2011 Leon Sleepers

  • Upload
    omar-c

  • View
    218

  • Download
    0

Embed Size (px)

Citation preview

  • 8/11/2019 2011 Leon Sleepers

    1/10

    A procedure to develop scalable models for the transient response of sleepersin conventional and high-speed railway lines and implementation to thevertical vibration mode

    E. Leon, D.C. Rizos n, J.M. Caicedo

    Department of Civil and Environmental Engineering, University of South Carolina, 300 Main Street, Columbia, SC 29208, USA

    a r t i c l e i n f o

    Article history:Received 23 June 2010

    Received in revised form

    12 November 2010

    Accepted 17 November 2010

    a b s t r a c t

    This paper presents a procedure to develop scalable reduced models for the through-the soil interactionand traveling wave effects of distant sleepers in a long railway track. For development purposes, and,

    without loss of generality, the geometry of the sleepers is consistent with the UIC-60 track system

    commonly usedin European high speed railand the vertical vibration mode is considered. Theballastand

    the effects of soil layering are not considered in the present paper; however, it is the subject of ongoing

    research. The proposed reduced models are based on B-Spline impulse response functions (BIRF) of the

    sleepersonly as computed through boundary elementmethod (BEM) solutionsof thefull model,preserve

    the frequency content of the full models, and they are highly accurate within the assumptions of linear

    isotropic and homogeneous soil media. They are expressed in a scalable form with respect to soil

    propertiesand sleeperspacing.In particular, theBIRFs of distantsleepers can be accuratelyapproximated

    by appropriate scaling operations of time and amplitude of a reference sleeper BIRF while retaining all

    dynamic characteristics of the full model. Three main scaling parameters are proposed: (i) the apparent

    propagation velocity,(ii) the geometric dampingcoefficient, and(iii) the soil propertiesof a reference soil

    (i.e., the shear modulus and shear wave velocity). The models are validated through comparisons with

    other BEMsolutions,and the accuracy andefficiency areestablished. Theproposed models are developed

    as part of an NSF funded research on vibrations inducedby high-speed rail traffic and are consistent with

    the associated train and rail models and a multi-system interface coupling (MSIC) technique that were

    developed as a part of the project and presented in companion papers. The proposed procedure forms the

    framework for developing scaled reduced models for other vibration modes and different sleeper

    geometries and can be generalized to include any foundation type or layered soil profiles.

    Published by Elsevier Ltd.

    1. Introduction

    High-speed trains (HST) is a popular transportation mode in the

    world and plans are being made in the United States for several high-

    speed rail systems to be constructed in different areas of the country

    over the next several decades. The definition of HSTs typically

    suggests a passenger train traveling in excess of 160 km/h(100 mph), though HSTs in service may travel faster than 320 km/h

    (200 mph). As the HST speeds continue to increase, potential

    problems arise when the train travels over soft soils with inherently

    lower Rayleighand shear wavevelocities.In such cases, thewave field

    generated by the HST propagates at speeds lower than the speed of

    the trains that generated them, creating a phenomenon equivalent to

    the sonic boom and supersonic travel, as described for example in

    Ref. [1]. The characteristics of the wave propagation due to the

    passage of HST, e.g., frequency content and peak particle velocity

    differ from the vibrations caused by conventional trains. Although

    these wavefields do notcause large strains andsoil nonlinearities, the

    vibrations could be potentially damaging to both the trains and the

    infrastructure facilities and cause annoyance to passengers and

    residents impacting negatively the public perception on the benefitsof HSR. It is, therefore, evident that an accurate prediction of HST

    inducedvibrations needs to account forthetraveling waveeffects and

    the through the soil interaction between sleepers at relatively large

    distances in addition to the rail, track and vehicle dynamics. Such

    tasks are among the most computationally expensive and require

    efficient approaches and procedures.

    The finite element method (FEM), boundary element method

    (BEM), and hybrid FEMBEM are among the most popular techni-

    ques for transient analysis and wave propagation and suitable for

    such studies. General literature reviews on time and frequency

    domain BEM, FEM, and coupled BEFE methods for problems in

    transientanalysishave beenpresented in Refs. [24] among others.

    Contents lists available at ScienceDirect

    journal homepage:w ww.elsevier.com/locate/soildyn

    Soil Dynamics and Earthquake Engineering

    0267-7261/$- see front matter Published by Elsevier Ltd.

    doi:10.1016/j.soildyn.2010.11.006

    n Corresponding author. Tel.: +1 803 777 6166; fax: + 1 803 777 0670.

    E-mail addresses: [email protected] (E. Leon),[email protected] (D.C. Rizos),

    [email protected] (J.M. Caicedo).

    Soil Dynamics and Earthquake Engineering 31 (2011) 502511

    http://-/?-http://www.elsevier.com/locate/soildynhttp://localhost/var/www/apps/conversion/tmp/scratch_1/dx.doi.org/10.1016/j.soildyn.2010.11.006mailto:[email protected]:[email protected]:[email protected]://localhost/var/www/apps/conversion/tmp/scratch_1/dx.doi.org/10.1016/j.soildyn.2010.11.006http://localhost/var/www/apps/conversion/tmp/scratch_1/dx.doi.org/10.1016/j.soildyn.2010.11.006mailto:[email protected]:[email protected]:[email protected]://localhost/var/www/apps/conversion/tmp/scratch_1/dx.doi.org/10.1016/j.soildyn.2010.11.006http://www.elsevier.com/locate/soildynhttp://-/?-
  • 8/11/2019 2011 Leon Sleepers

    2/10

    Standard FEM procedures for wave propagation in infinite media

    truncate the infinite extents by introducing fictitious boundaries

    and cannot accurately represent the radiation condition in the soil

    region. Proposed alternatives include (i) modeling of large compu-

    tational domain with material damping, e.g.[5], (ii) use of springs

    and visco-elastic dampers along the fictitious boundaries, e.g. [6],

    and (iii) development of special types of elements that model the

    infinite extents at the fictitious boundaries, e.g. [710]. Such

    alternatives are either less accurate, or computationally expensiveor suitable for frequency domain analysis only. BEM for wave

    propagation and sleeper soil interaction analysis are developed in

    the frequency or time domain. The frequency-domain techniques

    are based on the calculation of transfer functions followed by

    suitable convolution operations to obtain the response to the train-

    induced moving load[1117]. Time domain BEM formulations for

    general 3-D wave propagation problems in the direct time can be

    classified in view of the associated fundamental solutions in three

    major types: (i) Dirac-d fundamental solutions, e.g.[18], (ii) step

    impulse fundamental solutions, e.g. [19], and (iii) high order B-

    Spline fundamental solutions, e.g.[20,21].

