Engineering Structures 142 (2017) 192210

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Truncation error analysis on modal flexibility-based deflections:

application to mass regular and irregular structures
Giacomo Bernagozzi , Luca Landi, Pier Paolo Diotallevi
Department DICAM, University of Bologna, Bologna, Italy

a r t i c l e i n f o a b s t r a c t

Article history: It is of interest in the fields of vibration-based structural identification and damage detection to ana-
Received 21 June 2016 lyze the truncation effects introduced on modal flexibility (MF) based deflections that are estimated
Revised 24 March 2017 using only a subset of structural modes. To address this problem, an approach for truncation error
Accepted 27 March 2017
analysis on MF-based deflections of structural systems subjected to a generic load is proposed in
this paper. The approach is based on the determination of the relative contribution of each mode
to the deflection by means of a proposed load participation factor (LPF). This factor, as derived ana-
Keywords:
lytically, depends both on the applied load and on the distribution of the structural masses. The val-
Flexibility truncation error
Modal flexibility-based deflection
idation of the proposed approach was carried out both on numerical models of shear-type frame
Load participation factor buildings and on experimental data of a steel frame structure tested under ambient vibrations
Uniform load (i.e. the benchmark study sponsored by the IASC-ASCE Task Group on SHM). In both cases, results
Mass proportional load show that the LPF factors can give an a priori indication of the truncation effects expected on
Mass irregular structure the MF-based deflections. The relationship between the proposed approach and the approach based
on the mass participation factors, introduced by Zhang and Aktan (1998) for the case of uniform
load (UL) deflections, is discussed since the two approaches are equal only if a special load, which
is a mass proportional load (MPL), is considered. Thus, the application of this MPL load for mass
irregular structures is also investigated. Numerical analyses performed both on a shear-type frame
building and on a simply-supported beam, showed that for the great majority of the analyzed con-
figurations, the truncation errors on the MF-based deflections due to the MPL are lower compared to
those related to the UL.
2017 Elsevier Ltd. All rights reserved.

1. Introduction on the principle that relevant damages in structures lead to

Recent advances in vibration-based testing technologies lead to
the possibility of applying the techniques of the emerging field of
Structural Health Monitoring (SHM) to existing civil structures
and infrastructures. These testing techniques together with the
application of structural system identification and damage
detection algorithms can be adopted in order to evaluate possible
structural deteriorations and support the decision-making on
maintenance and retrofit programs.
Among different approaches for vibration-based damage
detection (VBDD) [15], modal-based methods are founded
Abbreviations: LPF, load participation factor; MF, modal flexibility; MPF, mass
participation factor; MPL, mass proportional load; MPL-PF, mass proportional load
participation factor; STFB, shear-type frame building; SSB, simply-supported beam;
UL, uniform load; UL-PF, uniform load participation factor.
Corresponding author.
E-mail address: giacomo.bernagozzi2@unibo.it (G. Bernagozzi).
changes in their dynamic properties. Such properties (i.e.
modal parameters) can be derived using modal identification
techniques from the measurements of ambient or forced
vibrations of the structures [68]. Moreover, modal flexibility
(MF) matrices can be assembled starting from identified
modal data and, as shown in [9], the coefficients of such
matrices are more sensitive to damages than the modal
parameters individually. Due to this, promising VBDD methods
are based on the calculation of MF matrices [1013], includ-
ing methods specifically developed for the case of output only
identification [1416].
As defined by Zhang and Aktan [17], the uniform load (UL)
surface is the deflection profile that can be calculated from
experimental MF matrices by applying uniform loads at all
the DOFs of a structure. This surface can be adopted as a
damage-sensitive feature [1719] since eventual modifications
in the UL deflection, when comparing at least two structural

http://dx.doi.org/10.1016/j.engstruct.2017.03.057
0141-0296/ 2017 Elsevier Ltd. All rights reserved.
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210
states, can be related to structural damages, such as stiffness
reductions. Similar approaches have been specifically devel-
oped for the damage detection of shear-type buildings [20
23] using special loads, termed positive shear inspection loads.
These methodologies have also been applied to beam-like
structures [24].
In modal identification of civil structures the high-order
modes cannot be extracted or are poorly estimated. Thus, the
use of such incomplete modal models [25] for assembling modal
flexibility (MF) matrices leads to inevitable errors, which are
defined as flexibility truncation errors [17], with respect to static
flexibility matrices. These errors affect also the uniform load
deflection which is calculated from MF matrices. As also defined
in [17], truncation error analysis is related to the investigation
of how many modes need to be included in order to obtain good
estimates of the MF matrix and the UL deflection derived from
experimental data. This analysis is commonly adopted in pretest
design using analytical or numerical models in order to determine
the frequency range to be tested, so that the truncation errors can
be minimized [17].
As indicated by Zhang and Aktan in [17], there exist three
main approaches for truncation error study: the first one is to
consider a number of modes such that the cumulative mass par-
ticipation factor of the structure is higher than a selected thresh-
old; the second approach is to directly compare the truncated MF
matrix with the exact matrix; the third one is equal to the second
approach but the UL deflection is considered instead of the MF
matrix. The mentioned authors [17] performed a series of analy-
ses using the second and the third approaches, on a 10 DOF mass-
spring model and a bridge FEM model, demonstrating that the UL
deflection is less sensitive to truncation errors than the MF
matrix.
The first objective of the present work is to propose an
approach for truncation error analysis that is applicable to modal
flexibility-based deflections of structural systems subjected to a
generic load. The main advantage of the proposed approach is
that it does not imply a direct comparison between the trun-
cated and the exact solutions, and it can be used to obtain an
a priori estimate of the truncation effects expected on the deflec-
tions. In particular, it is demonstrated that the proposed
approach is equal to the first approach for truncation error anal-
ysis, not applied in the numerical analyses of Zhang and Aktan
[17], only if a special load, which is a mass proportional load
(MPL), is considered. The second objective is then to compare
the proposed mass proportional load with the uniform load
[17], by evaluating the corresponding truncation errors on the
MF-based deflections of structures characterized by mass
irregularities.
The paper is organized as follow. At first the proposed gen-
eral approach for truncation error analysis is outlined, and then
the analytical expressions for evaluating the truncation errors
on the interstory drifts of the MF-based deflections of shear-
type frame buildings (a special case) are derived. Secondly,
the validation of the proposed approach, carried out both using
numerical simulations and experimental data, is presented.
Numerical models of shear-type frame buildings were employed
for this validation. In addition, the approach was verified on a
steel frame structure tested under ambient vibrations, using the
experimental data of the benchmark problem (phase II) pro-
vided by the IASCASCE Task Group on Structural Health Mon-
itoring [26,27]. Finally, numerical analyses performed to
compare the truncation errors on the deflections due to the
MPL load with those related to the UL load are presented.
These analyses were conducted on mass irregular structures,
considering both a shear-type frame building and a simply-
supported beam.
193
2. Theoretical background
2.1. Modal flexibility matrix and uniform load deflection estimated
from vibration data
According to the equations of motion and the dynamic charac-
teristic equations of an undamped MDOF structural system, the
modal flexibility matrix F r n
n can be expressed using r modes as
F r Ur K1 T
r Ur 1
where Ur n
r is the mass-orthogonal and mass-normalized mode
shape matrix, Kr r
r is the spectral matrix, which contains the
square of the first r natural circular frequencies xi2 on the main
diagonal, i = 1. . .r is the mode index, and n is the number of the
DOFs. Each component of the matrix is expressed as
X
r
/j;i /k;i
f j;k with j; k 1 . . . n 2
i1
x2i
Each term in the summation is independent from the global
sign of the i-th mode shape, i.e. it does not change if the mode
shape is multiplied by 1, and only the diagonal components of
the matrix are always positive quantities [17]. If r = n and consid-
ering exact modal data, the static flexibility matrix F n is obtained.
If r < n each component of the matrix is approximated with respect
to the exact value [25], and the difference between the two terms
is called residual flexibility [28]. Nevertheless, accurate estima-
tions of the flexibility matrix can be obtained using the first modes
since the contribution of each i-th mode to the matrix depends on
the term 1/xi2 [15,25]. Adopting Eq. (1), experimental modal flex-
ibility matrices can be assembled starting from modal parameters
extracted using system identification techniques [68] from the
measurements of ambient or forced vibrations of structures. Sev-
eral procedures can be applied in order to obtain the required
mass-normalized mode shapes both for the case of input-output
[7,29] or output-only modal identification [14,15,2023,30,31].
For the particular case of structures with closely spaced modes, a
multiple input test and suitable modal identification techniques,
available in the literature, are required to extract the structural
modes [8]. However, as described in [32,33], the case of closely
spaced modes inherits some relevant properties from the theoret-
ical case of repeated eigenvalues. One of these properties is that,
according to the dynamic characteristic equations of an undamped
system, the orthogonality relation between two modes character-
ized by the same eigenfrequencies is not assured [32,33]. A similar
property may thus be related to identified closely spaced modes,
and issues may arise if modes characterized by such property are
employed to calculate the modal flexibility matrices using Eq.
(1). With reference to this last point, however, it is worth noting
that there exist examples in the literature where the modal flexi-
bilities were adequately estimated in presence of closely spaced
modes (one of these examples is reported in [34]).
The deflection of an MDOF structure calculated starting from
the modal flexibility matrix F r and due to a uniform load (UL)
u 1 1 . . . 1T is defined as the uniform load surface [17], and it
is expressed as xu;r F r u where the j-th component of the deflec-
tion vector xu;r is
!
X
r
/j;i X
n X
r
/j;i
xu;r;j /k;i si 3
i1
x 2
i k1 i1
x2i
The convergence of the UL deflection to the exact solution is
more rapid with respect to the components of the modal flexibility
P
matrix due to the presence in Eq. (3) of the term si nk1 /k;i .
Indeed, the summation of the i-th mode shape coefficients can be
considered as the contribution of that mode, and in general higher
194 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210

