Mechanical Systems and Signal Processing: Oriol Guasch, Carlos Garcı A, Jordi Jove, Pere Artı S
Mechanical Systems and Signal Processing: Oriol Guasch, Carlos Garcı A, Jordi Jove, Pere Artı S
Mechanical Systems and Signal Processing: Oriol Guasch, Carlos Garcı A, Jordi Jove, Pere Artı S
a r t i c l e i n f o abstract
Article history: Transmissibility functions have received renewed interest given the important role they
Received 25 February 2012 play in operational modal analysis and operational transfer path analysis. However,
Received in revised form transmissibilities can also be used in the framework of classical transmission path
28 November 2012
analysis. This avoids some of the problems associated to the latter, such as the
Accepted 5 January 2013
Available online 15 February 2013
measurement of operational loads, or the need to remove the active parts of the system
to measure frequency response functions. The key of the transmissibility approach to
Keywords: classical transfer path analysis relies on the notion of direct or blocked transmissibilities,
GTDT method which can be computed from standard measurable transmissibilities. The response at
Blocked subsystems
any degree of freedom to a system external load can then be decomposed in terms of the
Direct transmissibility
remaining degrees of freedom responses and the system direct transmissibilities.
Transfer path analysis
Global transmissibility Although the theory supporting this approach has been known for long, no experimental
validation test has been reported to date. It is the purpose of this paper to provide such a
test by applying the method to a simple mechanical system for which an analytical
solution can be derived. For different congurations, it will be shown that direct
transmissibilities computed from measured transmissibilities compare fairly well with
analytical results. This opens the door to apply the method to more complex situations of
practical interest with condence.
& 2013 Elsevier Ltd. All rights reserved.
1. Introduction
So far, several experimental approaches have been followed to determine vibro-acoustic transfer paths in a mechanical
system. These are commonly referred to as transmission path analysis (TPA) techniques. In the late 1970s, approaches
based on multiple and partial coherence analysis were developed [14] (see [5] for a compact formulation on conditioned
spectral density functions). These approaches were one step methods in the sense that only operational data was required
to compute the system transfer paths. Although useful in many situations, they presented the problem of separation of
partially correlated sources. Therefore, alternative TPA strategies were developed during the 1980s. The most celebrated
one is nowadays commonly known as classical TPA. Classical TPA is a two-step method that aims at knowing the inuence
of the operational loads acting on a system, on the response of some selected degrees of freedom (d.o.f.s) of the system
n
Corresponding author. Tel.: 34 93 290 24 76; fax: 34 93 290 24 70.
E-mail address: [email protected] (O. Guasch).
URL: http://www.salle.url.edu/~oguasch (O. Guasch).
0888-3270/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ymssp.2013.01.006
354 O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369
[69]. In a rst step, frequency response functions (FRFs) become measured, whereas in a second step operational loads
(basically mechanical forces and volume velocities) get measured. The operational response of a d.o.f can then be split in
terms of the contributions of the various operational loads acting on the system. The main drawbacks of classical TPA are
related to measurement issues, namely the problem of having to remove the active parts of the system when measuring
FRFs, and the difculties associated with the measurement of operational forces (see e.g., [10]). These and related topics
makes classical TPA still an active research area (see e.g. [11] and references therein, or [12] for developments in the time
domain, among many others).
In view of classical TPA measurement difculties, it would be certainly worthwhile to develop a TPA strategy that could
be, at least, free of some of them. A good option would be that of having a TPA based on transmissibilities, which would
allow decomposing the response at any selected d.o.f in terms of the responses at all remaining d.o.f.s. This possibility was
already proposed in the very early stages of TPA [13]. In [13], a two-step approach was developed that involved performing
a rst set of transmissibility measurements, and a second set of measurements of operational responses. The former
requires no demounting tests whilst the second poses no particular problem. In contrast, the main difculty with this
transmissibility approach to classical TPA is that measured transmissibilities cannot be used for the operational response
reconstruction but instead, the so-called blocked or direct transmissibilities are needed to do so. However, it was also shown
in [13] that direct transmissibilities could be computed from measured transmissibilities, which were latter referred to as
global transmissibilities. Hence, this transmissibility approach to classical TPA will be termed hereafter the GTDT (Global
Transmissibility Direct Transmissibility) approach to TPA, as coined in [14].
This paper focuses on some aspects of the GTDT approach to TPA. Although formulated long time ago, no detailed and
controlled experimental test that could help to understanding as well as assessing the GTDT validity has been published to
date (as far as the authors know). It is the purpose of this paper to provide such a test by analyzing the GTDT performance
on a simple mechanical system made of springs, dampers and masses. A seven d.o.f. system is considered, for which an
analytical solution can be derived for comparison with experiments. Moreover, the presentation of the relation between
global and direct transmissibilities differs from the standard one in [1315]. In this work, their connection with the
dynamic stiffness and transfer matrices will be emphasized, in an attempt to facilitate comprehension. Issues related to
measurement aspects and to the validation of results will be also discussed. In some sense, the present work complements
that in [15], where a theoretical analysis of the GTDT approach, when applied to a less complex discrete mechanical
system, was performed. Path blocking issues as well as a comparison with classical TPA were also addressed in [15]
(see also [11]).
Up to now, most research on the GTDT method has been theoretical. For instance, in [14,16] it was applied to
continuum systems. In the rst reference, the case of transfer paths of bending waves in a beam was addressed,
whereas in the second one, the connection between direct transmissibilities and the numerical discretization schemes of
some nite element and nite difference methods was established for the Helmholtz equation. Besides, a rst link between
direct transmissibilities and energy transmission paths in statistical energy analysis (SEA) was provided in [17]. This
connection was complemented and fully exploited in [18], to propose an experimental SEA formulation that does not
require input power measurements. In [19], similar concepts to the ones in the GTDT method were introduced, although
using different notation, and the role of the eigenvalues and eigenvectors of an equivalent direct transmissibility matrix
was analyzed. In contrast to these theoretical works, very few experiments have been reported to date. The
transmissibility approach to TPA gave rise to covering techniques in which the direct or blocked transmissibilities were
attempted to be measured instead of computed from global transmissibilities [20]. The main problem with this
experimental procedure is that it is extremely time consuming and that the vibroacoustic behavior of the system can
become too altered, depending on the materials used for the covering. In [21,22] industrial applications of the GTDT
approach were applied to study cabin noise in railways. However, only very general noise contribution results were
provided in those references and no detailed information on global and direct transmissibilities was shown. As mentioned
above, the aim of this paper is to begin to cover this lack of controlled experimental results on the GTDT approach to
transfer path analysis.
Finally, it is worthwhile mentioning that the transmissibility concept received renewed interest in the beginnings of the
2000s [23,24]. The interest has grown with the appearance of operational modal analysis [2527] and operational TPA
[2831]. Two-step transfer path methods such as classical TPA or the GTDT approach can be very effective but they
obviously have the problem of being time consuming. This can prevent their use for some industrial applications.
