Sie sind auf Seite 1von 11

Science of the Total Environment 568 (2016) 13331343

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage:

Vibrations inside buildings due to subway railway trafc. Experimental

validation of a comprehensive prediction model
Patrcia Lopes a,b, Jsus Fernndez Ruiz c, Pedro Alves Costa a,, L. Medina Rodrguez c, Antnio Silva Cardoso a
University of Porto, Faculty of Engineering, Porto, Portugal
Polytechnic Institute of Porto, Porto, Portugal
University of La Corua, Campus de Elvia, La Corua, Spain


A numerical approach for prediction of vibrations due to subway trafc is proposed;

The experimental validation of the numerical approach is performed;
The proposed numerical approach is useful for the study of mitigation solutions.

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

Article history: The present paper focuses on the experimental validation of a numerical approach previously proposed by the
Received 8 October 2015 authors for the prediction of vibrations inside buildings due to railway trafc in tunnels. The numerical model
Received in revised form 1 November 2015 is based on the concept of dynamic substructuring and is composed by three autonomous models to simulate
Accepted 3 November 2015
the following main parts of the problem: i) generation of vibrations (traintrack interaction); ii) propagation
Available online 15 November 2015
of vibrations (tracktunnelground system); iii) reception of vibrations (building coupled to the ground). The
experimental validation consists in the comparison between the results predicted by the proposed numerical
Vibrations due to railway trafc in tunnels model and the measurements performed inside a building due to the railway trafc in a shallow tunnel located
Numerical modeling in Madrid. Apart from the brief description of the numerical model and of the case study, the main options
Experimental validation and simplications adopted on the numerical modeling strategy are discussed. The balance adopted between
Vibrations inside buildings accuracy and simplicity of the numerical approach proved to be a path to follow in order to transfer knowledge
to engineering practice. Finally, the comparison between numerical and experimental results allowed nding a
good agreement between both, fact that ensures the ability of the proposed modeling strategy to deal with
real engineering practical problems.
2015 Elsevier B.V. All rights reserved.

1. Introduction Subway railway networks correspond to the most efcient mass trans-
portation system in highly populated areas. Since the regulations on air
The growing of world population and its concentration on urban pollution will prevent the generalized development of road transporta-
centers and metropolitan areas is arising new challenges and demands tion in urban regions, the development of railway subways will be an
to society and, consequently, to engineering. According to UN, the unavoidable reality in the next decades, comprising the construction
population living in metropolitan areas will reach 66% of the world of new networks or the amplication of the existing ones.
population around 2050 (United Nations, D.o.E.a.S.A., 2014). The abrupt Despite the advantages provided by subway railway systems, there
population growth in urban areas, from 29% in 1950 to 54% in 2014, will are also some drawbacks that deserve especial attention from both
bring new demands for several engineering topics, namely for the technical and scientic communities. One of them is related to vibration
development of efcient and eco-friendly transportation systems. and re-radiated noise inside buildings due to trafc. Actually, since few
times subway networks cross the ground at shallow depths, the vibra-
tions induced by the train passage are easily transmitted to nearby
Corresponding author.
buildings. Although vibrations induced by railway trafc usually do
E-mail addresses: (P. Lopes), (J.F. Ruiz), (P. Alves Costa), (L. Medina Rodrguez), not imply damage on modern buildings, the disturbance of inhabitants, (A.S. Cardoso). with consequences on living quality and health, is a usual consequence.
0048-9697/ 2015 Elsevier B.V. All rights reserved.
1334 P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343

Fig. 1. Schematic representation of the numerical modeling strategy followed.

However, this sentence cannot be true when handling with ancient or elastic waves on the ground reaches the foundations of the structures,
heritage buildings, where damage can be induced by trafc vibrations impinging the building and giving rise to perceptible vibrations and
(Ma et al., 2011; Franois, 2008; Vogiatzis and Kouroussis, 2015). re-radiated noise inside dwellings. As can be seen, the problem is
Moreover, the concern about vibrations induced by subway railway quite complex, engaging different subdomains with specic dynamic
trafc increases when dealing with specic facilities as, for example, behaviors, which demands the development of suitable prediction tools.
nanotechnology labs or factories. In those cases, the occurrence of vibra- As to prediction models, different proposals have been presented
tions can prevent the regular work of equipment, implying considerable during the latter years, ranging from scoping and empirical rules
economic losses. (Hansen, 2005; Verbraken et al., 2011; Connolly et al., 2014) to
Having in mind the concerns expressed above, technical and scien- advanced numerical approaches (Gupta et al., 2007; Yang and Hung,
tic communities have allocated considerable effort on attempts to 2008; Lopes et al., 2014a). Although empirical models have the advan-
achieve a better understanding of the problem as well as to develop pre- tage of being derived from experimental results, their application to
diction tools that can be used to mitigate this kind of situation. From the complex scenarios, where design of mitigation measures is required,
previous research, it is possible to establish an understanding of this is difcult and sometimes even impossible. On the other hand, the
complex problem (Lai et al., 2005; Gupta et al., 2009; Lombaert et al., numerical approaches can partially overturn some of those drawbacks,
2014): i) the dynamic interaction between vehicle and track is the but are sometimes followed by a huge computational effort that is
source of vibration; ii) vibrations are transmitted from the track to the incompatible with practical applications. Alternatively, when the
tunnel and, posteriorly, to the ground; iii) the energy propagated as problem's geometry is simple, analytical models, like the PiP developed

Fig. 2. Location of the section under analysis: a) picture; b) schematic representation.