    Coupled formulations based on thecombination of the FEMand

    BEM methods have also been presented, e.g.[2225], providing an

    attractive numerical toolfor problemspertaining to coupled media.

    Recent research work on the analysis of ground vibrations pro-

    duced by HST has been reported in the frequency domain[2629],

    and in the time domain[3033]. Time domain simulations of the

    full length of a railroad track using FEM or BEM or combined

    FEMBEM models are often time consuming since they involve a

    significantly high number of degrees of freedom. A number of

    techniques for reducing the size of the models have been reported

    in the literature. One of the earliest and most popular reduction

    methods used in FEManalysis is the staticor Guyanreduction [34],

    but it is not exactfor dynamic analysisdue tothe presence of inertia

    forces in the system. Model reduction techniques reported in the

    literature for dynamic problems include the dynamic reduction

    method [3537], theimprovedreductionsystem (IRS) method [38],

    the iterative IRS technique[39], the component mode synthesis

    (CMS) method[40], andthe system equivalent reduction expansionprocess (SEREP) method[41], among others. The aforementioned

    techniques operate directly on the system matrices and are either

    less accurate or require modal analysis to identify potential modes

    that can be eliminated (or ignored). The B-Spline impulse response

    techniques[21]present an alternate approach in model reduction

    and are based on establishing signature responses of systems

    subjected to impulses of B-Spline modulation using the BEM

    reported in Ref. [20]. Once these characteristic responses are

    established the time history of the system response to arbitrary

    excitations is computed by a mere superposition. In the case of a

    long straight railroad track these characteristic impulse responses

    of the sleepers are repetitive for any two sourcereceiver sleeper

    systems andscalable fordifferent soil types.Theyhavebeenusedto

    generate the B-Spline impulse response function (BIRF) matrix of alarge system based on the analysis of much smaller systems

    consisting of only a few sleepers [1]. Even though this approach

    is more efficient than implementing the BEM on the entire track,

    the calculation of the characteristic responses can be time con-

    suming as the distance between the sleepers increases and further

    optimization is in order.

    This paper presents a procedure for developing scalable reduced

    models of distant sleepers ina long railway track accounting forthe

    through-the soil interaction and traveling wave effects. For devel-

    opment purposes and without loss of generality the geometry of

    the sleepers is consistent with the UIC-60 track system commonly

    used in European High Speed Rail, and the vertical vibration mode

    is considered. The effects of the ballast and soil layering are not

    considered in the present paper; however, it is the subject of

    ongoing research. The proposed reduced models are based on

    B-Spline impulse response functions (BIRF) of the sleepers only as

    computed through boundary element method (BEM) analysis of

    the full model, preserve the frequency content and all dynamic

    characteristics of the associated full models, and they are highly

    accurate within the assumptions of linear isotropic and homo-

    geneous soil media. The output informationof the full model can be

    retrieved routinely on an as needed basis. The B-Spline BEM

    formulation within the framework of B-Spline impulse responsetechniques [20] presents an innovative and efficient way to

    simulate in almost real time the wave propagation in soils

    accounting for kinematic interaction effects with the sleepers.

    Inertia interaction effects are addressed through coupling with

    other solution techniques as reported in Refs. [1,24]. The reduced

    models are expressed in a scalable form with respect to soil

    properties and sleeper spacing. In particular, the BIRFs of distant

    sleepers can be accurately approximated by appropriate scaling

    operations of time and amplitude of a reference sleeper BIRF while

    retaining all dynamic characteristics of the full model. Three main

    scaling parameters are introduced: (i) the apparent propagation

    velocity, (ii) the geometric damping coefficient, and (iii) soil

    properties of a reference soil (i.e., the shear modulus and shear

    wave velocity). The proposed models are developed as part of an

    NSF funded research on vibrations induced by high-speed rail

    traffic and are consistent with the associated train and rail models

    and a multi-system interface coupling (MSIC) technique developed

    by the research team and presented in companion papers. The

    proposed approach forms the framework for developing scaled

    reduced models for other vibration modes and different sleeper

    geometries and can be generalized to include any foundation type

    or layered soil profiles. In the following sections, the proposed

    technique is discussed. First, the BIRF techniques are introduced

    briefly, thena description of the proposed model reduction concept

    follows and, the scaling operations that lead to the proposed

    scalable forms are presented. The validation of the proposed

    closed-form solutions is provided and, finally for a more compre-

    hensive use of these expressions, the reader is referred to an

    implementation example.

    2. Overview of B-Spline impulse response techniques

    A detailed formulation of the employed 3-D BEM is too

    extensive and beyond the scope of this paper and can be found

    in Ref. [1,42]. The BEM uses the time domain fourth-order B-Spline

    fundamentalsolutions of the3-D full space along with higher order

    spatial discretization of the boundary. These fundamental solu-

    tions are developed assuming an excitation of the form of a fourth-

    order B-Spline polynomials. The B-Spline polynomials are piece-

    wise smooth polynomials and are the basis functions in function

    interpolation techniques. The fourth-order polynomials, when

    used as the excitation functions in the development of thefundamental solutions, satisfy the continuity requirements. The

    B-Spline impulse response technique is based on: (1) the relation

    between the B-Spline impulse excitation and the associated

    B-Spline impulse response function (BIRF) of the elastodynamic

    system as computed through a BEM method using the proposed

    B-Spline fundamental solutions and (2) the representation of any

    function as a linear combination of B-Spline polynomials. Conse-

    quently,if the response of theelastodynamicsystem to an arbitrary

    excitation function is sought after, the response can be computed

    through a mere superposition of the BIRFs without requiring a

    rigorous BEM solution to be performed anew for each arbitrary

    excitation. The technique implies linearity of the problem.