Consequently, current research has been devoted to go back to the origins of TPA so as to develop one-step methods that
only require operational data measurements. This has lead to different operational TPA approaches. However, and as
quoted in [10], special care has to be taken when applying these methods as they are not free of problems (strongly
misleading results can be obtained for instance, if a path is missing). Therefore, operational TPA methods that try to
combine the accuracy of traditional TPA approaches with the time efciency of operational TPA have been also
attempted [32]. Whether to use in practice two-step classical TPA or GTDT approaches, or one-step operational TPAs,
usually depends on many factors (not necessarily scientic, e.g., economical or time constraints) that could tip the balance
towards one type of method or the other.
The paper is organized as follows. In Section 2 the basic features of the GTDT method will be reviewed. The meaning
and relation between global and direct transmissibilities will be described, as well as how to reconstruct the response at
any d.o.f. in terms of the remaining d.o.f.s responses. In Section 3, the experimental setup and measurement details will be
O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369 355
reported, whereas the analytical solution for the GTDT validation test will be provided in Section 4. Experimental
and theoretical results will be compared in Section 5. Finally, conclusions will be drawn in Section 6.
Let us consider a typical situation in automotive vibroacoustics, where we want to determine the contributions of the
engine operational forces, fi, to the acoustic pressure at the drivers position inside a car, p. Assume for simplicity that fi are
being input to the car body through three connection points and that only the vertical displacement is of interest. Classical
force TPA proceeds as follows to face this problem. In the rst step of the method, the engine is decoupled from the car
p
body and transfer functions Hij (frequency response functions, FRFs, for s io, s being the Laplace variable and i 1)
between the connection points and the interior acoustic pressure are measured. In the second step of the method, the
engine is xed to the car body and operational forces at the connection points become measured, as well as the interior
acoustic pressure. If the latter was only due to engine vibrations transmitted through the connections points, it should
P
happen that p 3i 1 Hij f j , each summand thus providing the contribution of a particular engine connection point to the
interior noise. In a more general situation where not only the acoustic pressure but other system d.o.f. responses were to
be analyzed, we could build a vector x of generalized responses (x may include displacements, angles, velocities,
accelerations or acoustic pressures) and a vector f of Nf generalized external loads acting on the system (f may include
external forces, moments or acoustic volume velocities). Having measured all frequency response functions between
responses and loads, the operational response at any d.o.f., xi, could be expanded as
Nf
X
xi Hij f j : 1
j1
However, classical force TPA has two well-known drawbacks. The rst one concerns the fact that the active part of the
system has to be removed when measuring the FRFs (otherwise, when exciting one connection point forces could also
enter the other connection points, because of vibration transmission through the engine). The second one is related to the
difculties found when measuring operational forces given that, for example, the use of load cells for mechanical forces is
not always feasible. Alternative approaches have been developed through years, although once more, they are still not
exempt of problems. For instance, the mount stiffness approach is of limited application to particular type of connections
and has the inconvenience of obtaining reliable data for the stiffness mounts. The inverse force identication method is of
more general application, but it increases considerably the number of measurements to be made in a TPA. Moreover,
ill-conditioning problems can appear in the inversion of the measured FRF matrix and regularization techniques become
necessary [3335].
Though useful in many situations, the vibroacoustic engineer may encounter circumstances in which the above
limitations could turn classical force TPA inappropriate. This could be the case, for example, of a large industrial machine
where decoupling of the active part of the system was unfeasible. It would be then desirable to nd some way to perform
the analogous of decoupling through some mathematical postprocessing. Yet, in some instances little can be done to
modify the external loads acting on a system (e.g., the aerodynamic loads acting on the cabin of an airplane and generating
interior noise) so one could think of alternative strategies to classical TPA to analyze its dynamics. The GTDT approach to
TPA offers some possibilities in this sense. The method is based in splitting the response at a system d.o.f. in terms of the
other d.o.f.s responses, rather than in terms of the external loads. For instance, in the case of the interior noise of the
airplane cabin, this would be breakdown in terms of the contributions of the accelerations of the cabin panels, whereas for
the car example, the accelerations at the connection points would be used. At this stage it is important to note that there
are some signicant differences between the classical force TPA and GTDT approaches. The former allows knowing the
inuence of an external load on the response of a target d.o.f., that has been transmitted through all transmission paths
connecting the input d.o.f. with the target. The second expresses the inuences that the responses at a selected group of
d.o.f.s will have on the response of the target d.o.f., and allows identifying individual transmission paths in the energetic
formulation (out of the scope of this paper, see [15,17,18,36]).
If one considers the responses at N d.o.f.s in the system, the analogous to (1) for the GTDT reconstruction of the
response at one of them would be
X
N 1
xi TD D ext
ij xj T ii xi : 2
j 1,jai
Thus, the response at any d.o.f. xi, is decomposed into contributions from other d.o.f. responses (rst term in the r.h.s., right
hand side, of (2)) plus the portion of xi that is directly due to the external load acting on it (second term in the r.h.s. of (2)).
Note that no force appears in (2). The meaning of the terms T D D
ij and T ii (the so called direct or blocked transmissibilities) can
be made apparent from the equation. In the second term at the r.h.s. (right hand side), xext i stands for the operational
response the i-th d.o.f. would have if the external load f was only acting on it (i.e. f n din f i , din standing for Kroneckers
delta). Suppose that we just hammer the m-th d.o.f.; then xext i 0 (no force on it) and from (2) we will get that T D im xi =xm
356 O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369
provided all responses xj 0 8jam. That is to say, the terms T D ij (direct transmissibility from d.o.f. j to i) are nothing but the
response at i when j is excited and all the other d.o.f. responses become somehow blocked. Accordingly from (2), the term
TDii xi
ext
represents the response at the i-th d.o.f. when we excite it and the remaining system responses xj , jai, get blocked
(i.e., xj 0). If we denote this quantity by x0 T D ext D 0
ii xi , the term T ii x =xi
ext
(direct transmissibility from one d.o.f. response to
itself) is nothing but the quotient of the response at i when i is excited and all the remaining responses become blocked,
and the response at i when i is excited and there is no restriction on the other subsystem responses.
That is to say, in the GTDT approach a set of d.o.f.s are selected (not necessarily those at which external forces are input
but, ideally, those at which structural modications could be made, or those whose inuence on the target is deemed
important) and their operational responses and direct transmissibilities get measured/obtained (see below). Then the
response at any d.o.f. can be decomposed in terms of these quantities. Observe that the decomposition in (2) could be
useful for the vibroacoustic engineer because it allows quantifying the effects that a certain reduction in a direct
transmissibility, would have on the response of a given target d.o.f. For instance, one could see how the interior noise at
the airplane cabin would diminish when reducing the T D ij of a given panel by means of changing material properties,
rigidication, or insertion of double partition walls. Similarly, one could analyze what happens with the noise at the
drivers position in a car when modifying direct transmissibilities at the engine connection points, but also at other points
located in the chassis.