P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343 1335

Fig. 3. Schematic representation of the cross-section of the tunnel.

at U. Cambridge (Forrest and Hunt, 2006; Hussein et al., 2014; Hussein suspensions (Zhai and Cai, 1997; Sheng et al., 2004; Alves Costa,
and Hunt, 2009), can be applied. Attending to the complexity engaged 2012). A bit more complex is the simulation of the nearby buildings.
with the 3D propagation of waves on the tunnel-ground system, several Actually, buildings are usually 3D and cannot be simulated by a 2.5D
proposals have been presented in order to establish a compromise approach, demanding an autonomous simulation and coupling with
between computational effort and versatility of the model. Periodic the tunnelground system. However, the inherent complexity of the
models based on FEMBEM coupling have been proposed by Clouteau interaction between domains simulated with distinct techniques can
et al. (2005) aiming to reduce the computational effort. Alternatively, be reduced assuming that the presence of the building does not affect
when it is possible to assume the tunnelground system as an elastic the source of vibrations. This aspect was recently studied by Coulier
invariant domain, 2.5D approach can be applied. This approach, which et al. (2014), showing that it is a reasonable simplication. However,
can be applied to BEM, FEM, MFS or to the coupling between any of the soilstructure interaction must be attended to obtain reliable results
the numerical methods, involves strong reduction of the computational inside buildings (Lopes et al., 2014b). The 3D FEM is the most suitable
cost without losing the 3D character of the wave propagation solution method for dealing with complex structures of buildings and the SSI
(Lopes et al., 2013; Galvn et al., 2010; Amado Mendes et al., 2015; problem can be rigorously solved by a 3D FEMBEM approach, where
Rieckh et al., 2012). As results from a balance between complexity and the BEM is adopted for the modeling of the ground. Nevertheless, in
computational effort, Lopes et al. proposed a 2.5D approach based an attempt to minimize the computational effort required as well as
on the coupling between FEM and PML (perfectly matched layers) for the complexity of the problem, Lopes et al. (2014b) showed that
dealing with vibrations transmitted through the tunnelground system. lumped parameter models can be used to represent the dynamic
As highlighted above, a comprehensive modeling approach should ground behavior without a considerable loss of accuracy.
also attend to the mechanism of vibration generation and to reception From the description presented above, it is clear that the main
of vibrations inside buildings. Due to the complexity of the problem, theoretical bases for the formulation of the problem are established.
the sub-structuring approach is a rational methodology for dealing Anyway, the transfer of the knowledge to engineering practice demands
with this kind of problem. Regarding the simulation of the source, it is the experimental validation of the numerical tools in order to show
usual to assume that vibrations are generated due to unevenness of their reliability and accuracy. The present paper aims to contribute to
the track, being the vehicle simulated through a simple multi-body this particular aspect, presenting an experimental validation of a
approach where rigid masses, which represent the main masses of the numerical approach previously presented by Lopes et al. (Lopes et al.,
vehicle, are connected by springdashpots to attend to the train's 2014a; Lopes et al., 2014b). The selected case study corresponds to a

Fig. 4. Detail of railway track.

1336 P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343

Fig. 5. Monitored building: a) picture; b) structural plan.

shallow railway tunnel in Madrid which crosses the ground beneath a the case study. Then, the particular aspects of the numerical model are
building. Experimental measurements of vibrations induced by railway described and the results provided by the model are compared with
trafc were performed by CEDEX (CEDEX, 2003), due to complaints of the experimental measurements. Finally, the main conclusions are
the building's inhabitants, and were reported in Fernandez (2014). highlighted.
The measurement campaign performed by CEDEX comprised the
measurement of accelerations on the railway track (on the rail and 2. Brief description of the numerical approach
slab) and also vibration velocity at the building oors. Both measure-
ments are compared with the results obtained from the numerical The numerical approach used on the present studies was previously
model. Along the study presented below, the options and simplica- presented by Lopes et al. (Lopes et al., 2014a; Lopes et al., 2014b). By
tions assumed on the numerical modeling are discussed. that reason, only a brief description of the model is here presented,
The reasonable agreement obtained between experimental and the reader is advised to consult those previous works in order to obtain
numerical results shows the reliability of the model proposed by a deep understanding of the followed approach.
Lopes et al. (2014b). This model has the advantage of being relatively In an attempt to reach high numerical performance, the sub-
simple and can be faced as a valuable tool for the prediction and design structuring approach was followed, being the global model constituted
of vibration mitigation countermeasures. by 3 sub-models, as schematically depicted in Fig. 1.
Regarding the paper organization, a brief description of the numeri- Each sub-model is dedicated to one of the main parts of the whole
cal approach is presented rstly, being followed by the presentation of problem: i) generation; ii) propagation; iii) reception. This classication

Fig. 6. Schematic representation of the tunnel-building system.