    The boundary integral equation associated to the Navier

    Cauchy governing equations of motion is expressed in a discrete

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511 503

  • 8/11/2019 2011 Leon Sleepers

    3/10

    form yielding a system of algebraic equations at step Nrelating

    displacementsu to forcesfat discrete boundary nodes in the BEM

    model and at discrete time instants tjand tj, as

    cutN XN 1

    n 1

    UtNftNn 2TtNutNn 2 N 1,2,. . .,Nend 1

    where c is a discontinuity term, and U and T are the BEM

    coefficient matrices associated with the B-Spline fundamental

    solutions. Eq. (1) can be solved in a time marching scheme forNend time steps yielding theB-Spline impulse response (BIRF)of the

    system. In order to derive equivalent, time-dependent flexibility

    matrices associated with the loaded part of the BE model

    (e.g. points of application of external excitation), a unit B-Spline

    impulse force djf, perturbs each active degree of freedom,

    j 1,y,NN, at a time and successive solutions of Eq. (1) yield the

    B-Spline impulse response vectors, bNj utN. These BIRFs can be

    collected in matrix form as

    BN bN1 b

    N2 . . . b

    Nj . . . b

    NNN

    h i 2

    Matrix B represents the BIRF associated to the displacements of

    the BEM nodes due to concentrated unit forces applied in the

    directions of thedegreesof freedom that vary in time according to a

    B-Spline function, and captures the traveling wave characteristics.Subsequently, the response, uN, of the elastodynamic system

    subjected to arbitrary external forces fP applied at the nodes

    are computed at time step Nas

    uN XN 1

    n 1

    BNPNn 2 3

    By separating known from unknown quantities at time step N,

    and assuming that PN 0, Eq. (3) can be expressed as

    uN ZFN HN 4

    where

    Z 2B1 B2; HN XN 1

    n 3

    BnFNn 2B1FN1 5

    Matrix Z represents the flexibility matrix of the loaded BEM

    region and vector HN represents the influence of the response

    history on the current step. The flexibility matrix is independent of

    time; however, vector HNneeds to be evaluated at every time step.

    The proposed method is implemented in two major phases. The

    first phase calculates the BIRF matrices of the boundary of the

    domain. The evaluation of the BIRF matrices, BN, is computationally

    intensive; however, theyare independent of the externalexcitation

    and need to be evaluated only once for a given model geometry.

    The second phase calculates the response of the boundary of the

    domain given a known excitation force time history. In the

    presence of rigid bodies in contact with the free surface, such as

    the case of the railroad sleepers considered in this work, the same

    method canbe used along with the rigid surface boundary elementreported in Ref. [43]to compute the BIRF functions of the soil

    sleeper interface.

    3. Proposed model reduction

    3.1. The concept

    For the analysis of a real system consisting of a large number of

    sleepers where all sleepers are loaded in an asynchronous pattern

    due to themoving load, the calculationof theBIRF matrix, Eq.(2), is

    a computationally expensive task if all sleepers and an extended

    free field are to be considered simultaneously in the full model. In

    such a complete approach, every sleeper in the system needs to be

    loadedby a B-Spline impulse andthe response of all sleepers needs

    to be computed. Although this approach is an accurate representa-

    tion of the railsleeper system, it may be simplified by observing

    that: (1) the response of each rigid sleeper is described by six

    degrees of freedom,i.e., three translations andthree rotations of the

    center of the sleeper, (2) onlythe vertical andhorizontal translation

    degrees of freedom of the sleepers are coupled with the rail, and

    (3) repetitive responses of the sleepers are expected when a single

    representative sleeper is loadeddue to the assumption of a straighttrack. Therefore, only one sleeper should be loaded by a B-Spline

    excitation and the BIRFs of all sleepers should be computed. Since

    thefreefield effects arealready accounted forin thecomputed BIRF

    functions of the sleepers, the size of the full model is therefore

    reduced to the number of degrees of freedom of all sleepers in the

    system. This represents a first level of model reduction. A second

    level of reduction is attained next. In view of the wave attenuation

    and the HST speed, only a finite number of sleepers are expected to

    have a significant contribution to the response at any given time.

    Furthermore, depending on the geometry and spacing of sleepers,

    cross interaction effects are expected to be important for only the

    first twoto four sleepers adjacentto the loadedone, as discussed in

    Ref. [1] andverified in this work. Therefore, it is feasible to generate

    BIRF matrices of a large system based on the analysis of much

    smaller systems consisting of only a few sleepers. Consider, for

    example, the system shown in Fig. 1 that consists of a source

    sleeper,s , and m receiver sleepers on each side of the source. As

    proposed in Ref. [1] and assuming for demonstration purposes that

    the cross-interaction effects are significant within two sleepers

    away of the source, two groups of sleepers are identified. The first

    group (Group A) pertains to the source and the two adjacent

    sleepers oneachside.For this groupthe BIRF is computedthrougha

    rigorous BEMsolution. The second group (Group B) consists of pairs

    of source/receiver sleepers with the receiver sleeper not belonging

    to the first group. In this case all other sleepers between the source

    and receiver (light shaded sleepers in Fig. 1) are omitted and it is

    assumed that cross-interaction effects and wave scattering due to

    neighboring sleepers are negligible. This simplifying assumption is

    expected to produce conservative results since the sleepersbetween source and receiver act also as wave barriers. In Ref.[1]

    the BIRFs for this group are also computed numerically through a

    rigorous BEM solution with all sleepers between source and

    receiver being omitted. Subsequently, the BIRF of theentire system

    is formed by combining the BIRFs of the two groups. However, as

    the model considers larger number of sleepers, the BEM solution

    even for group B will become computationally expensive, since

    large regions of the free field should be modeled to account for the

    surface waves correctly.

    The proposed reduced model can be efficiently obtained for the

    systems in group B if the BIRF of any receiver sleeper in the group

    can be accurately approximated by an appropriate scaling of the

    BIRF ofa reference sleeper that is computedbased onBEM solutions

    of the full model. To demonstrate this concept, a representativesourcereceiver sleeper system in Group B (e.g. S-j in Fig. 1) is

    considered as the reference system. The distance between the

    source and receiver system isDrefand the BIRF is computed by the

    BEMmethod forthisreference system. Inthe present work theBIRF

    of another sleeper at distance DDi, may be computed through

    appropriate scaling of the reference system with respect to time

    and amplitude.

    3.2. The features of BIRFs of 2-sleeper systems

    In order to develop the proposed reduced models, the char-

    acteristics of the B-Spline impulse response functions (BIRF) need

    to be explored first. To this end, the BEM methodology outlined in

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511504

  • 8/11/2019 2011 Leon Sleepers

    4/10

    section 2 is used to calculate the BIRFs of a series of source-receiver

    sleeper systems spaced at variable distances.Fig. 2illustrates such

    BIRF responses for six different sourcereceiver spacings for the

    vertical vibration mode.