In what concerns sensor locations and inuence of possible structure modal behavior this is not a problem in the GTDT
approach provided the system under analysis is linear (which is assumed in all TPA approaches) and within the standard
experimental limitations e.g., typical problem of having placed a sensor at a nodal point for a given frequency, lack of space
to properly excite all d.o.f.s in the appropriate directions, etc. Note in this sense, that these difculties also arise in classical
force TPA because, as mentioned, some of the most usual methods to determine operational forces, namely the mount
stiffness approach and the inverse force identication method, precisely rely on response measurements (e.g., acceleration
is measured at various points in the chassis of a car to determine by inversion methods the operational forces at the engine
connection points). Thus, experimental difculties in the GTDT method do not arise from the measurement of the
operational response vector x in (2) (it will be also shown below that xext can be easily obtained from x). The main problem
with (2) resides in obtaining the direct transmissibilities T D ij given that this would imply blocking all d.o.f.s but j and i,
when measuring T D ij . One could attempt at physically blocking these d.o.f.s to obtain the direct transmissibilities, as done
with covering techniques [20]. However, and as already mentioned earlier, this approach suffers from many drawbacks.
Another possibility would be to try to obtain T D ij directly from operational data, using least squares techniques, somehow
analogously to what is done in operational TPA. This could be an alternative (again not free of problems) although it will
not be considered herein, being left for future works. The proposal of the GTDT approach to address this issue is to
compute T D ij from standard transmissibilities, which are easy to measure given that they do not require any demounting
test. This is precisely the key point of the GTDT approach to TPA: establishing a connection between standard or global
measurable transmissibilities and blocked or direct transmissibilities [1315].
To end this subsection and in what concerns terminology, transmissibility will be used hereafter to designate the
quotient between the responses at any pair of d.o.f.s, when the system is excited by an external load. The responses do not
necessarily have to correspond to the same physical magnitudes. For instance, we could have acoustic pressure/
acceleration transmissibilities, as well as velocity/velocity transmissibilities. The term transfer functions will be kept to
designate quotients between the system response at a d.o.f. and the external load applied on it, or at any other system d.o.f.
This nomenclature distinction between transmissibilities and transfer functions was not considered in previous works on
the GTDT approach to TPA (e.g., [1315]).
In order to relate blocked transmissibilities T D ij with standard measurable transmissibilities (hereafter termed global
transmissibilities and designated by T Gij ) we will follow an argumentation line very different from the original one in [13].
A mechanical system under harmonic excitation characterized by a dynamic stiffness matrix Z, its inverse H Z1 being
the system transfer function matrix, will be considered. The system dynamics will be driven as usual by Zx f or,
alternatively, by x Hf . First, it will be shown how global and direct transmissibilities can be expressed in terms of Z and
H. Once this is done, the link between T D G
ij and T ij arises quite straightforwardly saving several proof steps in previous works
(see e.g., [1315]). At this point and in what concerns notation, we remark that that the terms T D G
ij and T ij respectively stand
for direct and global transmissibilities from the j-th d.o.f. to the i-th d.o.f., in accordance with the standard convention used
for Zij and Hij. This has not been the option chosen in previous works on the GTDT approach to TPA [13,17,14,15,18], which
make use of the transposed notation commonly used for SEA coupling loss factors.
The global transmissibility T Gij simply corresponds to the ratio between the response at the i-th d.o.f. and the response at
the j-th d.o.f., when the latter is the only d.o.f. being excited (e.g., by impact hammer) and there is no blocking restriction
on the remaining system d.o.f.s. The system response vector to such a situation is given by
j j
x Z1 f Z1 0,0, . . . ,1, . . . ,0,0> H0,0, . . . ,1, . . . ,0,0> , 3
O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369 357
On the other hand, and as explained in the previous subsection, the direct transmissibility T D
ij , iaj, corresponds to the ratio
between the responses at the i-th d.o.f. and at the j-th d.o.f. when j is excited and all remaining d.o.f.s, except i and j,
become blocked. Think for instance of a system made of N masses connected by springs and dampers that can only move in
the vertical direction. The system dynamics is governed by Zx f but if we block the movement of all masses with the
exception of two masses connected one to the other, i and j, the dynamics of this reduced system when solely exciting j
will be described by the 2 2 matrix equation (see [15,18])
! !
Z ii Z ij xi 0
, 5
Z ji Z jj xj 1
which yields
Z ij Z ii
xi , xj : 6
Z ij Z ji Z ii Z jj Z ij Z ji Z ii Z jj
In what concerns the direct transmissibility from one subsystem to itself T D ii , it has been explained that it corresponds to
the quotient between the response at the i-th d.o.f. when it is excited and all the other d.o.f.s remain blocked, and the
response at i when it is excited but there is no restriction on the other d.o.f.s responses. If we block all d.o.f.s in Zx f but
the i-th one, where an external unit input is applied, this matrix system reduces to the single equation Z ii xi 1. Therefore,
the numerator of T D D
ii comes straightforwardly from the solution xi 1=Z ii to this equation. Besides, the denominator of T ii is
given by the response xi when i is the only d.o.f. being excited and we pose no restriction on the other system responses.
It then easily follows from the original non-reduced system (3) that xi Hii . Therefore we get,
1
TD
ii : 8
Z ii Hii
Whereas Z and H are symmetric matrices, it follows from (4), (7) and (8) that this will not be the case for the global and
direct transmissibilities matrices TG and TD . Note also that T Gii 1.
Eqs. (4), (7) and (8) offer a clear way to compute direct transmissibilities from measured global transmissibilities.
1
Writing (4) in matrix form we get TG H diag1=H11 . . . 1=HNN so that TG diagH11 . . . HNN Z. In components this reads
1
TG 9ij Hii Z ij : 9
Dividing T D D
ij in (7) by T ii in (8) and taking into account (9) gives
TD
ij 1
TG 9ij , iaj, 10
TD
ii
Eqs. (10) and (11) constitute the key relations we were looking for, as they show how to compute direct transmissibilities
1 1
from measured global transmissibilities. Taking into account that T D D G
ij T ii T 9ij 8iaj and that 1 T D
ii T
G
9ii , we can
DE
compact these expressions into a unique matrix equation by building a matrix T that has non-diagonal entries
1 1
T DE D D G
ij T ij T ii T 9ij 8iaj, and diagonal entries T DE D G
ii 1 T ii T 9ii . In words, TDE is nothing but the direct transmis-
sibility matrix with its diagonal replaced by 1 values. Writing these relations in matrix form yields
1
TDE KTD TG , 12
with KTD standing for a diagonal matrix whose entries are T D ii . Besides, the entries of T
DE
can be then expressed more
compactly as
Z ij Z ij
T DE D D
ij T ij dij T ij 1 dij 1 8i,j: 13
Z ii Z ii
358 O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369
Eq. (12) is the key relation of the GTDT approach to TPA and it has been derived herein following a very different
argumentation line from the original one in [13] (see also [14]). There is also a third possibility to prove it in the present
context, through the approach suggested in [18]. It has been included in Appendix A for completeness.
The computation of the direct transmissibility matrix involves the inversion of the measured global transmissibility
matrix, as seen from (12). This could lead to some numerical problems due to matrix ill-conditioning, especially when TG
involves different physical magnitudes [10]. Standard regularization techniques (see e.g., [3335]) can be applied to
mitigate these problems, although a presentation of them are out of the scope of this work (see [37,18] for some
applications in the framework of the GTDT method).