(Adapted from (CEDEX, 2003)).
P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343 1337

Table 1 of accuracy on the solution of the problem (Colao et al., 2015). This
Properties of the structural elements of the building. approach is followed in the present study.
Element Properties Dimensions In the present model, the solution of the traintrack dynamic inter-
E(GPa),, (kg/m3) action is obtained by a compliance formulation developed on the
Slabs 30, 0.2, 2500 Thickness: 0.25 m frequency domain. The source of dynamic excitation is given by the
Beams 30, 0.2, 2500 0.30 0.60 m2 rail unevenness, which is transformed to the frequency domain taking
Columns 30, 0.2, 2500 0.35 0.35 m2 into account the traveling speed of the train (Alves Costa et al., 2012).
The solution of this problem allows obtaining the dynamic loads
imposed by the train to the track. These loads are then combined with
can also be established taking into account the sub-domain that is the transfer functions, derived from the propagation modulus described
simulated: above, to produce the vibration elds generated on the tunnelground
system due to the train passage.
i). Modeling of tracktunnelground system (propagation)
The solution of the 3D wave propagation through the track iii). Modeling of buildings and soilstructure interaction
tunnelground system is obtained by a 2.5D FEMPML approach,
where the equilibrium equations are formulated on the
wavenumber-frequency domain. This approach uses the Fourier The most suitable numerical approach for simulation of 3D dynamic
transform regarding the tunnel development direction and, by behavior of buildings corresponds to the nite element method (FEM).
that reason, only the cross-section needs to be discretized into However, the soilstructure interaction (SSI) is a relevant aspect that
nite elements. Since the FEM is not suitable to deal with must be correctly attended to achieve accurate predictions of vibrations
unbounded domains, the discretized region is boxed by perfectly inside buildings due to the railway trafc in the tunnel (Lopes et al.,
matched layers, also formulated on the 2.5D domain, that avoid 2014b; Hussein et al., 2015; Kouroussis et al., 2014b). A precise
the spurious reection of waves that reach articial boundaries approach for dealing with the SSI is by the 3D FEMBEM coupling,
(which results from the limitation of the interest domain). A where the capabilities of the BEM to simulate the dynamic behavior of
deep explanation of the method, including its mathematical the ground are visible. On the other hand, alternative and simpler
formulation as well as theoretical validation examples, can be methods can also be explored such as the lumped parameter model
found in Lopes et al. (Lopes et al., 2013; Lopes et al., 2012). (Wolf, 1994). Lopes et al. (2014b) showed that the solution obtained
This numerical model is used to obtain transfer functions considering the SSI from the lumped parameter approach can be very
between the rail and other points of the system as well as to similar to the one obtained using a detailed 3D BEM approach, but the
assess the impedance of the track which is used on the genera- former is much simpler in terms of implementation and can be easily
tion modulus for the solution of the traintrack dynamic interac- introduced in a commercial nite element code. This aspect is relevant,
tion problem. since the method can then be easily transferred from academia to engi-
ii). Modeling of traintrack interaction (generation) neering practice. Based on the conclusions previously found by authors
(Lopes et al., 2014b), the impedance of footings is considered here
using a lumped parameter model (monkey-tail model) (Ibsen and
The dynamic excitation mechanisms due to train passage can be Liingaard, 2006a; Ibsen and Liingaard, 2006b).
divided into two distinct mechanisms: i) the quasi-static mechanism; Since it is assumed that the presence of the building does not affect
ii) the dynamic mechanism. Regarding the rst, the solution is trivial the vibration generation source (Coulier et al., 2014), the vibration elds
since it corresponds to the movement of distribution of the dead weight at the free-eld, obtained from the application of the modulus
of the train by its axles. On the other hand, the dynamic mechanism mentioned above, are used as excitation source to the structure,
arises from the generation of inertial forces on the train due to the which, on other hand, is coupled to the ground. Studies performed by
traintrack interaction. These inertial effects can have different sources Lopes et al. (2014b) show that it is generally acceptable to neglect the
as, for instance, the track unevenness. The assessment of the dynamic presence of the tunnel on the assessment of the building footings
traintrack interaction loads demands the solution of an interaction impedance. This simplication allows assuming the ground as a half-
problem between both domains, where the dynamics of the train space without any perturbation given by the cavity of the tunnel.
must be taken into account. Focusing on the present problem, the sim-
ulation of the train can be performed using simple formulations as, for 3. Case study description
example, a multi-body approach where the main masses of the train
are simulated as rigid bodies interconnected by springdashpot 3.1. General description
elements to represent the suspensions (Zhai and Cai, 1997; Alves
Costa, 2012; Colao et al., 2015; Kouroussis et al., 2014a). Anyway, The selected case study corresponds to an old shallow tunnel that
since suspensions of passenger trains isolate the sprung (car body) belongs to a stretch of the railway network of Madrid. The exact location
and semi-sprung masses (bogies) for relatively low frequencies, the of the stretch under analysis is depicted in Fig. 2.
modeling of the train can be considerably simplied considering only The shallow tunnel crosses the ground just beneath an existing
the unsprung masses of the train (wheelsets) without signicant loss building. Fig. 3 shows the geometry of the cross section of the problem.