values are obtained for the lower modes [17]. In addition, the con- both the structural masses and the mode shape components
tribution of a single mode in Eq. (3) does not depend on the global are included.
sign of the mode shape and, due to the double summation effect The deflection xp;r of a generic MDOF structure due to a generic
that is present in Eq. (3), the UL deflection is less sensitive to exper- load p can be determined starting from the modal flexibility matrix
imental errors than the deflection of the structure evaluated for a F r assembled using r modes
single load at only one DOF [17].
xp;r F r p 8
2.2. Truncation error analysis on uniform load deflections Pn
Introducing the term cp;i k1 pk /k;i , which is the work done
by the external load p for the modal displacements of the i-th
Truncation error analysis, according to the first approach men-
mode shape /i , the j-th component of the deflection is expressed as
tioned by Zhang and Aktan [17], can be performed on numerical !
structural models by checking if, for the considered r modes, the X
r
/j;i X
n X
r
/j;i
xp;r;j pk /k;i cp;i 9
cumulative mass participation factor (MPF) of the structure
P i1
x 2
i k1 i1
x2i
lr ri1 li is higher than a selected threshold (e.g. higher than
90%), where The expression for computing the modal flexibility-based
2 deflection using r modes i.e. Eq. (8) can be reformulated by con-
C
li i
4 sidering the static flexibility matrix F n and an r-mode equivalent
CT C load pr such that
is the mass participation factor related to the i-th mode, and xp;r F n pr 10
Pn
Ci j1 mj /j;i is the i-th modal participation factor if a structure
with a diagonal mass matrix and mass-normalized mode shapes In this way, all the truncation errors are included in the load and
are considered. In addition, in this case it is worth noting that the not in the flexibility matrix. By comparing the terms on the right
P hand side in Eqs. (8) and (10) the r-mode equivalent load can be
denominator of Eq. (4) is CT C mtot nj1 mj , where mtot is the
obtained
total mass of the structure and mj is the mass related to the j-th
DOF. pr F 1
n Frp 11
Alternatively, truncation error analysis on modal flexibility-
Since the modal flexibility matrix is assembled using the contri-
based deflections can be performed by directly comparing the P
bution of each single mode F i i.e. F r ri1 F i the r-mode
approximated components of such deflections with the true ones Pr
(i.e. the third approach mentioned in [17]). The relative error on equivalent load can be expressed as pr i1 pi , where pi is a
the j-th component of the MF-based deflection due to the uniform single-mode equivalent load defined as
load and assembled using r modes (e.g. the error on the displace-
pi F 1
n F i p 12
ment xu;r;j ), is
xu;r;j  xu;n;j Substituting Eq. (1) evaluated for r = n in place of F n and Eq. (1)
ex;u;r
j
5 evaluated for the i-th mode only i.e. F i /i x12 /Ti in Eq. (12),
xu;n;j i

and due to the orthogonality relationships of the mode shapes with

where xu;n;j is the exact displacement computed from the static flex-
ibility matrix F n . Moreover, to have an estimation of the truncation respect to the mass matrix i.e. UTn MUn I, where M n
n is the
errors on the whole deflection of the structure, the root-mean- mass matrix and I n
n is the identity matrix the single-mode
square (RMS) criterion can be applied to all the single-DOF trunca- equivalent load is reformulated as
tion errors ex;u;r
j
1 1
v pi UT 1
n KUn /i /Ti p MUn KUTn M/i /Ti p M/i /Ti p 13
u X x2i x2i
u1 n j2
eRMS
x;u;r t ex;u;r 6
n j1 Thus, the modal load is independent from the natural circular
frequency xi , and it depends only on the mass matrix, the i-th
In addition, the maximum of the single-DOF truncation errors mode shape and the assumed load. Considering structures with a
on the deflection, in terms of absolute values, can also be evaluated diagonal mass matrix, each element of the modal load vector pi
is expressed as
eMAX
x;u;r max jex;u;r j
j
7
!
X
n
pi;j mj /j;i pk /k;i mj /j;i cp;i 14
3. Proposed approach for truncation error analysis on modal- k1
flexibility based deflections due to a generic load
The normalized modal contribution of each mode in the modal
An approach for truncation error analysis on modal flexibility- decomposition of the applied load p is defined as the load partici-
based deflections of structural systems subjected to a generic pation factor (LPF) and is expressed as
load is proposed herein. The approach has been formulated with pi;TOT cp;i Ci
reference mainly to plane structures, and considering the modal vp;i T 15
pTOT cp C
flexibility-based deflections of these structures in one prevalent
Pn
direction (which is the direction of the applied load). The reason where pTOT j1 pj is the summation of all the components of the
for proposing an approach valid for a generic load is due to the P
assumed load p and pi;TOT nj1 pi;j is the summation of all the
preliminary observation that follows. On one side, each modal
components of the single-mode equivalent load pi . Due to Eq.
term in Eq. (3), which is related to the uniform load deflection,
(14), this last term is also expressed as
depends on the modal contribution si , which is the summation
of the i-th mode shape components. On the other side, the trun- X
n X
n
cation error approach based on the cumulative mass participa- pi;TOT mj /j;i cp;i cp;i mj /j;i cp;i Ci 16
tion factor [17] involves the participation factor Ci , in which j1 j1
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210 195

Moreover, as demonstrated in the Appendix A, the total load is MPL, the MPL-PF vm;i is a positive quantity. Moreover, for all the
the scalar product between the vectors cp and C, which contain the possible values of r the following relationship holds: 0 6 vm;r 6 1.
terms cp;i and Ci respectively It is worth noting that these conditions, valid for the MPL, are
not always fulfilled for generic loads, including the uniform load.
pTOT c Tp C 17
In order to make a comparison with the mass proportional load,
Starting from this, the contribution of the first r modes in the a scaled version of the uniform load can be considered. This mod-
P
modal decomposition of the applied load, is defined as the cumu- ified uniform load is defined as pu m a, where m 1n nj1 mj is
lative load participation factor the average mass of the MDOF structure. Adopting this load, each j-
Pr th modal flexibility-based displacement is equal to Eq. (3) scaled
pr;TOT X r
i1 cp;i Ci
vp;r vp;i 18 by the factor am , where a is a unitary acceleration. The proposed
pTOT i1
c Tp C load participation factor evaluated for the uniform load (UL-PF)
P P and for the i-th mode is defined as
where pr;TOT nj1 pr;j ri1 pi;TOT is the summation of all the
components of the r-mode equivalent load pr . Since all the trunca- cu;i Ci si Ci
vu;i T 21
tion errors are included in the load, according to Eq. (10), the pro- c Tu C s C
posed cumulative LPF evaluated for the first r modes is also
where if the mode shapes are mass normalized, then sT C n. In
related to the contribution of such modes to the modal flexibility-
the same way, the cumulative UL-PF is indicated as vu;r .
based deflection. Moreover, it is evident that if all the modes are
included (i.e. r = n), the cumulative LPF becomes vp;n 1. Due to
this, the overall truncation error that occurs on the whole deflection 5. Analytical expressions for truncation error analysis on modal
evaluated using the first r modes can be related to the cumulative flexibility-based deflections of shear-type frame buildings
LPF of the modes, from r + 1 to n, that are not included in the calcu- (STFB)
lation, i.e. v
 p;r 1  vp;r . Moreover, it is worth noting that if the
missing contribution to a target solution is positive, the related The problem of evaluating the truncation errors on modal
truncation error is necessarily negative. Thus, in general the overall flexibility-based deflections is specifically addressed for plane
truncation error on the whole deflection can be related to the term shear-type frame buildings (STFB). Firstly, the analytical expres-
vp;r  1, computed by changing the sign of v p;r . Finally, if the estima- sion for evaluating the truncation errors on the interstory drifts
tion of the overall truncation error on the deflection is computed computed from modal flexibility matrices and due to a generic load
using the RMS criterion or by evaluating the maximum of the applied to these structures is derived. Secondly, for the special
single-DOF truncation errors in terms of absolute values, according case of the displacement related to the first DOF (i.e. the lowest
to Eqs. (6) and (7) but applied for the generic load p, the above- one) of STFB structures, which is also equal to the drift of the first
mentioned term can be properly considered with the absolute story, the approach proposed in Section 3 and related to the overall
value, i.e. jvp;r  1j. This last quantity can thus be calculated using truncation error on the whole deflection, is analytically derived.
only the first r structural modes included in the truncated deflec-
tion, and it can be used to obtain an a priori estimate of the trunca- 5.1. Truncation error analysis on modal flexibility-based interstory
tion effects expected on the deflection, as shown in the numerical drifts of STFB structures
and experimental analyses presented in Section 6.
Referring to an n-story frame building modeled as an n-DOF
shear-type system and if xp;r is the MF-based displacement profile
4. Structural deflections calculated for a mass proportional load
due to a generic load p and computed using r modes, the vector of
(MPL)
the interstory drifts dp;r can be calculated