    The intercept of each BIRF indicates the distance between thesource and receiver while the actual amplitude of the BIRF is

    omitted.The repetitive nature of the BIRFs is evident. Thevariations

    in amplitude, and number of high amplitude peaks with distance

    from the source areapparent. The observed horizontal shift in time

    is due to the quiescent past of the response while the amplitude

    decrease is due to attenuation.Despite these variations, a similarity

    in the overall shape of the response curves is observed and can be

    interpreted by considering the characteristics of the wave propa-

    gationfromthe source towardthe receiversleepers. The BIRF of any

    sleeper is affected by both body waves (Pressure P- and Shear S-)

    and surface waves (Rayleigh R-) propagating through the

    semi-infinite soil. The general characteristics of the stress wave

    propagation theory are observed in the sleeper BIRFs illustrated in

    Fig. 2. Additional features inherent to the wave generation and

    propagation in soil due to the particular sleepers arrangement are

    observed:

    1. The motion between the first P- and S-wave arrivals becomes

    more pronounced and the number of peaks between these

    arrival times increases as the distance from the excitation

    source increases. This part of the motion includes contributions

    from: (a) the incident P-wave field emanated at the source

    sleeper and propagating with velocity cp; and (b) interaction of

    the scattered P-waves with the S-wave incident field.

    2. The part of the motion between the first S- and R-wave arrival

    includes contributions from, (a) the incident S-wave field

    emanated at the source sleeper and propagating at cS; and

    (b) interaction of the scattered S-waves propagating at cSwith

    the R-wave incident field propagating atcR. The contribution of

    scatteredP-waves is notanticipated to be significant in thistime

    interval.

    3. The response of the sleeper following the first R-wave arrival

    seems to be composed mainly of the incident R-waves ema-

    nated at the source sleeper propagating with velocity cR. The

    most likely contribution might be that of the scattered S-waves

    since their velocity is only about 10% higher than the velocity of

    R-waves.

    Comparing the BIRFs of the receivers in any 2-sleeper systems,time and amplitude of a reference receiver BIRF can be scaled with

    respect to distance and the soil properties to obtainaccurate match

    of the response of the other.

    3.3. Scaling of time

    During any time intervalDtthe group of waves of the same type

    emanated from the sourcesleeper travels a distance cDt, where cis

    the wave velocity of the medium. If the group of waves reaches

    receiver sleeperjat timetjand receiver sleeperiat timetithen the

    horizontal distance between the two sleepers is covered within

    Dt titj DiDj

    c 6

    Fig.2. Comparisonof thevertical componentof displacement of receiverslocated at

    increasing distances from the source as indicated by the intercept. The source

    function is a B-Spline impulse.

    Fig. 1. An m-sleeper system and the computation of the BIRF function.

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511 505

  • 8/11/2019 2011 Leon Sleepers

    5/10

    Hence, the problem of calculating a scaled time depends on the

    availability of the wave propagation velocities, c. At each time

    instant,t, the group of waves reaching the receiver sleeper can be

    thought of as combination of P-, S-, and R-waves. In lieu of

    considering the separate wave velocities, a parameter henceforth

    calledapparent propagation velocity Cis introduced to describe the

    propagation velocity of the wave packet of different wave types.

    The apparent velocity is a function of time, in contrast to the wave

    velocity of a particular wave type. This parameter captures theeffects of the geometry of the excitation source and the reflections

    from the receiver sleeper which, as discussed in the previous

    section, are perceived as factors that contribute to the complexity

    of the wave packet that travels from the source to the receiver.

    Assuming that C is known, the desired scaled time tscaled is a

    function of the apparent propagation velocity itself and is readily

    calculated as

    tscaled tref DDrefC 7

    The procedure to estimate the apparent propagation velocity is

    discussed in section 4.2

    3.4. Scaling of amplitude

    The attenuation of amplitude of vibration with distance attrib-

    uted only to geometric (radiation) damping may be described by

    Bornitzs equation[44]

    wi wj DjDi

    n8

    where wi and wj are vibration amplitudes at distance Di and Dj from

    a source of vibration and n is a geometric damping coefficient.

    Values ofn have been provided in the literature[45]for various

    types of excitation sources and types of propagating waves. For

    example, the amplitude of body waves at distance R from the

    source, decreases in proportion to 1/R (n1) except along the

    surface of the elastic half-space where the amplitude decreases in

    proportion to 1/R2

    (n2). The amplitude of the Rayleigh wavesdecreases in proportion to 1/R0.5 (n 0.5). Nevertheless, the values

    ofnfor the system inFig. 1are anticipated to differ from the latter

    ones due to theeffects of thegeometryof the excitation sourceand

    the reflections from the receiver sleeper. Therefore, for the accurate

    prediction of the wave attenuation due to applied loading in the

    current problem,nneeds to be quantified as a function of time for

    the specificgeometryof theproblem.The procedure to estimate the

    geometric attenuation coefficient is discussed in section 4.3. Using

    the estimated n values, the desired scaled amplitude wscaled is

    readily calculated as

    wscaled wref Dref

    D

    n9

    3.5. Soil Property scaling

    Scaled time and amplitude, as defined in Eqs. (7) and (9),

    respectively, are associated with the soil medium of a reference

    system. This implies that thesame soil medium properties exist for

    the reference system and the system whose scaled response is

    desired. However, different soil medium properties can be con-

    sidered for the two systems. Considering that the arbitrary system,

    i, consists of sleepers of the same dimensions as the reference

    system andGi,cs,i,riand Gref,cs,ref,rrefare the shear modulus, shearwave velocity and mass density of the arbitrary and the reference

    system, respectively, the scaled response at distance D can be

    computed as

    witi wscaled GrefGi

    wscaled cs,refcs,i

    2rrefri

    , ti tscaled cs,refcs,i

    10

    The proposed scaling method is only valid for sleepers having

    the same widths and is applicable only in the absence of

    connecting rails.

    4. Parameter estimation and development of scalable BIRF

    In order to obtain the closed-form solutions for the BIRF of a

    receiver sleeper as a function of distance from the source, the

    scaling functions in Eqs. (7) and (9) involve the estimation of

    the scaling parameters, i.e., the apparent propagation velocity andthe geometric damping coefficient. This section discusses the

    procedure to estimate these parameters and presents the scaling

    functions to obtainthe BIRF solutions of thereducedmodel. It must

    be noted that the scaled model should capture all the significant

    information of the numerically obtainedBIRFs. To this end,the BIRF

    of a receiver sleeper is obtained numerically for a number of

    sourcereceiver distances. Subsequently, discrete values of the

    apparent velocity and damping coefficient are obtained at char-

    acteristic time instants from the BIRFs. Finally, the apparent

    velocity and damping functions are established through a regres-

    sion analysis of the discrete values.