Fig. 7. Geometry and wheelset load distribution of the train 446 simple conguration (RENFE).
1338 P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343

Fig. 8. 2.5D FEMPML mesh.

The soil properties are also indicated in the gure. These properties The vibration measurement was performed by CEDEX during the
were derived from the studies performed by Melis et al. (2011) on the passage of a passenger train type 446 of RENFE (double composition).
geotechnical characterization of the ground for the construction of The geometry and load distribution per wheelset of this train are
recent tunnels in the proximity of this case study. As can be seen, the depicted in Fig. 7. The unsprung mass of the vehicle is around 1500 kg
tunnel is quite shallow being the distance between the roof of the per wheelset.
tunnel and the building of about only 5.5 m. It should be referred that In what the experimental data collected during the tests is concerned,
the building was constructed after the excavation of the tunnel. the vertical accelerations of rail and track slab were measured as well as
The liner of the tunnel is on stone masonry, common in old tunnels. the vertical velocities of 5th and 7th oors at a position located above the
A Young modulus of 5 GPa and a Poisson ration of 0.2 for the elastic tunnel (see Fig. 6).
dynamic properties of the material were estimated. From the description made until now, the lack of information
The railway is a STEDEF type, as illustrated in Fig. 4. The properties of regarding the track unevenness is evident. Unfortunately, it was not
the elements that constitute the track are also detailed on the gure. As possible to perform measurements of the track unevenness. Neverthe-
can be seen, the concrete slab rests directly on the tunnel invert, with less, an inversion procedure was adopted where a synthetic railway
any inclusion of a resilient element. The sleepers are spaced 0.6 m on prole was generated in order to nd a good agreement between the
the longitudinal direction. measured and predicted dynamic response of the rail. It was assumed
The building above the tunnel was constructed in the mid 50s of the that the rail unevenness amplitudes can be described by a PSD function
last century. Fig. 5a shows the faade of the building. The structure is with the following equation (Braun and Hellenbroich, 1991):
made of concrete and the plant of the regular oors is depicted in Fig.  
5b. There is 1 buried oor and 8 elevated oors as illustrated on the  kx w
Skx S kx;0 1
scheme presented by Fig. 6. kx;0
The properties of the building's structural elements are indicated in
Table 1. where kx,0 = 1 rad/m, S(kx,0) is a constant which comprises the geomet-
In addition to the dead weight of the structural elements, a load of rical quality of the track unevenness and w is a constant that usually
450 kg/m2 distributed by the slabs surface was considered in order to assumes a value between 3.0 and 4.0.
take into account nonstructural masses of the building.
Regarding foundations, the building is founded on shallow footings 3.2. Monitoring campaign
with an area of 2.75 2.75 m2.
The monitoring campaign consisted in the measurement of vibration
levels induced by the passage of double compositions of the train is
represented in Fig. 7 at regular conditions of exploration of the railway
system. During the passage of the train, the vertical acceleration of the
rail and of the slab was measured, as well as the vertical velocity of
the building slabs on the 5th and 7th oors (see Fig. 6). The measure-
ments were performed by CEDEX in 2002 and the authors of the present
paper didn't participate in the experimental activities (CEDEX, 2003).
Detailed description of the measurements performed on this case
study can be found in Fernandez (2014).
For the studies presented below, the passage of the double composi-
tion of the train depicted in Fig. 7 at the speed of 14.25 m/s (51.3 km/h)
was selected.

4. Model description

As mentioned previously, the sub-structuring approach is followed,

adopting different models and modeling techniques for the distinct
Fig. 9. Synthetic unevenness prole adopted in the present study. sub-domains. Concerning the tracktunnelground system, the dynamic
P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343 1339

Fig. 10. 3D nite elements mesh adopted for the simulation of the building.

response is assessed by a 2.5 D FEMPML approach. The variation of the considered uniformly distributed assuming the following properties:
ground elastic properties reproduced in Fig. 3 was attended in the 333 kN/mm/m and 67 kN/mm/m for the railpad and undersleeper pad
numerical model, where the maximum dimension of the nite respectively.
elements was dened taking into account the accuracy of the results For the simulation of the train and traintrack interaction, the two
for frequencies up to 80 Hz. The nite element region is bounded by more relevant excitation mechanisms were taken into account: i) quasi-
2.5D PML layers where a mesh adaptive procedure was implemented static mechanism; ii) dynamic mechanism. The former comprises the
in order to accurately simulate a half-space (Fig. 8). Detailed informa- movement of the static loads detailed in Fig. 7. For the assessment of
tion about that procedure can be found in Lopes et al. (2013). the latter, it is mandatory to solve the traintrack interaction model
Regarding the track, an EulerBernoulli beam was adopted to simu- where the structural simulation of the train is demanded. Alves Costa
late the rail and springdashpot elements were selected to simulate the et al. 2012), among others, shown that, for the purpose of prediction of
railpads and undersleeper pads. Since one of the requirements of the vibrations induced by railway trafc, the consideration of simplied
2.5D approach is the invariance of the geometry along the longitudinal models where only the unsprung masses of the train are taken into ac-
direction, the periodic support of the rails cannot be considered. By count is a reasonable approach. Therefore, the present case only considers
that reason, the sleepers were simulated as a uniformly distributed the unsprung masses of the train and they are simulated as rigid masses
mass (303 kg/m) and the resilient elements of the track were also (details about the modeling strategy can be found in Alves Costa, 2012).