Let us next consider that f represents an operational external load acting on the system. If we were able to determine
that part of the operational response at any subsystem, xext
i , that is solely due to the external load fi acting on it, then it is
clear that we could use the global transmissibility matrix to nd the overall operational response at any d.o.f. as x TG xext
(which is the analogous to x Hf in classical force TPA). This precisely offers the opportunity to nd xext as
1
xext TG x: 14
G 1 DE
Taking into account (12) it follows that T K1
TD T and substituting in (14) we get
ext DE D
KTD x T x T KTD Ix, 15
DE D
where in the second equality we have used (12) written in matrix form, i.e., T T KTD I. Rearranging this expression
we recover the result in [1416]
x TD KTD x KTD xext 16
D
(note that T KTD has zero entries in the diagonal). Eq. (16) is nothing but a generalization of Eq. (2) that includes all d.o.f.
operational response decompositions using direct transmissibilities in a single expression. In other words, if we explicitly
expand any row in (16), we recover the breakdown in (2) for the operational response at any system d.o.f., xi, in terms of
the contributions of the other system d.o.f. operational responses xj , jai, and the part of the operational response at i, xext
i ,
that is solely due to the external load acting on it. Given that we can obtain xext through (14) using the measured
operational response x and the global transmissibility matrix TG , and that use can be made of the latter to compute the
direct transmissibilities in the rst step of the GTDT method, it becomes possible to perform the decomposition in (16)
with the sole measurement of operational responses. These measurements pose no special difculty when compared to
operational load measurements in classical force TPA.
On the other hand, the usual way to proceed to check the accuracy of a classical force TPA analysis consists in
comparing the measured operational response data with the result arising from the summation of the contributions of all
operational forces. In other words, one compares the l.h.s. of Eq. (1) with its r.h.s. If the TPA has been correctly performed,
the differences will be small and probably due to possible errors in the measurement of the frequency response functions
in the rst step of the method, or in the measurement of operational loads in the second step of the method. However, if
one has to proceed analogously to check the validity of the results for the GTDT method using (2), or its generalization (16),
one should only consider the response at those d.o.f.s with no operational external loads acting on them. This is so because
Eq. (16) is in fact an identity arising from (14), i.e., it is a physical meaningful way to rewrite it. Whereas in classical force
TPA we independently measure H and f , and then compare Hf with the set of independently measured operational
responses x, this is not the case for Eq. (16), where TG and x are the only measured quantities. The consequence is that any
error in the computed direct transmissibilities that affects the rst summand of the r.h.s. of (16) is automatically
compensated in the second summand, so as to match both sides of the equation. Therefore and as said above, the
appropriate way to check the accuracy of the GTDT method is to use the response at a d.o.f. with no external load acting on
it. The second summand in (16) will then have a zero value and the differences between the l.h.s. and r.h.s. of (16) could be
attributed to the inexactitude in the global transmissibility measurements, to errors in the computation of the blocked
transmissibilities, to the possible missing of d.o.f.s, or to other related sources of errors.
3. Experimental setup
To test the performance of the GTDT method, a mechanical set up made of masses, springs and dampers has been
considered. The system consists of a large mass connected to ground by means of four columns, each one made of a foam
massfoam connection as shown in Fig. 1a. A schematic representation of the system with all variables and parameters
taken into account is provided in Fig. 1b. The values for the masses as well as for the stiffness and damping of every foam
are given in Table 1. The foam parameters have been obtained experimentally, by means of individual tests using the
standard circle t method (see e.g., [38]).
O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369 359
Fig. 1. (a) Experimental set up, (b) Schematic representation and parameter identication.
Table 1
Data of the experimental set up.
Seven d.o.f.s have been considered for the mechanical system, these being the ve masses vertical displacements z1, z2,
z3, z4, z5, and mass m1 rotated angles yx and yy , with respect to the x and y axes (see Fig. 1b). In the rst step of the GTDT
method the global transmissibilities between all them will have to be measured. The system global transmissibility matrix
TG will then consist of a 7 7 matrix with entries
17
360 O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369
where subscripts have been used to designate matrix elements and superscripts to indicate the involved d.o.f.s in each
transmissibility. It can be observed that TG has a block structure containing displacementdisplacement transmissibilities,
angledisplacement and displacementangle transmissibilities, and angleangle transmissibilities. On the other hand, in
the second step of the GTDT method we will have to measure the operational generalized response vector
x z1 ,z2 ,z3 ,z4 ,z5 , yx , yy > : 18
G
The transmissibility matrix T in (17) and the operational response vector x in (18) are the only quantities that need to
be measured in the GTDT approach to transmission path analysis. Once we get TG , direct transmissibilities can be obtained
from (12) and used in (16) to breakdown the operational response vector x.
The seven d.o.f. experimental set up has been chosen to test some aspects of the GTDT method for various reasons.
Thanks to its limited size and number of d.o.f.s, the system is very treatable though at the same time it allows to check
several aspects related to global and direct transmissibilities. In particular, we will be interested in addressing three
aspects of the GTDT method. First, we will compare global and direct transmissibilities, and test to which extent those
direct transmissibilities that should theoretically be zero (e.g., those between the displacements of masses from different
columns) in fact they are. Second, we will aim at checking how good is the reconstruction of the response at a given d.o.f.
in terms of the other d.o.f. responses according to (16), for a problem that involves different types of transmissibilities (see
TG in (17)). Third and last, we will see whether direct and global transmissibilities get modied when blocking a given
d.o.f. of the system. Results for all these three points will be provided in Section 5. Finally, we note that another advantage
of dealing with the proposed experimental setup is that it is simple enough to build an analytical model for comparison
with the experimental results (see Section 4).
3.2. Measurements
We have claimed earlier that one of the key issues of the GTDT method is that it involves very simple measurements.
The global transmissibility matrix is easy to obtain and measuring operational response data poses, in general, no special
difculty. This is in fact the case for most vibroacoustics applications where measurements only involve displacements
(read also velocities and accelerations) and acoustic pressure (see e.g., [21,22]). However, the global transmissibility matrix
of our mechanical system (17) contains several terms that require angle measurements and in turn, the ability to excite
pure moments on mass m1, so as to measure the corresponding transmissibilities. It is well-known that the measurement
of angles and the excitation of pure moments in structures is rather intricate (see e.g., [39]), but for the particular example
at hand we can proceed as follows to circumvent some of the associated difculties.
On the one hand, exciting a pure moment by impact hammer on the upper mass m1 without inducing a vertical
displacement on it, becomes almost unfeasible. Yet, we need to know the vector response to such an excitation to nd
some of the transmissibilities in (17), according to the denitions in Eqs. (3) and (4). Let us focus, for instance, on a pure
u
x-axis unit moment excitation (analogous results would follow for the y-axis moment), and denote by xMx the
corresponding system response vector, which, as stated, cannot be measured due to our inability to excite pure moments.