    4.1. Sleeper BIRF function

    The sleeper BIRF functions are calculated for the track geometry

    presented in Ref.[1]. The dimensionsof each sleeperare 0.285 m 2.5

    m witha centerto centerspacingof 0.955 m, yielding an edge-to-edge

    spacing of 0.67 m. The sleepers are assumed rigid and massless, and

    therefore, only kinematic interaction effects are accounted for.

    Complete interaction effects are readily accommodated through

    coupling of the proposed formulation with the FEM solution for

    the track-rail model where the inertia interaction due to sleepers

    mass is accounted for. The sleepersare assumed to rest directlyon the

    horizontal free surface of a linear elastic half-space with which they

    remain always in contact. A reference soil having a relatively low

    shear wavevelocity is usedin theanalysis andthe assumedproperties

    are shown inTable 1. As shown inFig. 3, the boundary of the half-

    space is discretizedinto 8-nodeboundary elements, andthe motionofthe sleeper is expressed by the 3 translations and 3 rotations of a

    reference node, R, at the sleeper center. Since only the vertical

    vibration mode is of interest in this work, only the vertical degree

    of freedom (DOF) of the source sleeper is excited with a fourth-order

    Table 1

    Input data for BEM rigorous solution.

    Property Value SI (US) Property Value SI (US)

    Sleeper length,L 2.5 m (98.425 in.) B-spline support, Dt 2 103 s

    Sleeper width,b 0.285 m (11.220 in.) Time step, dt 5 104 s

    Lames,l 5.19 107 KPa (7.52 106 lbf/in2) P-wave velocity,cp 246.07 m/s (9687.76 in/s)

    Shear modulus,G 3.46 107 KPa (5,02 106 lbf/in2) S-wave velocity, cs 131.53 m/s (5178.31 in/s)

    Density,r 2,000 Kg/m3 (0.002246 slugs/in3) R-wave velocity,cR 121.93 m/s (4800.28 in/s)

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511506

  • 8/11/2019 2011 Leon Sleepers

    6/10

    B-Spline impulse of duration Dt2 103 s, shown in the inset

    ofFig. 4. The associated time histories of the response of all DOFs

    are computed using the BEMmethod [21] at discretetimes tnnDt/4.

    The BIRF of the receiver sleepers pertaining to the vertical trans-

    lation is monitored and shown inFig. 4. The six curves representsix different sourcereceiver sleeper systems with the following

    source-receiver distances in increasing order: D22.85, D33.99,D44.845, D77.695, D10 10.545 and D1212.54 m, where sub-

    scripts indicate the number of sleepers between the source and

    receiver. It is noted that these distances are not always multiples of

    the spacing between the adjacent sleepers in order to facilitate the

    efficient discretization of the free field. However, this discrepancy

    does not affect in any way the proposed scaling operations. The latter

    is applicable to any spacing and not restricted to the UIC-60 track

    system.

    4.2. Determination of apparent propagation velocity

    The first step in determining the apparent velocity as a function

    of time is to compute discrete valuesbasedon the BIRFs obtained in

    the previous section. To this end, one response at a time is

    considered as the reference response where the sourcereceiver

    spacing is Di. The remaining responses, with sourcereceiver

    spacing Dj, are the target responses and are compared to the

    reference in view of their characteristic peaks. Thus,the time lapse,

    Dt, between the time,ti, of the reference response where a peak is

    observed and the time, tj, of the target response where the same

    peak takes place is measured. Subsequently, Eq. (6) can be solved

    for the velocity, C. This velocity is a measure of the apparent

    velocity at time ti for this pair of reference and target responses.

    Therefore, discretevalues of the apparentvelocity canbe calculated

    for all combinations of reference-target BIRFs. The calculated

    apparent velocity is presented in a graph form in Fig. 5 for four

    reference systems.

    It is observedthat the calculated Cfunctions are piecewisesmoothfunctionsdefined in four timeintervals, i.e.[tp, tc1], [tc1, tc2], [tc2, ts]and

    [ts,N) for each sourcereceiver distance. It is also observed that the C

    values corresponding to tp, tc1, tc2 and ts do not depend on the distance

    and are determined as Ctp cp 246m=s, Ctc1 2cpcs

    228m=s, Ctc2 cs 132m=s and Cts cR 122m=s. Times tpand ts correspond to the first arrival of the P- and S-waves,

    respectively, and tc1otc2 correspond to time instances where the

    computed apparent velocity is significantly affected by the convo-

    luted pressure and shear waves. A regression analysis,shown in Fig. 6,

    reveals a linear relationship between tc1,tc2and the distance of the

    reference sleeper from the source with an R2 value greater than 0.99.

    Generalized expressions for any reference soil with cs,refare derived

    tc1 1

    cs,ref 0:579 Dref 0:552, Dref44m; tc1in s 11

    tc2 1

    cs,ref 0:934 Dref0:776, Dref44m; tc2in s 12

    Fig. 6 also shows the characteristic times tc3 and tc4 that areused

    in the estimate of the geometric damping coefficient discussed in

    the next section. Because of definition tc1otc2, these linear

    relationships apply only for Dref44 m (more than 3 sleepers

    between source and reference sleeper). This is consistent with

    the findings in Ref. [1] where it is reported that the number of

    significant sleepers for cross-interaction effects is 24 on each side

    of the source sleeper.

    It is observed that in all but the secondinterval, [tc1, tc2],a linear

    relationship accurately describes the variation of the apparent

    Fig. 3. Discretized free field sleepersoil interface with 8 node boundary elements.

    Fig. 4. Discrete BIRF functions for the vertical translation as obtained from a

    rigorous BEM solution for six different sourcereceiver distances.

    Fig. 5. Calculated apparent propagation velocity with respect to time for four

    different distances of the reference sleeper from the source and the soilsleeper

    system ofTable 1. Curve corresponding to reference sleeper at D10 is selected to

    illustrate time limitsof thefourdiscrete parts that areobservedin allof thederived

    curves.

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511 507

  • 8/11/2019 2011 Leon Sleepers

    7/10

    velocity with time. The apparent velocity in the interval [tc1,tc2], is

    better approximated by a power function. In summary, the

    following closed-form expressions are derived for C

    Ctref

    cp,ref 2cp,refcs ,refcp,ref

    tc1,reftp,ref treftp,ref tp,refotrefotc1,ref

    a tbref

    tc1,refotrefotc2,ref

    cs,ref cR,refcs,refts,reftc2,ref

    treftc2,ref tc2,refotrefots,ref

    cR,ref ts,refotref

    8>>>>>>>>>>>:

    13

    where a 2cp,refcs,ref=tbc1,ref

    cs,ref=tbc2,ref

    and b ln2cp,refcs,ref=cs,ref=lntc1,ref=tc2,ref are estimated by enforcing the con-

    tinuity conditions at the ends of interval [tc1,tc2].

