Engineering Structures 174 (2018) 396–406

Contents lists available at ScienceDirect

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

Design of railway bridges for dynamic loads due to high-speed traffic T

⁎
Sebastian Schneider , Steffen Marx
Institute of Concrete Construction, Leibniz University Hannover, Appelstraße 9a, 30167 Hannover, Germany

A R T I C LE I N FO A B S T R A C T

Keywords: This contribution deals with the dynamic behaviour of railway bridges for high-speed traffic and investigates the
Continuous beams impact of changes to the structural system on the dynamic response due to crossing trains. Multi-span simply
Dynamics supported beams represent the most unfavourable system for high-speed trains, structurally as well as aesthe-
Dynamic increase tically. Therefore, alternative structural systems were analysed to find out whether the dynamic characteristics
Railway bridges
of railway bridges can be adjusted in the design stage. Because of the strong interaction between the crossing
Design
High-speed traffic
train and the bridge structure the impact of changes to the structure is very difficult to estimate a priori. The
Resonance internal forces in continuous beams with lengths exceeding 30 m are generally smaller than those in single-span
beams with the same cross section and the speeds at which they can be crossed are significantly higher. By
adding haunches to the beams those eigenfrequencies whose eigenmodes exhibit curvatures at the supports can
be increased. Shortening the end spans leads to an increase in all eigenfrequencies and hence in resonance
speeds. Using the findings from this article the dynamic stability of high-speed railway bridges can be improved
at the preliminary design stage.

1. Introduction dynamic optimisation of structures are given. Simply supported single-

Railway bridges are subjected to large static and extremely high
dynamic loads. The dynamic load is the governing factor for the design
of high-speed railway bridges and needs to be taken into account even
at the concept design stage, when the fundamental system properties
are defined [1,2]. At this stage, the type of structural system, the
stiffness and mass distribution and the system damping are of crucial
significance. These parameters in particular determine whether a
bridge will experience significant vibration or even resonance during
train crossings [3,4]. Currently, labour-intensive dynamic investiga-
tions are necessary to determine whether the general rules for the op-
timum static design of bridges (for determining the span lengths of
continuous beams, for example, or for deciding where to add haunches)
result in a structure which will experience favourable or unfavourable
structural effects during train crossings. There are several further
parameters that barely affect the static loading of the structure but that
have a big impact on its dynamic behaviour due to train crossings, such
as the number of spans of continuous beams.
This contribution aims to aid engineers in the concept design of
high-speed railway bridges. Based on extensive parameter studies, the
main parameters influencing the dynamic stability of bridges under
high-speed rail traffic are explained and recommendations for the
span beams and continuous beams with varying span lengths and dif-
ferences in haunch configuration are compared.
2. Dynamic analyses of bridges for high-speed traffic
For road and railway bridges used by moderate-speed traffic the
load increase due to dynamic traffic load can generally be estimated
sufficiently accurately using system-dependent, particularly span-
length-dependent, dynamic coefficients. The forces governing the de-
sign are determined with static equivalent load models and multiplied
by a dynamic coefficient. But the dynamic coefficients do not apply to
railway bridges experiencing resonance. For such bridges it is necessary
to carry out more detailed dynamic analyses with calculated simula-
tions of train crossings. Resonance excitation can be caused by a train
with relatively uniform axle distances and uniform speed, if the induced
excitation frequency corresponds to an eigenfrequency of the structure.
The corresponding resonance speeds vres,i,j,k, are the product of the j-th
flexural eigenfrequencies nj−1 and the wavelength of the excitation
frequencies λres,i,k.

⁎
Corresponding author.
E-mail address: schneider@ifma.uni-hannover.de (S. Schneider).
URL: http://www.massivbau.uni-hannover.de (S. Schneider).

https://doi.org/10.1016/j.engstruct.2018.07.030
Received 5 January 2018; Received in revised form 9 July 2018; Accepted 10 July 2018
Available online 30 July 2018
0141-0296/ © 2018 Elsevier Ltd. All rights reserved.
S. Schneider, S. Marx Engineering Structures 174 (2018) 396–406