Fig. 11. Time history of the vertical velocity of the rail. Fig. 12. Vertical velocity of the rail in the frequency domain.
1340 P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343

Fig. 13. Vertical velocity of the railway track slab in: a) time domain; b) frequency domain.

The unevenness prole of the track, which is the source of the approach is more inaccurate when compared to exact methodologies, as
dynamic traintrack interaction mechanism, was not measured. There- the 3D FEMBEM coupling, the balance between accuracy and simplicity
fore it was estimated by an optimization procedure in order to obtain a makes the former approach an interesting one. Consequently, the
reasonable match between predicted and measured frequency content monkey tail model was adopted to compute the impedance of the
of the vertical velocity in the rail. After this update procedure, it was footings (Ibsen and Liingaard, 2006a; Ibsen and Liingaard, 2006b). The
found that the synthetic unevenness prole depicted in Fig. 9 gives followed approach is quite simple but does not allow the inclusion
rise to a reasonable agreement between experimental and numerical of the ground layering. By that reason, for the computation of the
vertical velocity of the rail for the frequency range up to 80 Hz, as monkey-tail impedance coefcients, an equivalent homogeneous
shown and discussed in the following section. The synthetic unevenness ground with a shear wave velocity of 280 m/s was assumed instead of
prole was generated S0 = 1 106 m3 and w = 3 on the PSD equation the variation over depth illustrated in Fig. 3. A detailed description of
(see equation United Nations, D.o.E.a.S.A., 2014). the particular aspects of the modeling strategy can be found on previous
In regard to the building, a simplied model based on 3D FEM was works of the authors, namely on Lopes et al. (2014a); Lopes et al. (2014b).
constructed, where the main structural elements were attended, namely Finally, Rayleigh damping approach was followed, being the and
the ones mentioned on Table 1. Hence, only vertical dynamics is analyzed parameters selected in order to obtain a damping ratio around 1% for
and since the nearby buildings also have 1 buried oor, it was assumed the frequency range between 5 Hz and 80 Hz.
that the ground surface corresponds to the level of the footings. As will
be seen, this simplication is acceptable for the intended modeling. Fig. 5. Experimental validation
10 depicts the adopted nite elements mesh. The simulation of the
footings was performed assuming rigid behavior. By that reason the Fig. 11 compares the measured and computed time histories of the
footings were not discretized into nite elements. vertical velocity of the rail. The homologous results, but in the frequency
The SSI effects, which are quite relevant for achieving an accurate domain, are given in Fig. 12. As mentioned, the measured dynamic
assessment of the building response due to the train passage (Lopes response of the rail was used on the assessment of the unevenness
et al., 2014b), are considered using a lumped parameter approach in prole of the rails. Actually, the numerical results depicted in Fig. 11
order to represent the contribution of the ground. As previously and Fig. 12 were computed assuming the unevenness prole shown
shown by Lopes et al. (2014b), although the lumped parameter on the previous section. As can be observed, assuming the synthetic

Fig. 14. Time history of the vertical velocity of the slabs for different building oors: a) 5th oor; b) 7th oor.
P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343 1341

Fig. 15. Running rms of the vertical velocity of the slabs for different building oors: a) 5th oor; b) 7th oor.

unevenness prole shown in Fig. 9, it is possible to achieve a reasonable Even with a reasonable match found between measured and pre-
match between numerical and experimental dynamic responses of the dicted response of the track, the main goal of the study is the assessment
rail, with special focus on the frequency content illustrated in Fig. 12, of the capacity of the proposed numerical model to reproduce the
where the main components of the measured vertical velocity of the dynamic response of the building.
rail were reasonably reproduced by the numerical model. The high fre- Fig. 14 shows the computed and measured time histories of the ver-
quency content in the frequency range between 40 Hz and 60 Hz, which tical velocity in the 5th and 7th building oors. Both experimental and
is visible in both numerical and experimental results, is due to the reso- numerical results were ltered in order to remove the contribution of
nance of the wheelset over the track. This conclusion is reached assum- frequency components above 80 Hz. As demonstrated, a reasonable
ing an analogy with a one degree of freedom system where the mass is match between computed and measured results was obtained,
given by the wheelsets (1500 hg/wheelset) and the stiffness of the highlighting the capacity of the proposed numerical model to simulate
spring corresponds to the static stiffness of the track (1.4 108 N/m). the main characteristics and trends of the response. Although there
In spite of the articial generation of the irregularities prole, it are differences between measured and computed results, the peak
should be highlighted that only the vertical velocity of the rail was velocity of the dynamic response was well reproduced in the computed
used as target information. Therefore, the ability of the model to repro- results. However, since the dynamic response of the system is quite
duce the dynamic response of the system in other observation points is complex, with a wide-ranging frequency content, the use of other indi-
not compromised by the assumptions made on the assessment of the cators of the dynamic response is also suitable as, for example, the fast
rail unevenness prole. running root mean square (rms t = 0.125 s). This indicator was com-
Fig. 13 allows the comparison between experimental and numerical puted for both oors, being the comparison between experimental and
records of the vertical velocity of the railway track slab. Once again, the computed results performed in Fig. 15. Even though the model is nota-
reasonable agreement between experimental and numerical results is bly capable to predict the main trends of the response, in Fig. 15a it is
observed in both time and frequency records. However, the numerical possible to see a slight underestimation of the peak values of the rms
model underestimates the response for frequencies up to 40 Hz though while a better agreement is found in the 7th oor (Fig. 15b).
achieving a good match when the comparison focuses on the peak It should be highlighted that large uncertainty levels are always
values of the vertical velocity. implicit in these kind of studies, which are difcult or even impossible