The approach for truncation error analysis using the load partic- xp;r;j  xp;r;j1 for j 2 . . . n
dp;r;j 22
ipation factors (proposed in Section 3) is equivalent to the criterion xp;r;j for j 1
based on the mass participation factors, mentioned by Zhang and
Aktan in [17], only if a special load is applied to the structure. This On one hand, starting from the modal flexibility-based displace-
load, indicated as pm , is defined as a mass proportional load (MPL) ment xp;r;j i.e. Eq. (9) and due to Eq. (22) the j-th interstory drift
and, for structures characterized by a diagonal mass matrix, is calculated using r modes is expressed as
expressed as pm M a, where M is the mass matrix and a is an !
X
r
D/j;i X
n X
r
D/j;i
acceleration vector with constant terms a assumed equal to 1. dp;r;j pk /k;i cp;i 23
Starting from the modal flexibility matrix F r , the deflection of the i1
2
xi k1 i1
x2i
structure due to the MPL and using the first r modes is
where each term related to the i-th mode in the summation of Eq.
xm;r F r pm , where each component of the deflection vector is
(23) is indicated as dp;i;j and D/j;i is expressed as
!
X
r
/j;i X
n X
r
/j;i (
xm;r;j a mk /k;i a Ci 19 /j;i  /j1;i for j 2 . . . n
x2i x2i D/j;i 24
i1 k1 i1
/j;i for j 1
By evaluating the i-th mode load participation factor, reported
in Eq. (15), for a mass proportional load (MPL-PF), the mass partic- On the other hand, by applying a generic load p to a shear-type
ipation factor li , according to Eq. (4), is obtained frame building model, the exact j-th interstory drift can be evalu-
ated as
cm;i Ci C2 D/j;i D/j;i V j
vm;i T
Ti li 20 i i
dp;j f j V j Pn
Vj
25
cm C C C x2i x2i g j;UP
kj mk /k;i i
In the same way, the cumulative LPF, according to Eq. (18), cal-
i
culated for the MPL is the cumulative mass participation factor of where Vj is the story shear at the j-th story and f j is the exact flex-
the first r modes (lr . As evident in Eq. (20), when applying the ibility coefficient of the j-th story expressed in terms of the modal
196 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210

parameters of the i-th mode only as indicated by the superscript the modes that are not included in the calculation. As already men-
(i) and in terms of the structural masses, according to Eq. (B3) tioned in Section 4, when applying a mass proportional load, the
P
derived in Appendix B. The term g j;UP nkj mk /k;i is the portion proposed cumulative load participation factor vm;r is equal to the
i
of the participation factor Ci extended only to the DOFs that are cumulative mass participation factor lr . From this observation
above the selected j-th story. and according to Eq. (29), it is evident that for the specific case
In light of these premises, the contributions of the different of a shear-type building structure and considering its first DOF dis-
modes to the MF-based interstory drifts of the STFB structure are placement due to a mass proportional load, the truncation errors
derived, and then the analytical expression for evaluating the are explicitly related to the cumulative mass participation factors.
related truncation errors is determined. The relative contribution
ad;p;i
j
of the i-th mode in the estimation of the j-th interstory drift 6. Validation of the load participation factor for truncation
is the ratio between the truncated solution dp;i;j computed using error analysis
i
the i-th mode and the exact solution dp;j expressed in term of the
In this section the approach proposed in Section 3 to obtain an a
i-th mode only priori estimate of the truncation effects expected on the modal
D/j;i flexibility-based deflections (first objective of the work), is applied
dp;i;j x2
cp;i cp;i g j;UP and verified. In particular, the load participation factors are com-
ad;p;i
j
i
D/ i Vj
i
26
dp;j j;i Vj pared with the truncation errors calculated by means of the crite-
x2i j;UP
gi
rions defined in Eqs. (5)(7). The percent truncation errors on the
As evident in Eq. (26), the terms D/j;i and x2i cancel out. More- modally truncated deflections are thus calculated through a com-
over, the relative contribution of the first r modes in the estimation parison with the true deflections, and they represent the reference
of the j-th interstory drift is solutions. On the contrary, the load participation factors vp,r are
calculated, according to Eq. (18), only for the modes that are
Pr
X
r j;UP included in the computation of the truncated deflections, and then
i1 c p;i g i
ad;p;r
j
ad;p;i
j
27 the terms |vp,r  1| are considered as the mentioned a priori esti-
i1
Vj
mates of the truncation effects.
If all the modes are included in the evaluation of the j-th inter- The analyses were performed on MF-based deflections calcu-
story drift (i.e. r = n), the exact solution is derived, and of course the lated both using numerical simulations on structural models and
relative contribution of these n modes is ad;p;n
j
1. Due to this, if for an experimental vibration test. In the first case, the analyses
were performed on discrete models of plane frame structures,
the first r modes are known, the relative contribution a
 d;p;r
j
to the
which can be reduced to systems of lumped masses and springs.
j-th interstory drift of the missing (n  r) modes can be computed In the second case, a steel frame structure tested under ambient
as a
 d;p;r
j
1  ad;p;r
j
. Finally, the truncation error ed;p;r
j
on the j-th MF- vibrations was considered (i.e. the IASC-ASCE benchmark problem
based interstory drift can be evaluated by changing the sign of a
 d;p;r
j phase II) [26,27].
since, as already mentioned in Section 3, if the missing contribu-
tion to the target value is positive, the truncation error is negative 6.1. Numerical simulation on a shear-type frame building (STFB)
(and vice versa). Thus, the analytical expression related to the trun-
cation error on the j-th interstory drift is Shear-type models of a six-story reinforced concrete (RC) plane
frame building (Fig. 1) were considered in the numerical applica-
Pr j;UP
i1 c p;i g i
tion. The beams of the structure are supposed to be infinitely stiff
ed;p;r
j
ad;p;r
j
1 1 28 in comparison to the columns, and the stiffness of each story is
Vj
equal to kj=2.29
105 kN/m.
A parametric study using Monte Carlo simulations was con-
5.2. Truncation error analysis on the first DOF displacement of STFB ducted on the STFB structure to consider different distributions
structures: a special case of both the structural masses and the applied loads. The reason
for the choice of these parameters is that, as shown in Eq. (15),
Referring to the truncation errors on the first DOF displacement the proposed LPF factors depend both on the masses and on the
of a shear-type frame building (i.e. the first interstory drift for j=1),
two observations can be derived: firstly, the term g 1;UP i is equal to
the participation factor Ci and, secondly, the story shear at the base
of the structure is equal to the sum of all the applied loads, thus
V 1 pTOT c Tp C, according to Eq. (17). Starting from these observa-
tions, it is evident that the relative contribution a1d;p;i of the i-th
mode to the first interstory drift of a STFB is equal to the proposed
load participation factor vp;i , as reported in Eq. (15). Similarly, the
contribution of the first r modes a1d;p;r to the same interstory drift is
the cumulative load participation factor vp;r , according to Eq. (18).
For these reasons and due to Eq. (28), the truncation error on the
first interstory drift evaluated using r modes of a STFB can be
expressed as

e1d;p;r a1d;p;r  1 vp;r  1 29

This truncation error, which is of course equal to the error e1x;p;r

related to the first DOF displacement of the structure, is thus Fig. 1. Six-story plane frame building (dimensions in m) (a) and shear-type frame
explicitly related to the cumulative load participation factor of building (STFB) model (b).
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210 197

(a) (b)

random Numerical model

coefficients M K (n DOF)

Modal r=n
parameters

r<n
Modal
Fr Fn flexibility
matrices
random
coefficients p
j

xp,r xp,n MF-based

deflections

LOAD PARTICIPATION
p,(i) , p,r x,p,r
TRUNCATION
FACTORS (LPF) ERRORS

Fig. 2. Numerical simulation on a shear-type frame building: random variables in the Monte Carlo simulation (a) flow chart of the analyses (b).

loads. As reported in Fig. 2a, the mass of the j-th story of the STFB is
equal to mj = dj mref where mref = 100 kN s2/m, and the generic force
applied at each DOF to compute the deflection is pj = bj pref where
pref = 100 kN. In the Monte Carlo simulation both the coefficients
dj, which are the story mass ratios, and the coefficients bj, which
are the single-DOF load ratios, are considered as random variables.
For each run of the simulation, these parameters are generated
using the pseudo-random integer generator [35], and they are uni-
formly distributed in the range from 1 to 5.
On one side, the resulting distributions of the structural masses
correspond in some cases to realistic situations, in other cases to
more rare structural configurations. Anyway, such distributions
were considered to have a complete insight of the tendencies of
the results. On the other side, the generated random loads are such
that they induce positive shear forces at all the stories of the struc-
ture (i.e. these story shears have the same sign). This is a require-
ment for the inspection loads that can be applied, for example, on
modal flexibility matrices of building structures for damage detec-
tion purposes [2023], and a common choice is a uniform load.
However, random distributions of these positive shear inspection
loads were considered in this truncation error study to verify the
proposed LPF-based approach for more general loading conditions.
The calculations performed in each run of the Monte Carlo sim-
ulation are outlined in the flow chart reported in Fig. 2b. At first,
the mass matrix of the structure is assembled using the random
coefficients dj, and eigenvector analysis is performed on the
numerical model of the structure to compute natural frequencies
and mode shapes. Then, modal flexibility matrices are assembled

(a) (b)

6 r=6 6 r=6
r=5 r=5
r=4 r=4
5 5 r=3
r=3
r=2 r=2
4 r=1 r=1
4
j-th story
j-th story

3
3

2
2

1
1

0
0 0.005 0.01 0.015 0.02 -5 0 5 10 15 20 25
j
Displacement x p,r,j [m] Error | x,p,r | (%)

Fig. 3. MF-based deflections of the STFB structure with a generic distribution of the masses and for a generic load: displacements (a) truncation errors (b).
198 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210

considering all the possible subsets of included modes (i.e. for r = 1
. . . n  1), and the truncated MF-based deflections are calculated by
applying the load p, whose distribution depends on the random
coefficients bj. The displacements of these deflections are com-
pared with the true displacements obtained for r = n, and the trun-
cation errors at each DOF are calculated using Eq. (5). At the same
time, the load participation factors and the cumulative LPFs of the
first r modes are calculated using Eqs. (15) and (18), respectively.
An example of the MF-based deflections calculated by applying
a load vector b pref - where b = [1 3 1 3 1 1]T - to the STFB charac-
terized by a mass distribution d mref - where d = [3 1 3 1 1 1]T - is
shown in Fig. 3a. Each deflection is calculated for a different subset