However, if we apply a force f1 at a point located on the y-axis (nodal axis for y-axis moments), this will also result in an x-
u
axis moment M 1x r 1 f 1 , r1 standing for the lever arm. If we denote by xf 1 the system response to a unit force at the
barycenter of mass m1, it will follow by linearity that the system response to force f1 will be given by
u u
x1 f 1 xf 1 r 1 xMx : 19
Then, we just need to apply a force f2, with lever arm r2, to a second point located on the y-axis to get
u u
x2 f 2 xf 1 r 2 xMx : 20
Mux
From Eqs. (19) and (20), the system response to a pure x-axis moment excitation x can be nally obtained as
Mux f x2 f 2 x1
x 1 : 21
f 1 f 2 r 2 r 1
On the other hand, in what concerns the measurements of the rotated angle yx of mass m1 we have used the following
approach (the procedure being analogous for yy ). Two accelerometers have been placed on the x-axis, at the upper surface
and at the maximum distance apart from mass m1 barycenter. Let s1 and s2 denote the displacements measured at these
accelerometers. The vertical displacement of one of the edges of mass m1 with respect to its barycenter will be
ze 9s1 s2 9=2. In the limit of small angles sin yx yx , and the latter will be given by yx ze =r, r standing for half the width
of mass m1. All operations to obtain yx are carried out in the time domain, yx being Fourier transformed to the frequency
domain at the end of the process. Finally, we note that the transmissibilities in the rst block of matrix (17) do not involve
any angle and pose no special measurement difculty.
In what refers to measurement technical specications, transmissibility and operational response measurements have
been carried out using a multichannel spectrum analyzer OR38 set in recorder mode, with a sampling frequency of
51.2 kHz. Data are acquired from eight synchronized channels equipped with Visong Test 14100 ICP accelerometers
(see Fig. 1a). Five averaged impact hammer excitations have been considered for the transmissibilities and coherence
functions have been checked throughout the measurement process, to assess the linear behavior of the mechanical system.
O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369 361
A resolution of 0.5 Hz corresponding to 400 lines for the [0 200] Hz frequency range has been used. However, results
shown in Section 5 have been limited to the frequency range of interest, which comprises [12 120] Hz.
4. Analytical model
An analytical model can be derived quite straightforwardly for the mechanical setup presented in the previous section,
by establishing the force and torque balances for each mass. For the mass vertical displacements we will have (see Fig. 1b)
X int X
mi o2 zi fi f i , i 1 . . . 5, 22
int
with f i standing for the internal forces due to springs and dampers acting on each mass mi, and fi (with no superscript)
standing for the external ones. On the other hand, the torque balance for mass m1 will give
X int X
In o2 yn Mn M n , n x,y, 23
with In denoting the momenta of inertia of mass m1 with respect to the x and y axes, M int
n corresponding to the internal
momenta due to springs and dampers acting on mass m1, and Mn (with no superscript) denoting the external moments.
Grouping all inertia terms with internal forces and moments into a dynamic stiffness matrix Z results in
2 30 1 0 1
Z 11 Z 12 Z 13 Z 14 Z 15 Z 16 Z 17 z1 f1
6 Z 22 B
Z 23 Z 24 Z 25 Z 26 Z 27 7B z2 C B f 2 C
7 C B
6 C
6 7B C B C
6 Z 33 Z 34 Z 35 Z 36 Z 37 7 Bz C B f C
6 7B 3 C B 3 C
6 B
Z 44 Z 45 Z 46 Z 47 7B z4 C B f 4 C
7 C B
6 C, 24
6 7B C B C
6 Z 55 Z 56 Z 7
57 7BB z5 C B f C
6 C B 5 C
6 Z 66 Z 67 7 B C B C
4 Sym: 5@ y x A @ M x A
Z 77 yy My
the expressions for the dynamic stiffness matrix entries Zij being provided in Appendix B. The parameters used in the
computation of Zij are those of Table 1.
Once we have Z, we can compute its inverse H, and obtain the global and direct transmissibility matrices following the
procedure detailed in Section 2.2. These numerically computed transmissibilities, arising from the analytical model in (24),
will be compared to measured and processed data resulting from the application of the experimental GTDT method to the
mechanical setup.
5. Results
The tested mechanical system has seven eigenfrequencies. Three of them are located at 20.5 Hz, 25.5 Hz and 26 Hz and
correspond to the vertical displacement, x and y axis rotations of mass m1, whilst the remaining four can be found at
76.5 Hz, 87 Hz, 89 Hz and 91 Hz, and involve displacements of masses m2 to m5 in different combinations. Although peaks
at these modal frequencies will be present in most system transfer functions, this will not be necessarily the case when
plotting measured transmissibility functions. The reader is referred to [23,24,14] for explanations concerning peak and
valley locations in transmissibility functions.
In Fig. 2 we show some measured global transmissibilities together with their direct or blocked counterparts, computed
by means of Eqs. (10) and (11) (or (12) in matrix form). In Fig. 2a the logarithms of the squared moduli of the global and
direct transmissibilities, T G12 and T D 12 , between the displacement of mass m2 and mass m1 are plotted. Obviously, these
functions are necessarily different given that T G12 stands for the ratio between the responses x1 and x2, when m2 is excited,
whereas T D 12 corresponds to the same ratio but keeping all the remaining system d.o.f.s xed. A similar behavior can be
appreciated in Fig. 2b, for the transmissibilities between the displacement of mass m3 and the x-axis rotation of mass m1.
On the other hand, in Fig. 2c and d, we have respectively plotted the transmissibilities between masses m2 and m3, and
between masses m5 and m4. Given that neither m2 is directly connected to m3, nor m4 is connected to m5, the blocked
transmissibilities T D D
32 and T 45 should be zero and hence become minus innity in a logarithmic plot. However, this is not
the case for the direct transmissibilities in Fig. 2c and d, although they are clearly smaller than their respective global
transmissibilities T G32 and T G45 . It can be observed that the differences between direct and global transmissibilities are larger
than 10 dB for the whole frequency range in Fig. 2c. The same differences can be appreciated in Fig. 2d, except for the
frequencies comprising, more or less, from 70 Hz to 80 Hz. It is apparent that if a direct transmissibility that should be null
it is not, the operational response reconstruction (16) could get polluted. Whether or not the error in the reconstructed
operational displacement could be negligible will depend on how small the involved direct transmissibility becomes with
respect to all the other direct transmissibilities, and on the responses of the d.o.f.s it connects as well. For the present
example, it can be observed from Fig. 2 that those transmissibilities that should be null in Fig. 2c and d are considerably
smaller than the non-null ones in Fig. 2a and b, except again for the approximately [70 80] frequency range, where T D 45 in
Fig. 2d becomes larger than T G12 in Fig. 2a. To check if this has some polluting effect on the operational response
362 O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369
30 30
20 20
10 10
20log(|T|) [dB]
20log(|T|) [dB]
0 0
10 10
20 20
30 30
40 40
20 40 60 80 100 120 20 40 60 80 100 120
Frequency [Hz] Frequency [Hz]
30
30
20
20
10
10
20log(|T|) [dB]
20log(|T|) [dB]
0 0
10 10
20 20
30 30
40 40
20 40 60 80 100 120 20 40 60 80 100 120
Frequency [Hz] Frequency [Hz]
Fig. 2. Measured global transmissibilities versus direct transmissibilities. (a) T G12 : continuous, T D G D G
21 : dashed, (b) T 63 : continuous, T 63 : dashed, (c) T 32 :
continuous, T D G D
32 : dashed, (d) T 45 : continuous, T 45 : dashed.
reconstruction, we can mathematically force to zero the direct transmissibilities that should be null, by suppressing the
appropriate rows and columns in (12) and solving the system again (see [15,18]). We have already performed this
operation for the present example and nd no signicant departure from the original direct transmissibilities. The
pollution is thus negligible in this case.
