, but it can be used to detect the linearly dependent joint co-ordinate
pairs as well. Using MTC to detect linearly dependent joint co-ordinate pairs is simpler and
also more effective than other methods such as the SVD algorithm.
It also has been demonstrated that by removing linearly dependent co-ordinate pairs by
using the MTC method, and by removing random and inconsistent systematic errors by
using the modal analysis techniques, accurate synthesis results can be achieved.
–50
(a)
–75
H(1,1) (dB)
–100
–125
–150
–175
–50
(b)
–75
H(1,1) (dB)
–100
–125
–150
–175
103
(c)
102
101
100
α
10–1
10–2
10–3
0 500 1000 1500 2000 2500 3000
Frequency (Hz)
Figure 17. Results from data with consistent errors. (a) —Q—, Results from coupling three pairs; —w—, exact.
(b) —Q—, Results from coupling two pairs; —w—, exact. (c) The corresponding a values: —Q—, pair (9, 15);
—R—, pair (13, 19); —w—, pair (11, 17).
864 . . .
REFERENCES
1. A. P. V. U 1989 Ph.D. Thesis, Imperial College of Science, Technology and Medicine,
University of London. Dynamic analysis of coupled structures using experimental data.
2. W. C. H 1965 American Institute of Aeronautics and Astronautics Journal 3(4), 678–685.
Dynamic analysis of structural systems using component modes.
3. S. H 1969 Shock and Vibration Bulletin 40(4), 25–39. Review of modal synthesis techniques and
a new approach.
4. M. I, D. A. R and D. J. E 1987 Proceedings of the 5th International Modal Analysis
Conference, London, 1136–1141. Structural modification and coupling dynamic analysis using
measured FRF data.
5. J. S. T and Y. F. C 1988 Journal of Sound and Vibration 125, 487–502. The identification
of dynamic characteristics of a single bolt joint.
6. Y. R and C. F. B 1993 Proceedings of the 11th International Modal Analysis Conference,
Florida, U.S.A., 868–871. A generalised receptance coupling technique.
7. Y. R 1992 Ph.D. Thesis, Imperial College of Science, Technology and Medicine, University of
London. The analysis and identification of friction joint parameters in the dynamic response of
structures.
8. Y. R and C. F. B 1993. Proceedings of the 11th International Modal Analysis Conference,
Florida U.S.A., 872–876. A new multi-step two coordinate coupling technique and its application
for detecting linearly dependent coordinates.
9. D. O, J. L, D. V, H. G and R. A 1990 Proceedings of the 8th
International Modal Analysis Conference, Florida, 213–220. Coupling of structures using measured
FRFs by means of SVD based data reduction techniques.
N
ca n cb N
Cn
Hab= s n
=s , (B1)
n=1
(1+ihn )vn −v2 n=1 (1+ihn )vn2−v 2
2
where ca and cb are the elements in the mode shape vector, vn is the natural frequency and
h is the damping value; subscript n refers to the nth resonance, and N is the DOF of the
system. Then the FRFs from measurement can be simulated as
where
E is a random number within a range of 12e, and e is the magnitude of the corresponding
error. Each random number depends only upon the values and sequence of the variables
inside the parenthesis. For example, E(3, a, b, n) is related to co-ordinates a, b and mode
number n, but it is independent of frequency v, and also E(3, a, b, n)$E(3, b, a, n).
For FRF data regenerated from modal data, because FRFs can be measured only in a
limited frequency range, the modal data for modes outside that range are usually not
available. To compensate the effects of modes outside the measured frequency range, residue
866 . . .
terms can be used. The following formula can be used to calculate the regenerated FRFs
from modal data:
nc a n c b
n2
Crl Crh
Hab= 2+ s + . (B4)
v −v n=N 1 (1+ihn )vn2−v 2 vrh2 −v 2
2
rl
Here n=N1 , . . . , N2 represents modes in the measured frequency range, Cr and vr refer to
the modal constant and natural frequency of the residual terms, and the second subscripts
l and h represent the lower and higher frequency residue terms respectively. When vrl and
vrh are selected, equation (B4) is a linear equation with Crl and Crh being the only unknowns;
hence they can be easily determined from Hab at two different frequencies.
In real cases, modal data extracted from measured FRFs inevitably contains errors.
However, because the same set of modal data is used, regenerated FRFs are consistent. To
simulate a regenerated FRF with consistent errors, the following formula can be used:
APPENDIX C: NOMENCLATURE
f force
h receptance element
x displacement
{f } force vector
[H] receptance matrix
{x} displacement vector
[T] matrix related to the conditions of equilibrium and compatibility
a threshold of the MTC method
v angular frequency
Subscripts
A assembly
s substructure
a non-joint region on substructure system
b joint region on substructure system
c joint region on substructure system
n non-joint region on assembly system
j joint region on assembly system; also quantities related to joint