    4.3. Determination of geometric damping coefficient

    The geometric damping coefficient that appears in Eq.(9) can be

    determined in a manner similar to the apparent propagation

    velocity described in the previous section. From the amplitudes

    of the vertical displacement at characteristic peaks along two

    discrete BIRF functions, the geometricdamping coefficient n can be

    calculated from Eq. (8). Scaled in time responses are now selected

    consecutively to represent the reference response and coefficient n

    is calculated as the one required to match the amplitude values of

    the reference response with those of the remaining target

    responses simultaneously. The calculatedcoefficientn withrespect

    to time is illustrated inFig. 7(a)(d).

    Similarly to the closed-form expression derivation process for C,

    distinct limits are selected for the derivation of the closed-form

    expressions for n. Beginningfromthe left these are: [tp, tc1], [tc1, tc2],

    [tc2, tc3], [tc3, ts], [ts, tR], [tR, tc4] an d [tc4, N). The n valuescorresponding totp,tc1,tc2,tc3,ts,tRand tc4do not depend on the

    distanceand are determined as n(tp)1.5, n(tc1)1.5, n(tc2) 1.0,

    n(tc3) 2.0n(ts)0.0,n(tR)0.7 andn(tc4) 1.0. Timestc1, tc2,

    tc3, tc4, have been presented inFig. 6 and similarly to tc1 and tc2,

    linear relationships between tc3, tc4 and the distance of the

    reference sleeper from the source are derived

    tc3 1

    cs,ref 0:973 Dref0:789, Dref44m; tc3in s 14

    tc4 1

    cs,ref 1:026 Dref 1:960, Dref44m; tc4in s 15

    In all intervals a linear relationship accurately describes the

    variation of the geometric damping coefficient with time.In

    Fig. 6. Arrival timetc1andtc2as-obtained from the discrete BIRF functions ofFig. 4

    for six different sourcereceiver distances and the soilsleeper system ofTable 1.

    Arrival time tc3 and tc4 discussed in Section 4.3 are also plotted. Best fit curves

    indicate a linear relationship with respect to distance of the reference sleeper from

    the source.

    Fig. 7. Calculated geometric damping coefficient with respect to time at (a) D4;

    (b)D7; (c)D10 and (d) D12. A smoother curve, discretized in seven time intervals,

    representing the derived closed-form expressions for n is also illustrated.

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511508

  • 8/11/2019 2011 Leon Sleepers

    8/10

    summary, the following expressions are derived for n:

    ntref

    1:5 tp,refotrefotc1,ref

    1:5 2:5tc2,reftc1,ref treftc1,ref tc1,refotrefotc2,ref

    1:0 1:0tc3,reftc2,ref treftc2,ref tc2,refotrefotc3,ref

    2:0 2:0ts,reftc3,ref treftc3,ref tc3,refotrefots,ref

    0:7tR,refts,ref

    trefts,ref ts,refotrefotR,ref

    0:7 1:7tc4,reftR,ref treftR,ref tR,refotrefrtc4,ref

    1:0 tc4,refotref

    8>>>>>>>>>>>>>>>>>>>>>>>>>>>>>:

    16

    4.4. Closed-form expressions

    Closed-form expressions for scaling of time and amplitude are

    summarized in Tables 2 and 3, respectively. These expressions

    provide a simple algorithm that needs only a reference response at

    a distance Drefto predict the response at a greater distance D due to

    thesame excitationforce.The reference response is associatedwith

    the referencesystem whose geometrywas presented in Table 1 and

    soil medium properties as represented by cp,ref, cs,ref, cR,ref. The

    predicted response is associated with an arbitrary system of the

    same geometry as the reference but different soil medium asrepresented by cs,i. With such expressions the computational effort

    is significantly minimized since only the reference response at a

    short distance Dref needs to be calculated through rigorous

    boundary element analysis. For longer distances and the discrete

    time intervals over theduration of the reference response shown in

    Tables2 and 3, thecorresponding closed-form expressions are used

    in lieu of boundaryelement analysis. The limits of the discretetime

    intervals of the reference response are: (a) the arrival times of the

    primary waves which are computed as: tp,refDref/cp,ref,ts,refDref/

    cs,ref and tR,ref Dref/cR,ref; and (b) arrival times tc1,ref, tc2,ref,

    tc3,ref, tc4,ref, which are calculated from Eqs. (11), (12), (14) and

    (15), respectively. The proposed closed-form expressions are onlyvalid forDref44 m and for sleepers having the same geometry as

    the ones of the reference system presented in Table 1.

    5. Model validation and implementation

    The performance of the closed-form solutions is evaluated

    quantitatively on the actual response of a sleeper located at a

    distance D D1616.245 m from the source. The source sleeper is

    excited with a fourth-order B-Spline impulse force and the BIRF of

    the receiver is computed numerically using the BEM method. It

    should be emphasized that this BIRF was not considered in the

    developmentof the scaling parameters and serves as a benchmark.

    To validate the proposed model theBIRF of the receiver of thesame

    system is computed by the scaling of the two reference responses,

    DrefD77.695 m, and DrefD1212.54 m. A comparison of the

    three BIRFs is shown in Fig. 8. The agreement of the BIRFs as scaled

    from the two reference responses with the BIRF obtained numeri-

    cally is evident.

    The time history of the response of the receiver atD16due to an

    arbitrary load is shown inFig. 9. The load function is shown as an

    inset in the figure. The response is computed numerically based on

    the superposition scheme described by Eq. (3).

    In the particular example Bis a scalar and is represented by the

    two scaled BIRFs and the computed BIRF presented in Fig. 9,

    whereasP is the load function. Comparison between the real and

    the scaled responses indicate a high accuracy of the scaled BIRFs.

    Therefore, it is demonstrated that the proposed method can

    reproduce the BEM solutions accurately for all practical purposesin an extremely efficient manner.

    6. Conclusions

    Thephysical railsleepersystem that consists of a large number

    of sleepers loaded in an asynchronous pattern due to the moving

    Table 2

    Closed-form expressions for scaling of time.