(a) 40 (b) 40
corr. =1; slope sl=100 corr. = -0.092
35 35
| (%)

| (%)
30 30
1st DOF errors | x,p,r

1st DOF errors | x,p,r

j=1

j=1
25 25

20 20

15 15

10 10

5 5

0 0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
proposed LPF |p,r - 1| MPF |r - 1|

(c) 40 corr. =0.99; slope sl=100

(d) 40 corr. = -0.093
35 35
| (%)
| (%)

30 30
MAX
MAX

MAX errors | x,p,r

MAX errors | x,p,r

25 25

20 20

15 15

10 10

5 5

0 0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
proposed LPF |p,r - 1| MPF |r - 1|

(e) 40
corr. =0.96; slope sl=54 (f) 40
corr. = -0.025
35 35
(%)

(%)

30 30
RMS

RMS
RMS errors x,p,r

RMS errors x,p,r

25 25

20 20

15 15

10 10

5 5

0 0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
proposed LPF |p,r - 1| MPF |r - 1|

Fig. 4. Correlation analysis on the truncation errors of the STFB for r = 1: correlation with the proposed Load Participation Factor (a, c, e); correlation with the Mass
Participation Factor (b, d, f).
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210 199

of included modes, and it is evident that the displacement profile
obtained for r = 1 shows the major discrepancies with respect to
the true deflection obtained for r = 6. As shown in Fig. 3b and espe-
cially for r = 1 and r = 2, the maximum of the percent truncation
errors, in terms of absolute values, occur at the first DOF of the
structure (j = 1). The displacement of the first DOF of a STFB struc-
ture subjected to a generic positive shear inspection load is, indeed,
the lowest one, and thus in general the modal truncation effects
have a significant impact on the displacement of this DOF. As also
shown in Fig. 3b, if higher modes are included (e.g. for r  3), the
truncation errors along the height of the building are almost of
the same order.
The Monte Carlo method was then applied, and 500 simulations
of the procedure outlined in Fig. 2b were performed. The trunca-
tion errors and the LPF factors derived from all the simulations
were compared by means of correlation analysis. The results are
presented in the scatter plots on the left hand side of Fig. 4, where
the truncation errors and the terms |vp,r  1| are reported on the y-
axis and the x-axis, respectively. As discussed in Section 3, vp,r is
the cumulative LPF of the r modes that are included in the calcula-
tion of the truncated deflections. According to the first approach
for truncation error analysis mentioned by Zhang and Aktan in
[17] (Section 2.2), the cumulative mass participation factors lr
were also calculated. The factors |lr  1| are compared with the
truncation errors in the scatter plots on the right hand side of
Fig. 4. In particular, Fig. 4 refers to the first DOF errors (Fig. 4a,
b), the maximum errors (Fig. 4c, d) and the root-mean-square
errors (Fig. 4e, f). The results reported in this figure were obtained
for the case r = 1 (i.e. only the first mode is considered), while
results obtained with higher modes included will be discussed
later in this section.
Very high values of the correlation coefficients q (as shown in
Fig. 4a, c, and e) were found between the LPFs and the truncation
errors. On the contrary, for the considered structural cases with
random distributions of both the masses and the applied loads,
the correlation coefficients reported in Fig. 4b, d, and f indicate
no correlation between the mass participation factors and the
truncation errors. As evident in Fig. 4a, there is an exact relation-
ship between the truncation errors related to the first DOF dis-
placement of the STFB and the LPFs, as also analytically derived
in Section 5.2. In fact, the correlation coefficient is q = 1. In addi-
tion, the linear regression that passes through the origin has a
slope sl = 100 (or equivalent to 1 if relative errors are considered
instead of percent errors). Then, by comparing the maximum
errors with the LPFs, it is evident that, differently from the case
of the first DOF errors, an exact linear relationship is not present
anymore. However, a very high correlation (i.e. q = 0.99) was found
between the maximum errors and the LPFs calculated for the 500
simulations on the considered STFB structure (Fig. 4c), and, in addi-
tion, the slope of the linear regression is also equal to 100. Finally,
by comparing the RMS errors with the LPFs a correlation coefficient
equal to 0.96 was found (Fig. 4e). Thus, the proposed LPF factors
are also correlated with the truncation errors averaged on the
whole deflection. However, the slope of the linear regression is
remarkably lower than the previous case, and this is expected since
the RMS errors are lower than the maximum errors.
All these observations on the correlation between the LPFs and
the truncation errors are also confirmed if higher modes are
included in the calculation of the MF-based deflections. This is
shown by the results of the same analyses performed for the cases
r = 2 and r = 3 (reported in Fig. 5).
6.2. Application to real-data: IASC-ASCE benchmark problem for SHM
(phase II)
The proposed approach for truncation error analysis was also
applied to the experimental data of the IASC-ASCE benchmark
structure [26,27]. This structure is a one-third scale, four-story,
two-bay by two-bay steel frame structure (Fig. 6) that was tested
on August 47, 2002 at the Earthquake Engineering Research Lab-
oratory of the University of British Columbia, Vancouver, Canada.
Various type of excitations (impact hammer, electrodynamic

r=2 (b) r=2 (c) r=2

(a)
40 40 40
=1; sl=100 =0.98; sl=100 =0.96; sl=47
1st DOF errors | | (%)

MAX errors | | (%)

RMS errors (%)

30 30 30

20 20 20

10 10 10

0 0 0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
LPF | - 1| LPF | - 1| LPF | - 1|
p,r p,r p,r

r=3 (e) r=3 (f) r=3

(d)
40 40 40
=1; sl=100 =0.96; sl=100 =0.95; sl=46
1st DOF errors | | (%)

MAX errors | | (%)

RMS errors (%)

30 30 30

20 20 20

10 10 10

0 0 0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
LPF | - 1| LPF | - 1| LPF | - 1|
p,r p,r p,r

Fig. 5. Correlation analysis between the truncation errors of the STFB and the proposed Load Participation Factor: r = 2 (a, b, c); r = 3 (d, e, f).
200 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210
Fig. 6. IASC-ASCE benchmark structure: (a) 3D view with accelerometer locations (dimensions in m); (b) plan view.
shaker, and ambient vibrations) were used in these tests, which are
benchmark studies for SHM and damage detection on building
structures. The structure was thus tested considering different con-
figurations with imposed stiffness reductions (e.g. some diagonal
wall braces of the structure were removed and some beam-
column connections were loosened) [26].
In the truncation error analyses presented in this paper only the
ambient vibration data and only the first of the tested configura-
tions (i.e. the pristine or undamaged structure) were considered.
Fifteen acceleration sensors, schematically represented in Fig. 6a,
were used to collect the ambient vibration data. The measure-
ments provided in [27] have a length of time of 300 s and a sam-
pling frequency of 200 Hz.
Only the data acquired with the sensors located in the center of
the structure and aligned to the x direction (i.e. sensors A14-x,
A11-x, A8-x, A5-x, A2-x) were used in this work to estimate the
modal flexibility of the structure in x direction. In the considered
configuration, the structure was characterized by a plan-
symmetric distribution of the story stiffness at each level. In addi-
tion, as reported in [26], four steel plates (Fig. 6b) were positioned
at each level of the structure to make the mass distribution reason-
ably realistic. Even if the plates aligned to the west-east direction
were shifted towards north direction, the coupling effects on the
structural modes induced by this disposition of the masses are
small, as shown in [36] where the modal identification was applied
using all the sensor data. A 2D analysis was thus performed in the
present work to estimate the modal flexibility matrix of the struc-
ture in x direction, under the simplified assumption of neglecting
the mentioned small coupling effects. Moreover, the structure is
Table 1
Modal parameters of the IASC-ASCE benchmark structure (undamaged configuration) identified from ambient vibration data of acceleration sensors in x direction.
Mode i Natural frequency f i (Hz) Modal damping ratio fi (%)
I 7.49 0.77
II 19.89 0.54
III 25.31 0.26
IV 28.21 0.23
composed by columns and beams that have double T sections
(B100
9 and S75
11, respectively) [26]. As shown in Fig. 6b
and according to the orientation of the column sections, the x axis
is the weak direction of the structure, while the y axis is the strong
direction. If one compares the structural rigidities of the columns
and the beams in both directions, it can be observed that a simpli-
fied modeling, such as a 4-DOF shear-type modeling, is much more
valid for the weak direction of the structure with respect to the
strong direction.
A time-domain identification algorithm (i.e. the Natural Excita-
tion technique [37] combined with the Eigensystem Realization
Algorithm [38]) was used to obtain the modal parameters of the
structure. Natural frequencies, modal damping ratios, and mode
shapes of the first four modes of the structure in x direction are
reported in Table 1, where the mode shapes are normalized to a
maximum value of unity. All the principal modes of the structure
in the considered x direction were thus identified. The system mass
matrix was estimated on the basis of the information available
about the structure [26,27], and it was used to obtain mass-
normalized mode shapes. This is a simple scaling approach that
was also applied in other works [2023], where modal flexibility-
based deflections were estimated from output-only vibration data.
Identified mode shapes and natural frequencies were employed
to estimate the modal flexibility of the structure in x direction. The
MF matrices were assembled using different subsets of modes (i.e.
for r from 1 to 4), and then the MF-based deflections were com-
puted. In particular, the deflection obtained with all the modes
(i.e. r = 4) was considered as the target deflection, and was used
for the calculation of the percent truncation errors. An example
Mode shape components wj;i for each DOF j
w4;i w3;i w2;i w1;i
1.000 0.838 0.630 0.418
0.989 0.052 1.000 0.964
0.453 0.610 0.272 1.000
0.532 0.991 1.000 0.954
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210 201

(a) (b)
4.5 4.5
r=4 r=4
4 4 r=3
r=3
r=2 r=2
3.5 3.5
r=1 r=1
3
3

j-th story
j-th story
2.5
2.5
2
2
1.5
1.5
1
1
0.5

0 0.5
0 1 2 3 4 5 6 -5 0 5 10 15 20 25
Displacement x p,r,j [m] x 10-3 Error | x,p,r j| (%)