On the other hand, in Fig. 3 a comparison is made between the direct transmissibilities obtained from measured
transmissibilities through Eqs. (10) and (11), and the direct transmissibilities obtained from the analytical model
presented in Section 4. The former are referred to as experimental direct transmissibilities in the gure legends.
The analytical ones have been computed by rst building the global transmissibility matrix using Eq. (4); the transfer
matrix entries in (4) being obtained from the inverse of the analytical dynamic stiffness matrix in (24). The direct
transmissibilities for the analytical model are nally computed using again (10) and (11). The results in Fig. 3 correspond
to different types of blocked transmissibilities. In Fig. 3a, the experimental and analytical displacement/displacement
direct transmissibilities between masses m2 and m1 are plotted, whereas in Fig. 3b angle/displacement direct
transmissibilities between mass m3 displacement and the angle yx are presented. Fig. 3c and d respectively contain
examples of direct transmissibilities from one d.o.f. to itself. This is so for the displacement of mass m3 in Fig. 3c, and for
the angle yx in Fig. 3d. In all cases, the resemblance between the experimental and analytical results is notorious, showing
that the analytical model represents to a good degree of precision the behavior of the mechanical set up and that, despite
of measuring errors and simplifying hypotheses, the direct transmissibilities computed from the global ones are quite
accurate.
The following operational situation is considered. External moments and a vertical force are applied to mass m1 by
means of hammer excitation at an off axis point of m1. Simultaneously, m2 becomes also excited with impacts on the
vertical direction. Operational response splittings as proposed in Eq. (16) are then provided for three different d.o.f.s,
namely the vertical displacements of masses m3, m4 and m1.
Results for mass m3 are given in Fig. 4. Fig. 4a and b corresponds to partial path contribution (PPC) plots (as termed in
[10]). The rst subgure shows the logarithm of the squared modulus of the measured displacement of mass m3, z3,
together with the moduli of the various displacement contributions T D 3i zi to z3. The second contains the same information
O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369 363
30
30
20
20
10
20log(|TD|) [dB]
|) [dB]
10
ij
20log(|TD
ij
0
10 10
20 20
30 30
20 40 60 80 100 120 20 40 60 80 100 120
Frequency [Hz] Frequency [Hz]
30 30
20 20
20log(|Tij |) [dB]
20log(|TD|) [dB]
10 10
D
ij
0 0
10 10
20 20
30 30
20 40 60 80 100 120 20 40 60 80 100 120
Frequency [Hz] Frequency [Hz]
Fig. 3. Experimental direct transmissibilities versus analytical direct transmissibilities. (a) Experimental T D D
12 : continuous, Analytic T 12 : dashed,
(b) Experimental T D D D D D
63 : continuous, Analytic T 63 : dashed, (c) Experimental T 33 : continuous, Analytic T 33 : dashed, (d) Experimental T 66 : continuous,
Analytic T D
66 : dashed.
but on a surface plot to help interpreting the results. As observed in Fig. 4a, the summation of all displacement
contributions (triangle symbols) almost perfectly match the measured displacement of m3 (continuous line). According to
the direct transmissibility breakdown in Eq. (16), and assuming that our mechanical set up behaves as predicted by the
analytical model (pure vertical displacement at the masses, etc.) only the d.o.f.s z1, yx and yy should contribute to z3.
As opposite, z2, z4 and z5 should have null contributions given that their respective direct transmissibilities T D D
32 , T 34 and T 35
D
should be zero. This is the tendency observed in Fig. 4a, being the moduli of the contributions of z1, yx and yy the
dominating ones. The fact that the modulus of the yx contribution exceeds the total displacement of m3 is not surprising,
given that a coherent reconstruction is being carried out and phase effects are crucial. This can be appreciated
in the phasor PPC plot of Fig. 4c, where contributions are given at a frequency of 40 Hz. As observed, the contributions
of yx and yy are almost out of phase with that of z1. On the other hand, if we have a look at the contributions of z2, z4 and z5
in Fig. 4a, we will appreciate that they moduli are of the order of 10 dB smaller than the dominating ones, almost for the
whole frequency range. Therefore, whatever their direct transmissibility phases could be relative to the other d.o.f. ones,
they will have negligible contributions to z3. Actually, we have checked that forcing the direct transmissibilities T D D
32 , T 34
D
and T 35 to zero in the system of equations does not result in any appreciable modication of the overall reconstructed
response z3.
In Fig. 5ac, we present the results for mass m4. Again, displacement contributions are provided, together with their
overall summation and with the displacement measured at m4. The two latter match to a high degree of precision. As for
the m3 case, contributions to z4 should only be due to z1, yx and yy , those of z2, z3 and z5 being negligible. This is in fact
what is observed. It can be appreciated that yx is the main contributor for almost the whole frequency range, followed by
z1 and yy . The contributions of z2, z3 and z5 are clearly inferior to them as expected. In this case, no particular d.o.f. has a
modulus contribution higher than the overall reconstructed value given that the dominating d.o.f.s are practically in phase
for the whole frequency range (see e.g., Fig. 5c for phasor contributions at 40 Hz).
Finally, in Fig. 6 the response of mass m1 is decomposed as the summation of the contributions from all other system
P5 D D D D ext
d.o.f. responses, j 2 T 1j zj T 16 yx T 17 yy , plus the term T 11 z1 corresponding to the external excitation acting on it.
The latter is absent in the reconstruction of the responses z3 and z4 in Figs. 4 and 5, given that no external force is acting on
masses m3 and m4. As commented in Section 2.3, the term T D ext
11 z1 would exactly compensate any error in the total
displacement contribution, so that their summation exactly coincides with the measured response z1. Therefore, in order
to validate the method, the comparison between measured and reconstructed responses should be made for masses with
no external force acting on them such as m3 and m4.
364 O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369
Fig. 4. Operational displacement reconstruction versus measured one for mass m3. (a) Line PPC plot. Measured z3: continuous, Reconstructed z3:
triangles-,, T D D D D D D
31 z1 : dashed, T 32 z2 : continuous-J, T 34 z4 : continuous-&, T 35 z5 : continuous-%, T 36 yx : dotted, T 37 yy : dash-dotted, (b) Surface PPC plot,
(c) Phasor PPC plot at 40 Hz.
O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369 365
40
20
20
40
60
D
T
47 y 40
TD
46 x
20
T D z5
45
D 0
T z
43 3
D
T 42 z
2 20
TD z
41 1
40
Reconstructed z4
Measured z4 60
6e007
150 30
4e007
Measured z 2e007
4 D
Reconstructed z T41z1
4
180 D 0
T46x D
T47y
210 330
240 300
270
Fig. 5. Operational displacement reconstruction versus measured one for mass m4. (a) Line PPC plot. Measured z4: continuous, Reconstructed z4:
triangles-,, T D D D D D D
41 z1 : dashed, T 42 z2 : continuous-J, T 43 z3 : continuous-&, T 45 z5 : continuous-%, T 46 yx : dotted, T 47 yy : dash-dotted, (b) Surface PPC plot,
(c) Phasor PPC plot at 40 Hz.