Fig. 16. Frequency spectra of the vertical velocity of the slabs for distinct building oors: a) 5th oor; b) 7th oor.
1342 P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343

Fig. 17. One-third octave spectra of the vertical velocity of the slabs at different building oors: a) 5th oor; b) 7th oor (gray line numerical results; black line experimental results).

to fully reproduce in the numerical analysis. Several studies point out for the simulation of the train; ii) a 2.5D FEMPML approach for the
differences around 10 dB between experimental and predicted results simulation of the tracktunnelground system; iii) the 3D FEM for the
(Hunt and Hussein, 2007; Jones et al., 2012), which deserves some simulation of the nearby building. Trying to reduce the complexity of
attention since 10 dB is a huge difference. the modeling, a lumped-parameter approach was followed to include
Hence one of the main descriptors of the dynamic response is the the soilstructure interaction behavior of the building. This simplica-
frequency content, Fig. 16 shows the frequency spectra of the time tion revealed a good performance with a large reduction of complexity
records detailed in Fig. 14. of the modeling strategy.
Focusing on Fig. 16a, it is possible to see, from a general point of The experimental data used in the validation of the model were col-
view, a good match between predicted and experimental results. Never- lected in a previous campaign performed by CEDEX (2003). A deep dis-
theless, the numerical model is overestimating the response on the low cussion about the modeling simplications and options was performed
frequency range, namely around the frequencies of 5 to 8 Hz. This level before the comparison between numerical and experimental results.
of accuracy is partially lost when analyzing the response of the 7th oor The most relevant lack of experimental data refers to the absence of
slab (Fig. 16b), where larger differences between observed and computed information about the track unevenness, which is essential for the run-
results are more obvious. However, it should be stressed that even the ning of a study as the presented one. However, to overturn this draw-
experimental results can contain an appreciable degree of noise. For back, an optimization procedure was developed in order to nd an
instance, the concentrated peak of the experimental response for the unevenness prole compatible with the measured vertical velocity of
frequency of 10 Hz in Fig. 16b is not compatible with any rational inter- the rail due to the train passage.
pretation, and should be related to noise or the working frequency of The comparison between experimental and numerical results
any equipment located in the surroundings of the observation point. revealed an acceptable agreement in both the track and the building.
Despite the differences pointed out above, from a general point of Regarding the response of the building, notwithstanding the good
view, it can be stated that a reasonable or even a good agreement was agreement found in the time domain representation, some differences
found between predicted and measured results. To avoid some very become more clear in the frequency domain analysis. Actually, in spite
particular aspects of the linear frequency spectra, which sometimes of the good agreement found on the frequency range between 20 Hz
don't have a real consequence on the response, the one-third octave and 60 Hz, where the main energy content in concentrated, in frequen-
spectra of the computed and measured vertical velocity of the slabs cy range up to 20 Hz the differences between numerical and experimen-
are represented in Fig. 17. The thresholds curves proposed by Gordon tal data are more relevant.
(1999) for sensitive equipment are also plotted in the gures. The ndings of this study are relevant because they allow concluding
As expected, there are some differences between computed and that the proposed numerical approach, based on a dynamic sub-
measured data, mainly in the low frequency range, i.e., up to 20 Hz. Any- structuring strategy, constitutes an interesting framework for the
way, in the most relevant frequency range, i.e., from 20 Hz to 60 Hz, a prediction of vibrations in urban environment due to subway railway
quite good agreement between numerical and experimental results trafc. Moreover, the versatility of the proposed sub-models, based on
was achieved, reporting differences smaller than 4 dB. Moreover, the nite element concepts, allows including complex geometries for both
numerical model was able to reproduce the main behavior trends of the building and the tracktunnelground system. For that reason, the
the system, namely the attenuation of vibration levels with the increas- proposed numerical model proves to be an interesting framework for
ing in height of the observation point. the design of mitigation measures based on a deep understanding of
the problem, allowing obtaining a holistic picture of it.
6. Conclusions