Fig. 7. MF-based deflections of the IASC-ASCE benchmark structure for a generic load 2D analysis in x direction: displacements (a) truncation errors (b).

of the structural deflections obtained by applying a load vector b not apply in this experimental case study. The considered masses
pref - where b = [1 3 1 3]T and pref = 10 kN - is reported in Fig. 7a. are, indeed, the real masses of the structure, and the modal param-
It is worth noting that for the considered structure and the applied eters were identified only once from the ambient vibration data
load the maximum errors occur at the first DOF for all the analyzed (Table 1). However, in each run of the Monte Carlo simulation a
cases r = 1, r = 2, and r = 3 (Fig. 7b). positive shear inspection load was assembled using the random
The parametric study using the Monte Carlo method, previously coefficients bj, which were generated in the same way as described
applied to the STFB numerical model, was also applied to the in Section 6.1.
modal flexibility of the benchmark structure. It is worth mention- Similarly to the case of the STFB model, 500 runs of the Monte
ing that the first two steps of the procedure outlined in Fig. 2b (i.e. Carlo simulation were carried out on the experimental model of
assembling models with random distributions of the masses and the benchmark structure. The results of the correlation analysis
computing the modal parameters using eigenvector analysis) do between the truncation errors (both in terms of maximum and

(a) r=1 r=2 (c) r=3

(b)
=0.98; sl=94 =0.98; sl=110 =1; sl=88
MAX errors | | (%)
MAX errors | | (%)

MAX errors | | (%)

30 30 30

20 20 20

10 10 10

0 0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3 0 0.1 0.2 0.3
LPF | - 1| LPF | - 1| LPF |p,r - 1|
p,r p,r

r=1 r=2 r=3

(d) (e) (f)
=0.98; sl=53 =0.98; sl=61 =0.99; sl=64
RMS errors (%)

RMS errors (%)

RMS errors (%)

30 30 30

20 20 20

10 10 10

0 0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3 0 0.1 0.2 0.3
LPF | - 1| LPF | - 1| LPF | - 1|
p,r p,r p,r

Fig. 8. Correlation analysis between the truncation errors of the IASC-ASCE benchmark structure and the proposed Load Participation Factor (r = 1, 2, 3): maximum errors (a,
b, c); RMS errors (d, e, f).
202 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210

RMS errors) and the load participation factors are reported in Fig. 8.
In particular, Fig. 8 shows the results for a number of included
modes equal to r = 1 (Fig. 8a, d), r = 2 (Fig. 8b, e), and r = 3
(Fig. 8c, f). For all the cases, very high values of the correlation coef-
ficients q were obtained between the LPF factors and the trunca-
tion errors (as already found for the STFB model). The linear
regression analysis was also applied on the mentioned outcomes
of the calculations. Referring to the maximum errors, the slopes
of the linear regressions for the cases r = 1,2,3 were found to be
similar to the values obtained for the STFB analytical model (with
maximum differences of approximately 10%). Considering the RMS
errors, the slopes of the linear regressions for r = 1,2,3 are remark-
ably lower than the ones related to the maximum errors, as already
found and discussed for the analytical STFB model. The results
obtained for the experimental case study thus confirm the out-
comes that were found through numerical simulations on the ana-
lytical models of plane shear-type frame buildings. As already
discussed in the previous paragraphs of this section and also
shown by the good agreement between experimental and analyti-
cal results, the modeling assumptions that have been made on the
analyzed experimental case study can be considered as valid (i.e.
assuming a simplified shear-type modeling for the structure in
the considered x direction, and performing a 2D analysis under
the assumption of neglecting the small coupling effects induced
on the structural modes by the mass distribution). The analyzed
experimental case study has thus proven to be suitable for the ver-
ification of the proposed approach for truncation error analysis by
load participation factors, which has been formulated with refer-
ence mainly to plane structures and considering the modal
flexibility-based deflections of such structures in one prevalent
direction. The extension of the approach to 3D structures with a
significant plan asymmetry (due to the distribution of the stiffness
and the mass) and the analysis of case studies characterized by
highly coupled structural modes will be the object of future devel-
opments of the research.
From the results obtained for the analyzed building structures
the following two observations can be formulated. Firstly, the LPFs
give a measure of the maximum truncation errors that can be
expected in the estimation of the MF-based deflections, using only
information about the structural modes considered for the compu-
tation of these truncated deflections. Secondly, the RMS errors are
in general lower than the maximum errors, and results showed
that these RMS errors are also correlated with the proposed LPF
factors. Thus, if two different structural cases are considered (e.g.
two structures with different mass distributions but the same stiff-
ness and/or two different loading conditions), the LPFs can give an
indication of which is the case with the higher RMS truncation
error, expected on the deflection.
7. Truncation errors on structures with mass irregularities:
comparison between MPL and UL loads
As explained in Section 1, the second objective of the work is to
compare the truncation errors on the MF-based deflections calcu-
lated by applying a mass proportional load with those of the uni-
form load. In the present section, the numerical analyses that
were carried out to attain this objective are presented. Discrete
models of plane structures i.e. a shear-type frame building (STFB)
and a simply-supported beam (SSB) were considered in the anal-
yses. These structures are characterized by irregular mass distribu-
tions, while the stiffness coefficients are constant for all the
structural members. As mentioned in Section 4, the MPL is a special
load for which the proposed load participation factor is equal to the
mass participation factor.

Table 2
Structural configurations of the STFB in terms of the story mass ratios (c = 1,2,3,4,5).

j-th DOF 1f 2f 3f 4f 5f 6f 7f 8f 9f 10f 11f 12f

6 c 1 1 1 1 1 c 1 1 1 c 1
5 1 c 1 1 1 1 1 c 1 1 1 c
4 1 1 c 1 1 1 c 1 c 1 c 1
3 1 1 1 c 1 1 1 c 1 c 1 c
2 1 1 1 1 c 1 1 1 c 1 c 1
1 1 1 1 1 1 c 1 1 1 c 1 c

(a) UL (b) MPL

6 r=6 6 r=6
r=5 r=5
r=4 r=4
5 5
r=3 r=3
r=2 r=2
4
j-th story
j-th story

4 r=1 r=1

3 3

2 2

1 1

-5 0 5 10 15 20 25 -5 0 5 10 15 20 25
j j
Error | x,u,r | (%) Error | x,m,r | (%)

Fig. 9. Truncation errors on the displacements of the STFB (case 7f; c = 3): (a) uniform load; (b) mass proportional load.
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210 203

(a) j=6 (b) j=5

6 1.5
Err. | x,p,r | (%)

Err. | x,p,r j| (%)

UL UL
4 MPL 1 MPL
j

2 0.5

0 0
1 2 3 4 5 1 2 3 4 5
Number of modes (r) Number of modes (r)
(c) j=4 (d) j=3
3 10
Err. | x,p,r | (%)

Err. | x,p,r | (%)

UL UL
2 MPL MPL
j

j
5
1

0 0
1 2 3 4 5 1 2 3 4 5
Number of modes (r) Number of modes (r)
(e) j=2 (f) j=1
20 30
Err. | x,p,r j| (%)

Err. | x,p,r | (%)

UL UL
MPL 20 MPL

j
10
10

0 0
1 2 3 4 5 1 2 3 4 5
Number of modes (r) Number of modes (r)

Fig. 10. Truncation errors on the j-th DOF displacement of the STFB (case 7f; c = 3) comparison between the UL and the MPL.

(a) (b)
1200 1200
=5 =5
=4 =4
1000 1000
=3 =3
=2 =2
800 =1 800 =1
m |si|

|i|

600 600
*

400 400

200 200

0 0
1 2 3 4 5 6 1 2 3 4 5 6
i-th mode i-th mode
Fig. 11. Single-mode participation factors of the STFB (case 7f) for different mass irregularities: (a) m* |si|; (b) |Ci|.

7.1. Application to a shear-type frame building (STFB)
The same shear-type frame building described in Section 6.1
was considered for the analyses presented herein (Fig. 1). Various
structural configurations characterized by different distributions
of the story masses were considered. These configurations are
reported from case 1f to 12f in Table 2, where the mass distribu-
tions are expressed in terms of the story mass ratios and the coef-
ficient c can assume the following values: 1, 2, 3, 4, 5. This
coefficient c was used to impose increasing amounts of mass irreg-
ularities on the analyzed structural configurations, and each
increased mass of the structure is mj = c mref. Referring to the
adopted nomenclature, the cases are ordered as indicated by the
first number of the acronym, and the letter f refers to the type
of structure, which is a frame structure.
7.1.1. One structural configuration with mass irregularity
At first, the truncation error study was performed for the con-
figuration 7f with a mass irregularity that is imposed by selecting
c = 3. After performing an analytical modal analysis on the
204 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210

(a) UL (b) MPL

1 1

Load part. factor m,r (MPL-PF)

Load part. factor u,r (UL-PF)

0.95 0.95

0.9 0.9

0.85 0.85
=5 =5
=4 =4
0.8 =3 0.8 =3
=2 =2
=1 =1
0.75 0.75
1 2 3 4 5 6 1 2 3 4 5 6
Number of modes (r) Number of modes (r)
Fig. 12. Cumulative load participation factors of the STFB (case 7f) for different mass irregularities: (a) uniform load; (b) mass proportional load.