366 O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369
We can gain some more insight into the behavior of direct transmissibilities by blocking one of the system masses
(see [15]). According to the denitions of global and direct transmissibilities, when this is done any global transmissibility
40
20
Displ. [dB ref. 1e6 m]
20
40
60
20
20
10
10
0
20log(|T|) [dB]
20log(|T|) [dB]
10 10
20 20
30 30
40 40
20 40 60 80 100 120 20 40 60 80 100 120
10 10
0 0
20log(|T|) [dB]
20log(|T|) [dB]
10 10
20 20
30 30
40 40
20 40 60 80 100 120 20 40 60 80 100 120
Frequency [Hz] Frequency [Hz]
Fig. 7. Results for system with blocked mass m3 (BM stands for blocked mass in legends). (a) Experimental BM T G14 : continuous, Analytical BM T G14 :
dashed, Experimental T G14 : dotted, (b) Experimental BM T D D D D
55 : continuous, Analytical BM T 55 : dashed, Experimental T 55 : dotted, (c) Experimental BM T 14 :
continuous, Analytical BM T D 14 : dashed, Experimental T D
14 : dotted, (d) Experimental BM T D
27 : continuous, Analytical BM T D
27 : dashed, Experimental T D
27 :
dotted.
O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369 367
between a pair of system d.o.f.s automatically changes. The same holds true for a direct transmissibility from one d.o.f. to
itself, given that the denominator in (8) involves the d.o.f. response when no blocking restriction is applied to the
remaining of the system. The opposite occurs for direct transmissibilities between any pair of d.o.f.s, different from the
blocked one. This is so because the denition of direct transmissibility between i and j already implies that all d.o.f.s
different from i and j are already blocked. Hence, T Dij cannot be affected when a d.o.f. k, with kai,j, is physically blocked.
To check the degree of accuracy of this point in our mechanical system, we have blocked mass m3. Then we have
measured all new global transmissibilities for the reduced system (now only having six d.o.f.s) and recomputed the system
direct transmissibilities. We have also blocked mass m3 in the analytical model and obtained the new analytical direct
transmissibilities. Results are presented in Fig. 7 and compared to those of the original system.
In Fig. 7a the logarithm of the squared moduli of the new experimental and analytical global transmissibility T G14 are
plotted, showing great resemblance. Both clearly differ from the same global transmissibility of the original system (dotted
curve), as expected. As mentioned above, the same should happen for direct transmissibilities from one d.o.f. to itself.
This can be appreciated in Fig. 7b, where it is shown how the new experimental and numerical T D 55 are close one to the
other but again, they are noticeably different from their counterpart in the original system. In contrast, if we have a look at
direct transmissibilities not involving mass m3, no discrepancies should be observed between the new and original system
ones. This can be readily observed in Fig. 7c and d, where results are presented for T D D
14 and T 27 . In both cases, the three
direct transmissibilities in each gure look extremely similar.
6. Conclusions
Tests on the GTDT approach to classical transfer path analysis have been performed on a simple mechanical setup of
seven degrees of freedom, for which an analytical model has been built. The system is made of masses, springs and
dampers. Working with such a controlled experiment has allowed to verify several issues related to the concept of direct
transmissibility. For instance, it has been shown experimentally how direct transmissibilities allow to breakdown the
response of a system d.o.f. in terms of the other system d.o.f.s and of its response to the external force acting on it. We have
checked the accuracy in doing so by analyzing the behavior of those direct transmissibilities that should acquire negligible
values. Moreover it has been experimentally conrmed that direct transmissibilities behave as predicted when blocking a
system d.o.f.
The GTDT approach to classical transfer path analysis constitutes an alternative to the more standard force TPA
approach. It involves much easier measurements, although probably, more intricate postprocessing. The herein presented
tests are intended to clarify some of the latter. The obtained results look fairly promising, indicating that the method could
start being applied to more complex situations of practical interest, with condence.
X D G
T Gij T jk T ki 8iaj, A:2b
jak
which are the original relations between direct and global transmissibilities derived in [13], from a different line of
reasoning. These equations can also be derived in the present context following the approach suggested in [18], as an
alternative to that in the main text.
Let us focus on Eq. (A.2a). Expanding the determinant of the dynamic stiffness matrix Z by minors using the i-th row
yields
X
N X
det Z 1i j Dij Z ij Z ii Dii 1i j Dij Z ij , A:3
i1 iaj
with Dij standing for the (i,j)-th minor of Z. Given that H Z1 , it follows:
1i j Dij
Hji Z 1
ji : A:4
det Z
Dividing (A.3) by det Z and taking into account (A.4) gives
X
1 Z ii Hii Z ij Hji : A:5
iaj
368 O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369
Finally, dividing (A.5) by Z ii Hii and using relations (4), (7) and (8) in text results in (A.2a). Eq. (A.2b) can be obtained
following a very similar procedure (see [18]).
For any connection between two masses in the model, mi and mj, characterized by a spring kij and a damper cij let us
dene
k ij kij iocij : B:1
Similarly, for a mass connected to the ground we will have k ig kig iocig . The entries Zij for the dynamic stiffness matrix
appearing in Eq. (24) in text can then be written as follows.
The diagonal terms become
X
5
Z 11 k 1n m1 o2 , B:2
n2
Z ii k 1i k ig mi o2 i 2 . . . 5, B:3
X
5
Z 66 r 2xn k 1n Ix o2 , B:4
n2
X
5
Z 77 r 2yn k 1n Iy o2 , B:5
n2
Z 17 r y2 k 12 r y3 k 13 r y4 k 14 r y5 k 15 ,
Z 2j 0, j 3 . . . 5, Z 26 r x2 k 12 , Z 27 r y2 k 12 , B:7
Z 3j 0, j 4,5, Z 36 r x3 k 13 , Z 37 r y3 k 13 , B:8
Z 45 0, Z 46 r x4 k 14 , Z 47 r y4 k 14 , B:9
Z 56 r x5 k 15 , Z 57 r y5 k 15 , B:10
Z 67 r x2 r y2 k 12 r x3 r y3 k 13 r x4 r y4 k 14 r x5 r y5 k 15 : B:11
In the above expressions, rxn and ryn respectively stand for the x and y components of the lever arm corresponding to the
force applied by the spring-damper k 1n at its connection point with mass m1.
References
[1] C.J. Dodds, J.D. Robson, Partial coherence in multivariate random processes, J. Sound Vib. 42 (2) (1975) 243249.
[2] J.S. Bendat, Solutions for the multiple input/output problem, J. Sound Vib. 44 (3) (1976) 311325.
[3] J.S. Bendat, System identication from multiple input/output data, J. Sound Vib. 49 (3) (1976) 293308.
[4] R. Potter, Matrix formulation of multiple and partial coherence, J. Acoust. Soc. Am. 61 (1977) 776781.