The present paper focuses on the experimental validation of a com- Acknowledgments

prehensive numerical approach previously proposed by the authors for
the prediction of vibrations inside buildings due to railway trafc in tun- The rst author wishes to acknowledge FCT for the nancial support
nels (Lopes et al., 2014a; Lopes et al., 2014b). The numerical approach provided by the grant SFRH/BD/69290/2010.
followed corresponds to a balance between accuracy and complexity. The collaboration of ADIF with the availability of the experimental
By that reason, different modeling approaches are adopted as function data used in the experimental validation of the model is greatly appre-
of the specicities of each subdomain, namely: i) a multi-body model ciated and acknowledged by the authors.
P. Lopes et al. / Science of the Total Environment 568 (2016) 13331343 1343

Authors wish to thank to Dr. Miguel Ferraz, from U. Porto, the sup- Hussein, M., H. Hunt, K. Kuo, P. Alves Costa and J. Barbosa, The use of sub-modelling tech-
nique to calculate vibration in buildings from underground railways Proc. Inst. Mech.
port given in the development of the numerical model of the building. Eng. F J. Rail Rapid Transit, 2015. 229(3): p. 303314.
Ibsen, L., Liingaard, M., 2006a. Lumped-Parameter Models. Aalborg University, p. 36.
References Ibsen, L., Liingaard, M., 2006b. Application of Lumped-Parameter Models, in DCE Techni-
cal Report No. 12. Aalborg University, Aalborg, p. 39.
Alves Costa, P., Calada, R., Cardoso, A., 2012a. Inuence of train dynamic modelling strat- Jones, S., Kuo, K., Hussein, M., Hunt, H., 2012. Prediction uncertainties and inaccuracies
egy on the prediction of track-ground vibrations induced by railway trafc. Proc. Inst. resulting from common assumptions in modelling vibration from underground rail-
Mech. Eng. F J. Rail Rapid Transit 226 (4), 434450. ways. Proc. Inst. Mech. Eng. F J. Rail Rapid Transit 226 (5), 501512.
Alves Costa, P., Calada, R., Silva Cardoso, A.A., 2012b. Trackground vibrations induced by Kouroussis, G., Connolly, D.P., Verlinden, O., 2014a. Railway induced ground vibrations
railway trafc: in-situ measurements and validation of a 2.5D FEM-BEM model. Soil a review of vehicle effects. Int. J. Rail Transport. 2 (2), 69110.
Dyn. Earthq. Eng. 32 (1), 111128. Kouroussis, G., Pauwels, N., Brix, P., Conti, C., Verlinden, O., 2014b. A numerical analysis of
Amado Mendes, P., Alves Costa, P., Godinho, L., Lopes, P., 2015. 2.5D MFS-FEM model for the inuence of tram characteristics and rail prole on railway trafc ground-borne
the prediction of vibrations due to underground railway trafc. Eng. Struct. http://dx. noise and vibration in the Brussels Region. Sci. Total Environ. 482-483, 452460. Lai, C., A. Callerio, E. Faccioli, V. Morelli and P. Romani, Prediction of railway-induced
Braun, H., Hellenbroich, T., 1991. Messergebnisse von strassenunebenheiten. VDI Ber. 877, ground vibrations in tunnels J. Vib. Acoust., 2005. 127(5): p. 503514.
4780. Lombaert, G., Franois, S., Verbraken, H., Degrande, G., Thompson, D.J., 2014. Numerical,
CEDEX, 2003. Estudio de vibraciones inducidas por la explotacin ferroviaria en la Cuesta experimental and hybrid methods for the prediction of railway-induced ground vi-
de San Vicente n36 y en el tnel del campo del moro en el pasillo verde ferroviario bration. In: Cunha, A., et al. (Eds.), 9th International Conference on Structural Dynam-
de Madrid. CEDEX, Madrid. ics, EURODYN 2014, pp. 9199 (Porto).
Clouteau, D., Arnst, M., Al-Hussaini, T., Degrande, G., 2005. Free eld vibrations due to dy- Lopes, P., Alves Costa, P., Calada, R., Silva Cardoso, A., 2012. Modeling of vibrations in tun-
namic loading on a tunnel embedded in a stratied medium. J. Sound Vib. 283 (12), nels using the nite element method with perfectly matched layers. In: Pombo, J.
173199. (Ed.), First International Conference on Railway Technology: Research, Development
Colao, A., Alves Costa, P., Connolly, D., 2015. The inuence of train properties on railway and Maintenance. Las palmas de gran canaria, Spain.
ground vibrations. Struct. Infrastruct. Eng. Lopes, P., Alves Costa, P., Calada, R., Silva Cardoso, A., 2013. Numerical modeling of vibra-
1025291. tions induced in tunnels: A 2.5D FEMPML approach. In: Xia, H., Calada, R. (Eds.),
Connolly, D.P., Kouroussis, G., Woodward, P.K., Giannopoulos, A., Verlinden, O., Forde, Trafc Induced Environmental Vibrations and Controls: Theory and Application.