    Reference time Time scaling

    Apparent propagation velocity, C Scaled time,ti

    trefotp,ref tref DDref

    cs,refcs,i

    tp,refotrefotc1,refcp,ref

    2cp,refcs,refcp,reftc1,reftp,ref

    treftp,ref tref DDref

    V

    cs,refcs,i

    tc1,refotrefotc2,ref a tbref

    a 2cp,refcs,ref

    tbc1,ref

    cs,ref

    tbc2,ref

    b ln 2cp,refcs,ref

    cs,ref

    =ln

    tc1,reftc2,ref

    tc2,refotrefots,ref cs,ref cR,refcs,refts,reftc2,ref

    treftc2,ref

    trefots,ref cR,ref

    Table 3

    Closed-form expressions for scaling of amplitude

    Reference time Amplitude scaling

    Geometric damping coefficient,n Scaled amplitude,wi

    trefotp,ref 0.0wref

    DrefD

    n

    cs,refcs,i

    2rrefri

    tp,refotrefotc1,ref 1.5

    tc1,refotrefotc2,ref 1:5 2:5

    tc2,reftc1,ref treftc1,ref

    tc2,refotrefotc3,ref1:0

    1:0

    tc3,reftc2,ref treftc2,ref

    tc3,refotrefots,ref2:0

    2:0

    ts,reftc3,ref treftc3,ref

    ts,refotrefotR,ref 0:7

    tR,refts,ref trefts,ref

    tR,refotrefotc4,ref 0:7 1:7

    tc4,reftR,ref treftR,ref

    trefotc4,ref 1.0

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511 509

  • 8/11/2019 2011 Leon Sleepers

    9/10

    train load is reduced to twosubsystems. The first one addresses thenear to the source group of sleepers and the through the soil cross-

    interaction effects and has been presented in previous work of the

    authors. The second subsystem addresses the far from the source

    sleepers andthe travelingwave effects andis thefocus of this work.

    This paper presented a procedure to develop scalable reduced

    modelsin the time domain forthe vibrationresponse analysisof far

    fromthe source railroad sleepers to a B-Splineimpulseexcitation. A

    two sleeper system is considered consisting of a source sleeper

    loaded by a vertical B-Spline impulse function and a receiver

    sleeper located at a large distance from the source where the

    B-Splineimpulse responsefunction will be computed. It is assumed

    thatfor the far sleepers the cross-interactioneffectswith the source

    are not important, and the scattering due to the presence of other

    sleepers between the source and receiver will yield non-conserva-tive results. In order to compute the BIRF of any two sleepers a full

    model that includes thefree surface in a discrete form is considered

    and BEM solutions are computed. The number of degrees of

    freedom of the full model is in the order of thousands. The BIRF

    function of the receiver sleeper itself represents a reduced model

    that can be used in lieu of the full modelto compute theresponseof

    the receiver sleeper to any arbitrary excitation and can be readily

    combined with the near the source sleeper subsystem. The BIRF

    model preserves the frequency content and all dynamic character-

    istics of the full model; however, it has a maximum of only six

    degrees of freedom in general. Furthermore,it is demonstratedthat

    this BIRF function model is scalable with respect to soil properties

    and distance from the source and the proposed scaling procedure

    introduced three scaling parameters, i.e., the soil shear wave

    velocity,the apparentwave propagation velocity andthe geometric

    damping coefficient. The soil shear wave velocityis used forscaling

    the model with respect to the soil properties and is constant. The

    apparent wave velocity and geometric damping coefficients are

    used forscaling the model with respect to distance from thesource,

    are piecewise smooth functions of time, and are presented in a

    closed form through regression analysis. The scaled models

    increase further the computational efficiency of the BIRF models

    since only the BIRF of a reference sourcereceiver sleeper systemneeds to be computed. BIRF models for the remaining receiver

    sleepers in the railroad track are accurately approximated through

    the scaling procedures. The scaling operations however are not

    valid for source receiver distances less than four sleepers apart.

    Althoughthe development of the procedure focused on the vertical

    vibration mode of the receiver sleeper, other important modes can

    be approximated in a similar way. The reduced models are

    validated through comparisons with other BEM solutions, and

    theiraccuracy and efficiency are established. The present approach

    can be extended to more complex problems like layered soil, effect

    of topography or coupled vibrations of nearby structures, as

    reported in forthcoming papers.

    Acknowledgement

    This work was supported by the National Science Foundation

    under Grant# CMMI-0800414. Any opinions, findings and conclu-

    sions or recommendations expressed in this material are those of

    the authors and do not necessarily reflect those of the National

    Science Foundation.

    References

    [1] OBrien J, Rizos DC. A 3D BEM-FEM methodology for simulation of high speedtrain induced vibrations. Soil Dynamics and Earthquake Engineering 2005;25:289301.

    [2] Beskos DE. Boundaryelementmethodsin dynamic analysis:Part II 19861996.ASME Applied Mechanics Reviews 1997;50(3). 149-19.

    [3] Mackerle J. Earthquake analysis of structures: FEM and BEM approaches Abibliography (19951998). Finite Elements in Analysis and Design 1999;32:11324.

    [4] Rizos DC, Karabalis DL. Soil-Fluid-Structure Interaction, Invited Chapter. In:KauselE, ManolisGD, editors. Wave MotionProblemsin Earthquake Engineer-ing. Southampton: Computational Mechanics Publications; 1999.

    [5] Song C, Wolf JP. Dynamic stiffness of unbounded medium based on damping-solvent extraction.Earthquake Engineeringand StructuralDynamics1994;23:16981.

    [6] Hall L. Simulations and analyses of train-induced ground vibrations; acomparative study of two- and three-dimensional calculations with actualmeasurements, PhD dissertation, Dept. of Civil and Environmental Engineer-ing, Royal Institute of Technology, Stockholm, Sweden, 2000.

    [7] LysmerJ, SeedHB, Udaka T, Huang RN,Tsai CF.Efficientfiniteelement analysisof seismic soil structure interaction, Report No. EERC 75-34, EarthquakeEngineering Research Center, University of California, Berkeley, 1975.

    [8] Gupta S, PenzienJ, LinTW, YehCS. Three-dimensional hybridmodelingof soil-structure interaction. Earthquake Engineering and Structural Dynamics

    1982;10:6987.[9] Kausel E, Tassoulas JL. Transmitting boundaries: a close form comparison,Bulletin Seismic. Soc. Am 1981;71:14359.