(a) UL (b) MPL

15 15
=5 =5
=4 =4
=3 =3
=2 =2
(%)
(%)

10 =1 10 =1
RMS
RMS

Error x,m,r
Error x,u,r

5 5

0 0
1 2 3 4 5 1 2 3 4 5
Number of modes (r) Number of modes (r)
Fig. 13. RMS truncation errors on the displacements of the STFB (case 7f) for different mass irregularities: (a) uniform load; (b) mass proportional load.

structural model, the modal flexibility matrices were assembled
using different subsets of included modes (i.e. r = 1 . . . n). Then,
the deflection profiles of the structure were calculated by applying
both the MPL load and the scaled version of the uniform load, as
defined in Section 4.
The percent error on each truncated MF-based displacement
was evaluated with respect to the target solution (r = n) and the
outcomes are reported in terms of absolute values in Fig. 9. The
truncation errors on the deflections calculated using the MPL
(Fig. 9b) are in general reduced with respect to the UL load errors
(Fig. 9a). For instance, considering the first DOF (j = 1) and for
r = 1 the errors are 21.1% and 9.8% for the UL and the MPL, respec-
tively. For the same DOF and for r = 2 the truncation error is
reduced from 12.1% (UL) to 5.3% (MPL).
The truncation errors are also plotted in Fig. 10 for each DOF
separately, and the trends of these errors are presented as a func-
tion of the number of the included modes (r). Firstly, the results
show that the errors related to the MPL are lower than the ones
related to the UL for the majority of the DOFs and the values of r,
with only two exceptions (i.e. the 5th DOF for r = 1 and the 3rd
DOF for r = 3). Secondly, the higher the number of the included
modes, the lower the single-DOF truncation error for both two
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210 205

loads. In addition, it can be observed that these decreasing trends
of the errors related to the MPL are more regular than the ones
related to the UL, as evident for example in Fig. 10b for j = 5.
7.1.2. Analysis of increasing mass irregularities on one structural
configuration
The truncation error analysis was then performed on the config-
uration 7f of the STFB structure and considering increasing
amounts of mass irregularities (i.e. c = 1, 2, 3, 4, 5). Of course, for
c = 1 the results obtained by applying the UL are equal to the ones
related to the MPL.
The participation factors Ci and si were evaluated for the differ-
ent structures and reported in terms of absolute values in Fig. 11,
where the coefficients si are scaled by m (Section 4). Each of these
factors can be considered as the weighting term of each mode in
Eqs. (19) and (3), which are related to the MF-based displacements
due to the MPL and the UL, respectively. The results show that an
increase in the coefficient c lead to an increase in the parameter
m|si|, especially for the high-order modes (i.e. the 3rd and the
5th modes). On the contrary, the modifications induced by the
mass variations on the participation factors jCi j of all the high-
order modes are lower than the previous case. The cumulative load
participation factors were also evaluated both for the UL and the
MPL, and the values of these factors are reported in Fig. 12 for
the different coefficients c and for the different number of included
modes. An asymptotic convergence to one of the MPL-PFs for
increasing values of r can be noticed in Fig. 12b, while the UL-PFs
can assume values greater than one for r < n, as shown for example
in Fig. 12a for r = 4.
An estimation of the truncation errors on the whole deflections
was carried out using the RMS criterion on all the single-DOF
errors. The results are reported in Fig. 13 for each number of
included modes and comparing the UL with the MPL. The figure
shows that the mass proportional load RMS errors (Fig. 13b) are
lower or equal than the uniform load errors (Fig. 13a) for all the
analyzed cases. For example, the error for r = 1 and c = 3 is equal
to 11.4% for the UL, while is 5.1% for the MPL. In addition, the trun-
cation errors related to one specific load (UL or MPL) can be com-
pared with the respective cumulative load participation factors,
evaluated for the modes included in the computation of the trun-
cated deflections (Fig. 12). Considering the structures with differ-
ent mass irregularities (i.e. c = 1,2,3,4,5) and considering the
results obtained for the lower values of r (e.g. r = 1,2 for the UL
and r = 1 for the MPL), it is evident that if the cumulative LPFs
decrease, the truncation errors increase (and vice versa). These
trends are in agreement with the proposed general approach for
truncation error analysis.
7.1.3. Analysis of different structural configurations with a fixed mass
irregularity
Herein, the results of the truncation error analysis are presented
for all the structural configurations from 1f to 12f of the STFB
(Table 2) and for a fixed mass irregularity (i.e. selecting c = 3).
These configurations are characterized by different positions and
different numbers of the imposed mass variations. The RMS trun-
cation errors related to the whole MF-based deflections were eval-
uated both for the UL and the MPL and for the different subsets of
included modes.
As shown in Table 3, in the 80% of the 72 analyzed cases (i.e. 12
configurations
6 values of r) the MPL provides RMS truncation
errors that are lower or equal than the errors related to the UL.
Moreover, it is worth noting that for the analyzed STFB structures,
the errors related to the MPL monotonically decrease if higher val-
ues of r are considered. On the contrary, this trend is not always
evident for the UL. Referring only to the results for r = 1, it is evi-
dent in the table that the MPL provides truncation errors that are
lower than the UL if the mass variations are applied at the upper
DOFs of the STFB. This occurs in the cases from 1f to 3f if one mass
variation is imposed, in the cases 7f and 8f for two mass variations
and in the case 11f if three mass variations are applied. Conversely,

Table 3
RMS truncation errors on the displacements of the STFB different structural configurations with a fixed mass irregularity (c = 3).

Case Load p RMS truncation errors eRMS

x,p,r (%)

r=1 r=2 r=3 r=4 r=5 r=6

1f UL 13.73 1.43 1.12 0.15 0.12 0.00
MPL 6.90 1.72 0.58 0.20 0.05 0.00
2f UL 10.66 1.76 0.56 0.74 0.01 0.00
MPL 5.84 1.31 0.77 0.32 0.08 0.00
3f UL 7.30 5.45 0.45 0.34 0.36 0.00
MPL 4.82 2.16 0.67 0.14 0.13 0.00
4f UL 4.21 4.68 0.86 0.11 0.11 0.00
MPL 4.48 2.13 0.37 0.28 0.02 0.00
5f UL 2.00 2.17 2.34 0.68 0.08 0.00
MPL 5.96 1.14 0.93 0.15 0.01 0.00
6f UL 3.25 1.13 0.45 0.16 0.05 0.00
MPL 9.41 1.05 0.13 0.03 0.01 0.00
7f UL 11.41 5.82 0.82 0.54 0.37 0.00
MPL 5.11 2.42 0.55 0.15 0.10 0.00
8f UL 7.30 4.56 0.74 0.75 0.09 0.00
MPL 4.66 1.70 0.40 0.24 0.22 0.00
9f UL 3.81 3.14 2.12 1.47 0.36 0.00
MPL 5.13 0.99 0.79 0.21 0.01 0.00
10f UL 2.06 2.55 0.90 0.94 0.12 0.00
MPL 6.99 2.29 0.09 0.07 0.00 0.00
11f UL 8.13 2.60 2.22 1.39 1.19 0.00
MPL 5.97 1.33 0.66 0.27 0.07 0.00
12f UL 5.43 1.99 1.10 0.92 0.96 0.00
MPL 7.18 2.13 0.07 0.06 0.02 0.00
206 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210

if the mass modifications are applied at the lower DOFs and for
r = 1, the UL provides lower truncation errors. The result is due to
the fact that for a STFB the first mode shape components are higher
at the upper DOFs and this distribution is relevant in the determi-
nation of the load participation factor of the first mode. Notwith-
standing that, as evident in Table 3, for higher values of r and
especially for multiple variations of the masses (i.e. cases from 7f
to 12f), the truncation errors are reduced by applying the MPL
instead of the UL. In the few cases where this reduction is not pre-
sent, the errors related to the MPL and the UL are comparable.
7.2. Application to a simply-supported beam (SSB)
The truncation error analyses on modal-flexibility based deflec-
tions were also carried out using structural models of a simply-
supported beam (SSB). This beam-like structure was examined
because it can represent an idealized model of a simple bridge
structure. The structure has a span length of 50 m and a uniform
cross section (EJ = 2.05
107 kN m2), and the mentioned parame-
Fig. 14. Simply-supported beam (SSB) modeled as a 6-DOF discrete lumped-mass
system (dimensions in m).
ters were defined with reference to an example of a bridge struc-
ture, modeled as a simply-supported beam, that was found in the
literature [39]. In addition, the beam analyzed in the present work
is modeled as a 6-DOF discrete lumped-mass system (Fig. 14), and,
similarly to the STFB, various structural configurations character-
ized by different distributions of the masses were considered.
These distributions are reported from case 1b to 12b and in terms
of the single-DOF mass ratios in Table 4, where the coefficient c
assumes the following values: 1, 2, 3, 4, 5. As already defined for
the STFB, each increased mass is mj = c mref where mref = 17.85 -
kN s2/m for the SSB. The same analyses performed for the STFB
(Section 7.1) were repeated for the SSB, and thus only the main
results are presented.
At first, the structural configuration 4b of the SSB with c = 3 was
considered. The modal flexibility-based deflections were calculated
by applying loads with positive signs (both UL and MPL loads)
along the z axis (Fig. 14). As evident in Fig. 15, the percent trunca-
tion errors on the displacements due to the mass proportional load
are lower than the errors related to the uniform load. Moreover,
the errors are higher at the DOFs of the SSB that are close to the
external constraints, both for the MPL and the UL.
Secondly, the configuration 4b of the SSB was analyzed in case
of increasing amounts of the mass irregularities (i.e. for c = 1, 2, 3,
4, 5). The cumulative load participation factors of the beam were
evaluated both for the UL and the MPL (Fig. 16). It is worth noting
that, in both cases, the second, the fourth and the sixth modes
provide no contributions to the cumulative LPF if a uniform mass
distribution is considered (i.e. for c = 1). This is due to the fact that,
for such symmetric configuration (i.e. 4b for c = 1), the modal
participation factors evaluated for the mentioned modes of the

Table 4
Structural configurations of the SSB in terms of the single-DOF mass ratios (c = 1,2,3,4,5).

j-th DOF 1b 2b 3b 4b 5b 6b 7b 8b 9b 10b 11b 12b

1 c 1 1 1 1 1 1 1 c c c c
2 1 c 1 1 1 1 1 c 1 1 1 1
3 1 1 c c 1 1 c 1 1 1 1 c
4 1 1 1 1 c c c 1 1 1 c 1
5 1 1 1 c 1 c 1 c 1 c 1 c
6 1 1 1 1 c 1 1 1 c 1 1 1

(a) UL (b) MPL

3 3
r=6 r=6
2.5 r=5 2.5 r=5
r=4 r=4
r=3 r=3
2 2
r=2
| x,m,r j| (%)

r=2
| x,u,r j| (%)

r=1 r=1
1.5 1.5

1 1
Error

Error

0.5 0.5

0 0

-0.5 -0.5
1 2 3 4 5 6 1 2 3 4 5 6
j-th DOF j-th DOF
Fig. 15. Truncation errors on the displacements of the SSB (case 4b; c = 3): (a) uniform load; (b) mass proportional load.
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210 207

(a) UL (b) MPL

1 1

Load part. factor m,r (MPL-PF)

Load part. factor u,r (UL-PF)

0.95 0.95

0.9 0.9

=5 =5
=4 =4
0.85 0.85
=3 =3
=2 =2
=1 =1
0.8 0.8
1 2 3 4 5 6 1 2 3 4 5 6
Number of modes (r) Number of modes (r)
Fig. 16. Cumulative load participation factors of the SSB (case 4b) for different mass irregularities: (a) uniform load; (b) mass proportional load.