[5] O. Guasch, F.X. Magrans, A compact formulation for conditioned spectral density function analysis by means of the LDLH matrix factorization, J.
Sound Vib. 277 (45) (2004) 10821092.
[6] W. Stahel, R.H. Van Ligten, J. Gillard, Measuring Method to Obtain the Transmission Paths and Simultaneous Real Force Contributions in a
Mechanical Linear System, Technical Report No. 80.21, Lab. Acustico Italiana Keller, 1980.
[7] J. Verheij, Multipath Sound Transfer from Resiliently Mounted Shipboard Machinery, Ph.D. Thesis, Technische Physische Dienst TNO-TH, Delft, 1986.
[8] H.R. Tschudi, The force transmission path method: an interesting alternative concerning demounting tests, in: Unikeller Conference 91 (1991) 1991.
[9] Transfer path analysis: the qualication and quantication of vibro-acoustic transfer paths, in: LMS International. Application Notes, 1995.
[10] P. Gajdatsy, K. Janssens, W. Desmet, H. Van der Auweraer, Application of the transmissibility concept in transfer path analysis, Mech. Syst. Signal
Process. 24 (7) (2010) 19631976.
[11] A. Inoue, R. Singh, G. Fernandes, Absolute and relative path measures in a discrete system by using two analytical methods, J. Sound Vib. 313 (35)
(2008) 696722.
[12] A. Gunduz, A. Inoue, R. Singh, Estimation of interfacial forces in time domain for linear systems, J. Sound Vib. 329 (13) (2010) 26162634.
[13] F.X. Magrans, Method of measuring transmission paths, J. Sound Vib. 74 (3) (1981) 311330.
[14] O. Guasch, F.X. Magrans, The Global Transfer Direct Transfer method applied to a nite simply supported elastic beam, J. Sound Vib. 276 (12)
(2004) 335359.
[15] O. Guasch, Direct transfer functions and path blocking in a discrete mechanical system, J. Sound Vib. 321 (35) (2009) 854874.
[16] F.X. Magrans, O. Guasch, The role of the Direct Transfer Function Matrix as a connectivity matrix and application to the Helmholtz equation in two
dimensions: Relation to numerical methods and free eld radiation example, J. Comput. Acoust. 13 (2) (2005) 341363.
[17] F.X. Magrans, Denition and calculation of transmission paths within a SEA framework, J. Sound Vib. 165 (2) (1993) 277283.
O. Guasch et al. / Mechanical Systems and Signal Processing 37 (2013) 353369 369
[18] O. Guasch, A direct transmissibility formulation for experimental statistical energy analysis with no input power measurements, J. Sound Vib. 330
(25) (2011) 62236236.
[19] F. Bessac, L. Gagliardini, L. Guyader, Coupling eigenvalues and eigenvectors: a tool for investigating the vibroacoustic behaviour of coupled vibrating
systems, J. Sound Vib. 191 (5) (1996) 881899.
[20] J. Mixson, J. Wilby, Interior noise, in: H. Hubbard (Ed.), Aeroacoustics of Flight Vehicles. Theory and Practice, Noise Control, vol. II, Acoustical Society
of America, 1995.
[21] F.X. Magrans, P.V. Rodrguez, G.C. Cousin, Low and mid-high frequency advanced transmission path analysis, in: Proceedings of the Twelfth
International Congress on Sound and Vibration ICSV12 (2005), Lisbon, Portugal, 2005.
[22] J. Jove, F. Guerville, A. Vallespn, Study of the aerodynamic contribution to the high-speed train cabin internal noise by means of hybrid FE-SEA
modeling, in: InterNoise 2010, Noise and Sustainability, June 1316, Lisbon (Portugal), 2010.
[23] N.M.M. Maia, J.M.M. Silva, A.M.R. Ribeiro, The transmissibility concept in multi-degree-of-freedom systems, Mech. Syst. Signal Process. 15 (1) (2001)
129137.
[24] A.M.R. Ribeiro, J.M.M. Silva, N.M.M. Maia, On the generalization of the transmissibility concept, Mech. Syst. Signal Process. 14 (1) (2000) 2935.
[25] C. Devriendt, P. Guillaume, The use of transmissibility measurements in output-only modal analysis, Mech. Syst. Signal Process. 21 (7) (2007)
26892696.
[26] C. Devriendt, G. De Sitter, S. Vanlanduit, P. Guillaume, Operational modal analysis in the presence of harmonic excitations by the use of
transmissibility measurements, Mech. Syst. Signal Process. 23 (3) (2009) 621635.
[27] F. Magalha~ es, A. Cunha, Explaining operational modal analysis with data from an arch bridge, Mech. Syst. Signal Process. 25 (5) (2011) 14311450.
[28] K. Noumura, J. Yoshida, Method of transfer path analysis for vehicle interior sound with no excitation experiment, in: Proceedings of FISITA 2006,
F2006D183, Yokohama, Japan, 2006.
[29] M. Lohrmann, Operational transfer path analysis: comparison with conventional methods, in: Fifteenth ICVS Conference, Daejon, Korea, 2008.
[30] H. Van der Auweraer, P. Mas, S. Dom, A. Vecchio, K. Janssens, P. Van de Ponseele, Transfer path analysis in the critical path of vehicle renement: the
role of fast, hybrid and operational path analysis, in: Proceedings of the SAE Noise and Vibration Conference, Paper 2007-01-2352, May 1417, 2007.
[31] G. De Sitter, C. Devriendt, P. Guillaume, E. Pruyt, Operational transfer path analysis, Mech. Syst. Signal Process. 24 (2) (2010) 416431.
[32] K. Janssens, P. Gajdatsy, L. Gielen, P. Mas, L. Britte, W. Desmet, H. Van der Auweraer, OPAX: a new transfer path analysis method based on parametric
load models, Mech. Syst. Signal Process. 25 (4) (2011) 13211338.
[33] A. Thite, D. Thompson, The quantication of structure-borne transmission paths by inverse methods. Part 1: improved singular value rejection
methods, J. Sound Vib. 264 (2003) 411431.
[34] A. Thite, D. Thompson, The quantication of structure-borne transmission paths by inverse methods. Part 2: use of regularization techniques,
J. Sound Vib. 264 (2003) 433451.
[35] H. Choi, A. Thite, D. Thompson, Comparison of methods for parameter selection in Tikhonov regularization with application to inverse force
determination, J. Sound Vib. 304 (2007) 894917.
[36] O. Guasch, A. Aragone s, Finding the dominant energy transmission paths in statistical energy analysis, J. Sound Vib. 330 (10) (2011) 23252338.
[37] O. Guasch, Regularization and Resampling Methods to Improve the Accuracy of the Global Transfer Direct Transfer Method of Transmission Path
Analysis, Technical Report. ICR No 21-09/2004, 2004.
[38] Z. Fu, J. He, Modal Analysis, Butterworth-Heinemann, 2001.
[39] Y. Champoux, V. Cotoni, B. Paillard, O. Beslin, Moment excitation of structures using two synchronized impact hammers, J. Sound Vib. 263 (3) (2003)
515533.