    [10] Guddati MN, Tassoulas JL. Space-time finite elements for the analysis oftransient wave propagation in unbounded layered media. International

    Journal of Solids and Structures 1999;36:4699723.[11] Clouteau D, Degrande G, Lombaert G. Numerical modelling of traffic induced

    vibrations. Meccanica 2001;36:40120.[12] Lombaert G, Degrande G. Experimental validation of a numerical prediction

    model for free field traffic induced vibrations by in situ experiments. SoilDynamics and Earthquake Engineering 2001;21:48597.

    [13] LombaertG, DegrandeG. Theexperimental validationof a numericalmodelforthe prediction of the vibrations in the free field produced by road traffic.

    Journal of Sound and Vibration 2003;262:3093 1.[14] Lombaert G, Degrande G, Clouteau D. Numerical modelling of the free field

    traffic-induced vibrations. Soil Dynamics and Earthquake Engineering2000;19:47388.

    [15] Lombaert G, Degrande G, Kogut J, Franc-ois S. The experimental validation of anumerical model for the prediction of railway induced vibrations. Journal of

    Sound and Vibration 2006;297:51235.

    Fig. 8. Comparison of scaled BIRFs with computed BIRF for D D1616.245 m.

    Fig. 9. Comparison of response time histories due to arbitrary force excitation for

    DD1616.245 m.

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511510

  • 8/11/2019 2011 Leon Sleepers

    10/10

    [16] Degrande G, Schillemans L. Free field vibrations during the passage of a ThalysHST at variable speed. Journal of Sound and Vibration 2001;247:13144.

    [17] Andersen L, Nielsen SRK. Boundary element analysis of the steady-stateresponse of an elastic half-space to a moving force on its surface. EngineeringAnalysis with Boundary Elements 2003;27:2338.

    [18] Banerjee PK, Ahmad S, Manolis GD. Transient elastodynamic analysis of 3-Dproblems by boundary element method. Earthquake Engineering and Struc-tural Dynamics 1986;14:93349.

    [19] Karabalis DL, Beskos DE. In: Beskos DE, editor. Dynamic soil-structureinteraction. Amsterdam: North Holland; 1987. Chapter 11.

    [20] Rizos DC, Karabalis DL. An advanced direct time domain BEM formulation for

    general 3-D elastodynamic problems. Computational Mechanics 1994;15:24969.

    [21] Rizos DC, Loya K. Dynamic and seismic analysis of foundations based on freefield B-Spline characteristic response histories. Journal of EngineeringMechanics 2002;128(4):43848.

    [22] Von Estorff O, Firuziaan M. Coupled BEM/FEM approach for nonlinear soil/structure interaction. Engineering Analysis withBoundary Elements 2000;24:71525.

    [23] Yazdchi M, Khalili N, Valliappan S. Dynamic soil-structure interactionanalysisvia coupled finite element-boundary element method. Soil Dynamics andEarthquake Engineering 1999;18:499517.

    [24] Rizos DC, Wang J. Coupled BEM-FEM solutions for direct time domain soil-structure interaction analysis. Engineering Analysis with Boundary Elements2002;26:87788.

    [25] Karabalis DL, Mohammadi M. 3-D dynamic foundation-soil-foundation interac-tion on layeredsoil. Soil Dynamics and EarthquakeEngineering 1998;17:13952.

    [26] Lombaert G, Degrande G. Ground-borne vibration due to static and dynamicaxle loads of InterCity and high-speed trains. Journal of Sound and Vibration

    2009;319:103666.[27] Auersch L. The excitation of ground vibration by rail traffic: theory of vehicletracksoil interaction and measurements of high-speedlines. Journal of Soundand Vibration 2005;284:10332.

    [28] Sheng X, Jones CJC, Thompson DJ. Prediction of ground vibration from trainsusing the wave number finite and boundary element methods. Journal ofSound and Vibration 2006;293:57586.

    [29] Xia H, Cao YM, Roeck GDe. Theoretical modeling and characteristic analysis ofmoving-train induced ground vibrations. Journal of Sound and Vibration2010;329:81932.

    [30] Galvn P, Domnguez J. High speed train-induced ground motion and inter-action with structures. Journal of Sound and Vibration 2007;307:75577.

    [31] Galvn P, Domnguez J. Analysis of groundmotion dueto moving surfaceloads

    induced by high-speed trains. Engineering Analysis with Boundary Elements

    2007;31:93141.[32] Galvn P, Domnguez J. Experimental and numerical analysis of vibrations

    induced by high-speed trains on the CordobaMalaga line. Soil Dynamics and

    Earthquake Engineering 2009;29:64157.[33] Galvn P, Romero A, Domnguez J. Fully three-dimensional analysis of high-

    speed traintracksoil-structure dynamic interaction. Journal of Sound and

    Vibration 2010;329:514763.[34] Guyan RJ. Reduction of stiffness and mass matrices. AIAA Journal 1965;3:380.[35] Leung AYT. An accurate method of dynamic condensation in structural

    analysis.InternationalJournalfor Numerical Methods in Engineering 1978;12:

    170515.[36] Leung AYT. An accurate method of dynamic substructuring with simplified

    computation. International Journal for Numerical Methods in Engineering

    1979;14:124156.[37] Paz M. Dynamic condensation. AIAA Journal 1984;22:7247.[38] OCallahan JC. A procedure for an improved reduced system (IRS) model. In:

    Proceedingsof the seventh international modal analysis conference,Las Vegas

    1989, p. 1721.[39] Friswell MI,Garvey SD, PennyJET. Modelreductionusingdynamic anditerated

    IRS techniques. Journal of Sound and Vibration 1995;186(2):31123.[40] Craig R, Bampton M. Coupling of substructures in dynamic analysis. American

    Institute of Aeronautics and Astronautics Journal 1968;6(7):13139.[41] OCallahan JC, Avitabile P, Riemer R. System equivalent reduction expansion

    process (SEREP). In: Proceedings of the seventh international modal analysis

    conference, Las Vegas 1989, p. 2937.[42] OBrien J. Traininducedvibrations in a ground-track system,MS thesis, Dept.of

    Civil Engineering, Univ. of South Carolina, 2004.[43] Rizos DC. A rigid surface boundary element for soil-structure interaction

    analysis in the direct time domain. Computational Mechanics 2000;26(6):

    58291.[44] Bornitz G. Uber die Ausbreitung der von Grosskolbenmaschinen erzeugten

    Bodenschwingungen in die Tiefe. Berlin: J. Springer; 1931.[45] Gutowski T, Dym C. Propagation of ground vibration: a review. Journal of

    Sound and Vibration 1976;49(2):17993.

    E. Leon et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 502511 511