(a) UL (b) MPL

2 2
=5 =5
1.8 1.8
=4 =4
1.6 =3 1.6 =3
1.4 =2 1.4 =2
(%)
(%)

=1 =1
RMS
RMS

1.2 1.2
Error x,m,r
Error x,u,r

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
1 2 3 4 5 6 1 2 3 4 5 6
Number of modes (r) Number of modes (r)
Fig. 17. RMS truncation errors on the displacements of the SSB (case 4b) for different mass irregularities: (a) uniform load; (b) mass proportional load.

SSB structure are equal to zero (i.e. Ci 0. In addition, the work
done by a uniform load for the modal displacements of a symmet-
ric mode shape is equal to zero (i.e. for such modes cp;i 0).
For the configuration 4b of the SSB the RMS truncation errors
related to the whole deflections were also evaluated for the differ-
ent values of c (Fig. 17). It is evident that the MPL RMS errors are
lower than the ones related to the UL, for all the cases character-
ized by mass irregularities and especially if r  2. According to
Zhang and Aktan [17], the truncation errors related to the uniform
load should be examined with reference to the cumulative mass
participation factors lr . Following this approach, the UL errors
(Fig. 17a) can be compared, for increasing values of c, with the
cumulative MPL-PFs, which are shown in Fig. 16b and are equal
to lr (Section 4). A not expected trend can be noticed when consid-
ering the above-mentioned parameters, for example, for r = 2. The
higher the cumulative mass participation factor, the higher the
truncation error related to the uniform load. On the contrary, by
comparing Figs. 17a and 16a it is evident that the trend of the UL
errors is similar to the trend of the UL participation factors (e.g.
for r = 2 and for increasing mass irregularities the lower the UL-
PF, the higher the UL truncation error). Then, by comparing
Figs. 17b and 16b it is evident that the trend of the MPL errors is
similar to the trend of the cumulative mass participation factors
(e.g. for r = 2 and for increasing c coefficients the higher the
MPL-PF, the lower the MPL truncation error). Referring to
beam-like structures, it is worth mentioning that it is not possible
208 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210

Table 5
RMS truncation errors on the displacements of the SSB different structural configurations with a fixed mass irregularity (c = 3).

Case Load p RMS truncation errors eRMS

x,p,r (%)

r=1 r=2 r=3 r=4 r=5 r=6

1b UL 1.63 0.25 0.12 0.01 0.01 0.00
MPL 1.07 0.61 0.02 0.02 0.00 0.00
2b UL 2.86 0.38 0.14 0.09 0.01 0.00
MPL 0.41 0.37 0.03 0.03 0.00 0.00
3b UL 1.75 1.09 0.07 0.02 0.02 0.00
MPL 0.54 0.46 0.03 0.02 0.00 0.00
4b UL 1.23 0.90 0.09 0.09 0.04 0.00
MPL 0.42 0.37 0.04 0.02 0.00 0.00
5b UL 2.44 0.93 0.07 0.04 0.03 0.00
MPL 0.78 0.64 0.03 0.02 0.00 0.00
6b UL 3.25 0.89 0.15 0.09 0.01 0.00
MPL 0.43 0.38 0.03 0.03 0.00 0.00
7b UL 1.43 1.43 0.00 0.00 0.00 0.00
MPL 0.41 0.41 0.02 0.02 0.00 0.00
8b UL 0.23 0.23 0.20 0.20 0.00 0.00
MPL 0.28 0.28 0.03 0.03 0.00 0.00
9b UL 0.21 0.21 0.03 0.03 0.00 0.00
MPL 0.71 0.71 0.02 0.02 0.00 0.00
10b UL 1.37 0.21 0.22 0.07 0.02 0.00
MPL 1.16 0.45 0.04 0.03 0.00 0.00
11b UL 0.80 0.83 0.27 0.02 0.01 0.00
MPL 1.23 0.54 0.02 0.02 0.00 0.00
12b UL 0.53 0.53 0.16 0.07 0.06 0.00
MPL 0.71 0.51 0.04 0.02 0.00 0.00

to analytically relate the truncation errors with the LPFs, as shown
by contrast for a shear-type building model in Section 5. However,
as discussed above, the general trends of both the RMS errors and
the cumulative load participation factors, obtained for the consid-
ered beam structure with mass irregularities and for the applied
loads (UL and MPL), can be well explained on the basis of the
LPF-based approach, proposed in this paper for truncation error
analysis.
Finally, for all the structural configurations from 1b to 12b
(Table 4) and for c = 3, the RMS truncation errors on the displace-
ments were computed. As shown in Table 5, the MPL provides
truncation errors that are lower or equal than the errors related
to the UL in the 85% of the 72 analyzed cases (i.e. 12 configura-
tions
6 values of r). Referring to the truncation errors for r = 1,
it is evident that for example, focusing on the cases 7b, 8b and
9b the effectiveness of applying the MPL instead of the UL is
higher if the mass increments are imposed at the middle of the
beam.
8. Conclusions
This work is an attempt to analyze how truncation effects are
introduced on modal-flexibility (MF) based deflections when
these deflections are estimated using only a subset of structural
modes. This is a common situation that occurs in practical appli-
cations of modal testing on real-life structures. To address this
problem, an approach for truncation error analysis on MF-based
deflections of structural systems subjected to generic loads has
been proposed. This approach is based on the calculation of a load
participation factor (LPF) that represents the relative contribution
of each mode to the whole deflection. This LPF, as deduced ana-
lytically, depends both on the applied load and on the distribu-
tion of the structural masses. By calculating the cumulative LPFs
of the modes included in the truncated deflection, the relative
contribution of the missing modes can be easily determined,
and this last quantity can be used to obtain an a priori indication
of the truncation effects expected on the deflection. In addition,
the relationship between the proposed approach and the
approach based on the mass participation factors, mentioned in
[17] for the case of uniform load (UL) deflections, was discussed
since the two approaches are equal only if a special load, which
is a mass proportional load (MPL), is applied. Thus, the applica-
tion of this MPL load for structures with mass irregularities was
also investigated, and the truncation errors on these deflections
were compared with the errors on the deflections due to the uni-
form load.
To verify the proposed LPF-based approach for truncation
error analysis both numerical and experimental case studies
were considered. Referring to the former, the analyses were con-
ducted on numerical models of shear-type frame buildings
(STFB), and Monte Carlo simulations were employed to analyze
a large number of structural cases characterized by generic dis-
tributions of both the masses and the applied loads. Results
showed that the proposed LPF factors are highly correlated with
the truncation errors on the MF-based deflections. On the con-
trary, for the considered structural cases no correlations were
found between the truncation errors and the mass participation
factors. In addition, the results showed that, for the analyzed
structures, the LPF factors can give a measure of the maximum
of the single-DOF truncation errors expected on the deflections.
The above-mentioned results were also confirmed for a steel
frame structure tested under ambient vibrations (i.e. the IASC-
ASCE benchmark problem) [26,27]. The identified modes and
an estimation of the structural masses were used to extract
the modal flexibility of the structure in one direction. Based on
this experimental model and using the Monte Carlo method to
consider a large number of MF-based deflections due to generic
loads, the correlation analysis was performed between the pro-
posed LPF factors and the truncation errors. This experimental
 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210 209

2 3
validation was made possible because all the principal modes in ..
the considered direction of the structure were identified, and 6fn  fj  f1  fj  f1 . f1 7
6 7
6 .. .. .. 7
thus the errors were computed by directly comparing truncated 6  . . . f1 7
6 7
and non-truncated deflections. On the contrary, according to the 6 .. 7
6 7
proposed approach the LPF factors were computed only for the 6 fj  f1  fj  f1 .f1 7
6 7
subset of modes used for calculating the truncated deflections. 6 .. 7
6 7
This is the main advantage of the approach, and it is expected 4    . f1 5
that the LPF factors can be used as a tool for truncation error f1 f1 f1 f1 f1
analysis on more complex real-life structures, where current lim- 2 30 1 0 1
mn 0       0 /n;i /n;i
its of testing technologies lead necessarily to the identification of 6 .. .. .. 7B . C B . C
only a subset of structural modes. 6 0 . 0 . . 7B .. C B .. C
6 7B C B C
6 . .. 7B C B C
To compare the truncation errors on the deflections due to the 6