M.C., 2014. Scoping prediction of re-radiated ground-borne noise and vibration Nova, pp. 133166.
near high speed rail lines with variable soils. Soil Dyn. Earthq. Eng. 66, 7888. Lopes, P., Alves Costa, P., Ferraz, M., Calada, R., Silva Cardoso, A., 2014a. Numerical model-
Coulier, P., Lombaert, G., Degrande, G., 2014. The inuence of sourcereceiver interaction ing of vibrations induced by railway trafc in tunnels: from the source to the nearby
on the numerical prediction of railway induced vibrations. J. Sound Vib. 333 (12), buildings. Soil Dyn. Earthq. Eng. 61-62, 269285.
25202538. Lopes, P., Alves Costa, P., Calada, R., Silva Cardoso, A., 2014b. Inuence of soil stiffness on
Fernandez, J., 2014. Estudio numrico de vibraciones provocadas por el trco ferroviario vibrations inside buildings due to railway trafc: numerical study. Comput. Geotech.
en tneles en el dominio del tiempo: anlisis geotcnico, validacin experimental y 61, 277291.
propuesta de soluciones, in Escuela Tcnica Superior de Ingenieros de Caminos, Ma, M., Markine, V., Liu, W., Yuan, Y., Zhang, F., 2011. Metro train-induced vibrations on
Canales y Puertos. Coruna, Universidad de Coruna. historic buildings in Chengdu. J. Zhejiang Univ. Sci. A (Appl. Phys. Eng.) 12 (10),
Forrest, J.A., Hunt, H., 2006. Ground vibration generated by trains in underground tunnels. 782793.
J. Sound Vib. 294 (45), 706736. Melis, M., 2011. In: C. de, E.T.S.d.I.d.C. ferrocarriles, Canales y Puertos (Eds.), Apuntes de
Franois, S., 2008. Nonlinear Modelling of the Response of Structures due to Ground Vi- introduccin al Proyecto y Construccin de Tneles y Metros en suelos y rocas
brations, in Departement Burgerlijke Bouwkunde. Leuven, Katholieke Universiteit blandas o muy rotas. Universidad Politcnica de Madrid, Madrid.
Leuven. Rieckh, G., Kreuzer, W., Waubke, H., Balazs, P., 2012. A 2.5D-FourierBEM model for vibra-
Galvn, P., Franois, S., Schevenels, M., Bongini, E., Degrande, G., Lombaert, G., 2010. A 2.5D tions in a tunnel running through layered anisotropic soil. Eng. Anal. Boundary Elem.
coupled FEBE model for the prediction of railway induced vibrations. Soil Dyn. 36, 960967.
Earthq. Eng. 30 (12), 15001512. Sheng, X., Jones, C., Thompson, D., 2004. A theoretical study on the inuence of the track
Gordon, C., 1999. Generic vibration criteria for vibration-sensitive equipment. Proc. SPIE on train-induced ground vibration. J. Sound Vib. 272, 909936.
3786, 2239. United Nations, D.o.E.a.S.A., 2014. Population Division World Urbanization Prospects: The
Gupta, S., Hussein, M., Degrande, G., Hunt, H., Clouteau, D., 2007. A comparison of two nu- 2014 Revision, Highlights (ST/ESA/SER.A/352).
merical models for the prediction of vibrations from underground railway trafc. Soil Verbraken, H., Lombaert, G., Degrande, G., 2011. Verication of an empirical prediction
Dyn. Earthq. Eng. 27 (7), 608624. method for railway induced vibrations by means of numerical simulations. J. Sound
Gupta, S., Degrande, G., Lombaert, G., 2009. Experimental validation of a numerical model Vib. 330 (8), 16921703.
for subway induced vibrations. J. Sound Vib. 321, 786812. Vogiatzis, K., Kouroussis, G., 2015. Prediction and efcient control of vibration mitigation
Hansen, H.M.M., 2005. High-Speed Ground Transportation. Noise and Vibration Impact using oating slabs: practical application at Athens metro lines 2 and 3. Int. J. Rail
Assessment. Federal Railroad Administration. Transport. 3 (4), 215232.
Hunt, H., Hussein, M., 2007. Vibration from railways: can we achieve better than Wolf, J.P., 1994. Foundation Vibration Analysis Using Simple Physical Models. Prentice
+/ 10 dB pridiction accuracy? ICSV14 14th International Congress on Hall, Englewood Cliffs.
Sound and Vibration (Cairns, Australia) Yang, Y., Hung, H., 2008. Soil vibrations caused by underground moving trains. J. Geotech.
Hussein, M.F.M., Hunt, H.E.M., 2009. A numerical model for calculating vibration due to a Geoenviron. 134 (11), 16331644.
harmonic moving load on a oating-slab track with discontinuous slabs in an under- Zhai, W., Cai, Z., 1997. Dynamic interaction between a lumped mass vehicle and a dis-
ground railway tunnel. J. Sound Vib. 321, 363374. cretely supported continuous rail track. Comput. Struct. 63 (5), 987997.
Hussein, M.F.M., Franois, S., Schevenels, M., Hunt, H.E.M., Talbot, J.P., Degrande, G., 2014.
The ctitious force method for efcient calculation of vibration from a tunnel embed-
ded in a multi-layered half-space. J. Sound Vib. 333 (25), 69967018.