vres,i,j,k = n j−1·λ res,i,k = n j−1· Table 1

Specifications of the HSLM-A.
i = 1, 2, 3, … and j = 1, 2, 3, …
Lc,k / i
n j−1 ⩽ n max (1) HSLM Number Coach Number Train Distance Axle Total
of length of axles length between loads axle
According to [5], the first flexural eigenfrequency is denoted by n0. coaches Lc [m] [–] Ltrain bogies d P [kN] loads
N [–] [m] [m] ΣP
The wavelength of the excitation frequency λres,i,k is defined as the ratio
[kN]
of the almost uniform axle distance (coach length Lc,k) of the respective
train load model k and an integral divisor i. The divisor i takes into A01 18 18 50 397.526 2.0 170 8500
account after how many eigenvibrations the j-th eigenmode is excited. A02 17 19 48 398.526 3.5 200 9600
A03 16 20 46 397.526 2.0 180 8280
For bridges with low-speed traffic and/or with certain eigenfrequency
A04 15 21 44 394.526 3.0 190 8360
limits as well as for structural systems for which sufficient data on the A05 14 22 42 389.526 2.0 170 7140
behaviour during operation is available (for example, single-span A06 13 23 40 382.526 2.0 180 7200
frames with certain minimum dimensions), a dynamic analysis does not A07 13 24 40 397.526 2.0 190 7600
need to be carried out [5]. A08 12 25 38 387.526 2.5 190 7220
A09 11 26 36 375.526 2.0 210 7560
A10 11 27 36 388.526 2.0 210 7560
2.1. Issues to consider in the concept design stage

The resonant excitation of a railway bridge can result in the criteria
for the ultimate and serviceability states not being met. Hence, if the
risk of resonance cannot be excluded with simple criteria, a computa-
tional simulation of the train crossings is required. To do this, the load
models of the currently operating trains and, if considering the inter-
operability criteria of European high-speed train routes, also the high-
speed load models (HSLM) must be taken into account. Fig. 1 shows the
configuration and Table 1 shows the specifications of the HSLM-A.
For ultimate limit state analyses the entire range of speeds up to the
legal speed limits or the highest possible train speed multiplied by a
factor of 1.2 needs to be investigated in small steps. This results in an
extremely large number of simulations and a very labour-intensive
design process, even with currently available high-performance com-
puters and software. If changes are made based on the results of the
simulations, the entire design process has to be repeated. Because of the
highly nonlinear behaviour of the reactions of the structure under re-
sonance and the still insufficient knowledge about the impact of the
various system parameters it is often necessary to run several iterations
to arrive at the optimum design.
Drastic measures are generally required to achieve noticeable im-
provements in the dynamic behaviour at the design stage, for example
changes to the fundamental geometric parameters of the structure such
as the span lengths, the span length ratio, the construction height or
modifications to the structural system. As these parameters influence
the entire design and the structural clearance in particular, the dynamic
stability for the range of design speeds of the bridge must be in-
vestigated and shown to be adequate at an early stage in the design
(concept design).
2.2. Calculation methods and design checks
Several dynamic train simulation methods exist, and they differ
considerably with respect to computational effort and the type of input
parameters. The simplest model which also yields conservative results
uses axle loads of a crossing train expressed as load-time functions. This
means that the equations of motion can be solved using modal analysis
or the more time-intensive method of linear time-step integration.
Particularly for bridges that have very high mass compared to the
weight of a train this method yields sufficiently accurate results. This
applies to any concrete bridges, but also to steel bridges with ballast
tracks or slab tracks. Merely for very lightweight open-track steel
bridges (for example auxiliary bridges) there can be larger differences
between the results of this calculation method and those of the ap-
proach where the trains are modelled as mobile mass-spring damper
systems [6–8].
The range of design speeds that need to be analysed needs to be
determined subject to the design check criteria. According to [5] the
guiding speed for determining the range is the speed limit vö, which is
the highest speed at which the bridge is to be crossed. This speed is
defined either as the design speed of the route (or a smaller value,
depending on the alignment on the bridge) or the maximum speed
vtrain,k of the train k. For the serviceability limit state and the fatigue
limit state speeds up to 1.0 · vö have to be investigated, whereas for the
ultimate limit state speeds of up to 1.2 · vö must be considered.
In estimating the risk of fatigue failure it is extremely important to
define the operating program including the train-crossing frequencies,
because a high-speed train crossing a bridge at high velocity can cause a
high number of damage-inducing load cycles [9]. In practice, however,
it is often almost impossible to define a realistic operating program. It is
generally not possible for the operator of a high-speed route to reliably
predict the type and number of trains crossing the bridge over its life-
time.
Even if there is no risk of resonance, the fatigue limit state analysis
does not yield sensible results with the high-speed load models in the
sense of a damage accumulation calculation, as they are merely dy-
namic load models and not real train configurations. Safe fatigue design
is currently only possible by restricting the range of maximum stresses
induced by the HSLMs to a level close to the endurance limit. It is re-
commended to use the value of the S-N curve at 109 load cycles. The
relevant mean stress also has to be taken into account in the design of
concrete bridges. Using these design provisions, a bridge with an as-
sumed life span of 100 years can safely be crossed 250 times per day by
the most unfavourable high-speed train.