6 . . 0 mj 0 . 7B /j;i C 1 B /j;i C B2
7B C x2 B C
mass proportional load (MPL) with those of the uniform load (UL), 6 7B . C i B C
6 .. .. .. 7B . C B .. C
various configurations of STFB models with increasing amounts of 4 . . 0 . 0 5@ . A @ . A
mass irregularities were considered. In addition, numerical analy- /1;i /1;i
0   0 m1
ses were also performed on a simply-supported beam. For both
structures, the results showed that in the large majority of the The adopted formulation is the dual of the one reported in [40]
analyzed configurations, the truncation errors due to the MPL and due to the special structure of the flexibility matrix for shear-
are lower compared to those related to the UL. In the few cases type frame buildings, the flexibility coefficients can be expressed in
where the result was not found, the errors related to the two terms of the modal data and the structural masses. Indeed, as evi-
loads were comparable. These results suggest that in real-life dent in Eq. (B2), the last row of F contains only the coefficient f 1 ,
applications of structural identification and damage detection which can be derived directly from the last equation of the system.
for structures with mass irregularities, a mass proportional load Then, all the other flexibility coefficients for j=2 . . . n can be deter-
can be adopted as an alternative to the uniform load, with the mined. The j-th flexibility coefficient can thus be expressed as
aim of reducing the truncation effects on the modal flexibility-
based deflections. i D/j;i 1
fj Pn B3
x2i kj mk /k;i
Acknowledgements
where D/j;i is defined in Eq. (24). The formulation is valid using the
The work done by the IASC-ASCE Task Group on the benchmark modal parameters of any mode i and for any normalization of the
studies for Structural Health Monitoring [26,27] is gratefully mode shape. The notation with the superscript (i) in Eq. (B3) indi-
acknowledged. i
cates that the exact flexibility coefficient f j is expressed in term
of the i-th mode only.
Appendix A. Evaluation of the total load: proof of Eq. (17)
References
The equation to be proven is pTOT c Tp C, where pTOT is the sum of
the components of the assumed load p. The vector c p , which con- [1] Doebling SW, Farrar CR, Prime MB, Shevitz DW. Damage identification and
health monitoring of structural and mechanical systems from changes in their
tains the scalar products between the load p and each mode shape,
vibration characteristics: a literature review Report LA-13070-MS. Los Alamos,
and the vector C, which contains the modal participation factors, NM, USA: Los Alamos National Laboratory; 1996.
can be expressed as [2] Sohn H, Farrar CR, Hemez FM, Shunk DD, Stinemates DW, Nadler BR, et al. A
review of structural health monitoring literature form 19962001 Report LA-
c Tp pT Un A1 13976-MS. Los Alamos, NM, USA: Los Alamos National Laboratory; 2003.
[3] Fan W, Qiao P. Vibration-based damage identification methods: a review and
comparative study. Struct Health Monit 2011;10(1):83111.
[4] Farrar CR, Worden K. Structural health monitoring: a machine learning
C UTn MI A2
perspective. 1st ed. Chichester, UK: John Wiley & Sons; 2013.
[5] Sabatino S, Ervin EK. Comparison of damage diagnosis algorithms on a spatial
where In
1 is a vector of all ones. The scalar product between cp and frame using vibration data. Adv Struct Eng 2015;18(5):73958.
C can be expressed as [6] Ewins DJ. Modal testing: theory, practice and application. 2nd ed. Philadelphia,
PA, USA: Research Studies Press Ltd; 2000.
c Tp C pT Un UTn MI pT I pTOT A3 [7] Alvin KF, Robertson AN, Reich GW, Park KC. Structural system identification:
from reality to model. Comput Struct 2003;81:114976.
[8] Brincker R, Ventura CE. Introduction to operational modal analysis. 1st
where the expression Un UTn M 1 is valid due to the mass normal- ed. Chichester, UK: John Wiley & Sons; 2015.
ization of the mode shapes. [9] Zhao J, DeWolf JT. Sensitivity study for vibrational parameters used in damage
detection. J Struct Eng 1999;125(4):4106.
[10] Pandey AK, Biswas M. Damage detection in structures using changes in
Appendix B. Flexibility coefficients of shear-type frame flexibility. J Sound Vib 1994;169(1):317.
buildings derived from the dynamic characteristic equations of [11] Pandey AK, Biswas M. Experimental verification of flexibility difference
method for locating damage in structures. J Sound Vib 1995;184(2):31128.
the undamped system [12] Bernal D. Load vectors for damage localization. J Eng Mech 2002;128(1):
714.
The dynamic characteristic equations of an undamped MDOF [13] Yang QW, Liu JK. Damage identification by the eigenparameter decomposition
of structural flexibility change. Int J Numer Meth Eng 2009;78:44459.
structural system, using the flexibility matrix F and the mass
[14] Bernal D. A subspace approach for the localization of damage in stochastic
matrix M, are expressed as systems. In: Chang FK, editor. Structural health monitoring: the demand and
challenges proceedings of the 3rd international workshop in structural
1 health monitoring. Boca Raton, FL, USA: CRC Press; 2001. p. 899908.
FM/i /i B1
x2i [15] Duan Z, Yan G, Ou J, Spencer BF. Damage localization in ambient vibration
by constructing proportional flexibility matrix. J Sound Vib 2005;284:
Considering the flexibility matrix of a plane shear-type frame 45566.
[16] Duan Z, Yan G, Ou J, Spencer BF. Damage detection in ambient vibration using
building model, in which the j-th interstory flexibility is f j , and a proportional flexibility matrix with incomplete measured DOFs. Struct Control
diagonal mass matrix, Eq. (B1) can be reformulated as Health Monit 2007;14(2):18696.
210 G. Bernagozzi et al. / Engineering Structures 142 (2017) 192210
[17] Zhang Z, Aktan AE. Application of modal flexibility and its derivatives in
structural identification. Res Nondestr Eval 1998;10(1):4361.
[18] Catbas FN, Brown DL, Aktan AE. Use of modal flexibility for damage detection
and condition assessment: case studies and demonstrations on large
structures. J Struct Eng 2006;132(11):1699712.
[19] Catbas FN, Gul M, Burkett JL. Damage assessment using flexibility and
flexibility-based curvature for structural health monitoring. Smart Mater
Struct 2008;17(1):015024.
[20] Koo KY, Sung SH, Park JW, Jung HJ. Damage detection of shear buildings using
deflections obtained by modal flexibility. Smart Mater Struct 2010;19
(11):115026.
[21] Koo KY, Sung SH, Jung HJ. Damage quantification of shear buildings using
deflections obtained by modal flexibility. Smart Mater Struct 2011;20
(4):045010.
[22] Sung SH, Koo KY, Jung HY, Jung HJ. Damage-induced deflection approach for
damage localization and quantification of shear buildings: validation on a full-
scale shear building. Smart Mater Struct 2012;21(11):115013.
[23] Amiri GG, Hosseinzadeh AZ, Bagheri A, Koo KY. Damage prognosis by means of
modal residual force and static deflections obtained by modal flexibility based
on the diagonalization method. Smart Mater Struct 2013;22(7):075032.
[24] Koo KY, Lee JJ, Yun CB, Kim JT. Damage detection in beam-like structures using
deflections obtained by modal flexibility matrices. Smart Struct Syst 2008;4
(5):60528.
[25] Berman A, Flannelly WG. Theory of incomplete models of dynamic structures.
AIAA J 1971;9(8):14817.
[26] Dyke SJ, Bernal D, Beck J, Ventura C. Experimental phase II of the structural
health monitoring benchmark problem. In: Proceedings of the 16th ASCE
engineering mechanics conference, Seattle, Washington, USA, July 1618,
2003.
[27] Dyke SJ. Report on the building structural health monitoring problem phase 2
experimental, 2011. Network for Earthquake Engineering Simulation
(database), <https://nees.org/resources/2412>.
[28] Doebling SW, Peterson LD, Alvin KF. Estimation of reciprocal residual
flexibility from experimental modal data. AIAA J 1996;34(8):167885.
[29] De Angelis M, Lus H, Betti R, Longman RW. Extracting physical parameters of
mechanical models from identified state-space representations. J Appl Mech
2002;69(5):61725.
[30] Lpez-Aenlle M, Fernndez P, Brincker R, Fernndez-Canteli A. Scaling-factor
estimation using an optimized mass-change strategy. Mech Syst Signal Process
2010;24(5):126073.
[31] Aenlle ML, Brincker R. Modal scaling in operational modal analysis using a
finite element model. Int J Mech Sci 2013;76:86101.
[32] Brincker R, Lpez-Aenlle M. Mode shape sensitivity of two closely spaced
eigenvalues. J Sound Vib 2015;334:37787.
[33] Brincker R. Implications of closely spaced modes in OMA. In: Proceedings of
the 6th international operational modal analysis conference (IOMAC), Gijn,
Spain, May 1214, 2015.
[34] Hogue TD, Aktan AE, Hoyos A. Localized identification of constructed facilities.
J Struct Eng 1991;117(1):12848.
[35] MATLAB. Natick, Massachusetts, United States: The MathWorks Inc.
[36] Ching J, Beck JL. Bayesian analysis of the phase II IASCASCE structural health
monitoring experimental benchmark data. J Eng Mech 2004;130(10):123344.
[37] James III GH, Carne TG, Lauffer JP. The Natural Excitation Technique (NExT) for
modal parameter extraction from operating wind turbines. Int J Anal Exp
Modal Anal 1995;10(4):26077.
[38] Juang JN, Pappa RS. An eigensystem realization algorithm for modal parameter
identification and model reduction. J Guid Control Dyn 1985;8(5):6207.
[39] HIVOSS Project. Human induced vibrations of steel structures: design of
footbridges - background document. RFS2-CT-2007-00033, Research Fund for
Coal and Steel; 2007, <http://www.stb.rwth-aachen.de/projekte/2007/
HIVOSS/download.php>.
[40] Wang JF, Lin CC, Yen SM. A story damage index of seismically-excited
buildings based on modal frequency and mode shape. Eng Struct 2007;29
(9):214357.