Fig. 1. Configuration of the HSLM-A.

397
S. Schneider, S. Marx Engineering Structures 174 (2018) 396–406

Fig. 2. Movement of axle loads along the beam.

3. Parameter studies

3.1. Fundamental approach

Because of the complexity of the dynamic response of bridge

structures to crossing trains it is very difficult to predict the qualitative
and quantitative consequences of changes to the structure at the design
stage. Hence, extensive parameter studies were carried out at the
Institute of Concrete Construction at the Leibniz Universität Hannover
to improve the design process for typical bridge cross sections and
structural systems. The calculations have been done using the FEM
program SOFiSTiK. All beams were modelled using beam elements,
assuming Bernoulli beam theory, with a single element length of 1/10
of the field length. Fig. 2 shows the principal of a simulated single span
beam subjected to the moving axle forces. The time step width for the
direct time integration method (Newmark method) was always set to
Fig. 4. Permissible deflection of railway bridges for a coach-acceleration of
0.002 s. The structural damping parameter ζ for prestressed concrete 1.0 m/s2 [10]
bridges as well as the additional damping parameter Δζ due to bridge-
train interaction given by [15] were used. Except for the investigations
shown in Fig. 6, the concrete properties of a C40 were used. single-span and multi-span bridges with constant stiffness distributions
To obtain a broad basis for comparison, single-span beams were were situated outside the grey area in Fig. 5. According to [5], dynamic
investigated in a first step. Next, continuous beams with equal spans, calculations do not have to be carried out for bridges within the grey
unequal spans and different haunch configurations were examined. The area if certain conditions are met.
structures were designed to comply with the common boundary con- The simulations were carried out for speeds between 0.9 · vres,1,1,A01
ditions of double-track railway bridges including their edge beams, and 420 km/h and hence cover the currently operating trains and
track structure (ballast track) and other railway equipment. Depending foreseeable developments in the European high-speed traffic.
on the span length, the bridge types chosen were slab bridges, slab-
with-cantilever and double-T beam bridges as well as box girder bridges 3.2. Single-span beams
(Fig. 3).
The beam height and hence the slenderness and stiffness were In the first parameter study the influence of the stiffness, mass
chosen so that the very strict deflection criterion shown in Fig. 4 for a distribution and damping on the dynamic reactions of single-span
crossing speed of 350 km/h was just met. Even so, the investigated beams was studied by varying the parameters in turn and investigating

Fig. 3. Investigated cross-section types.

398
S. Schneider, S. Marx Engineering Structures 174 (2018) 396–406

deflection criterion shown in Fig. 4 is sufficient for railway bridges

under high-speed traffic.
An evaluation of the parameter study of the single-span bridges
designed with the deflection criterion described in Section 3.1 showed
that only structures with span lengths of less than 35 m exhibited strong
resonance at the investigated speed ranges. For structures with span
lengths longer than 35 m the acceleration falls below the allowable
values [5,14] of 5.0 m/s2 (slab track limit) or 3.5 m/s2 (ballast track
limit) (Fig. 9). If the span length exceeds 40 m the calculated bending
moments are also smaller than those determined by applying load
model LM 71 multiplied by ϕ2 (Fig. 10). Considering the investigated
dependencies it can be seen that the dynamic bending moments for
bridges with the same span lengths are independent of the type of cross
section, as all cross sections possess the same stiffness according to
Fig. 5. Range of the first bending eigenfrequency depending on the span-length Fig. 4. It can hence be deduced that the comfort criterion from Fig. 4
[5] can only be applied in the cross-section design of bridges for crossing
speeds of 350 km/h if their span lengths exceed 45 m. For span lengths
≥45 m no dynamic analysis is required at the preliminary design stage
the effect on the bending moment, acceleration and deflection of a
and the internal forces can be determined by applying load model
single-span beam with a span length of L = 25 m that is crossed by the
LM 71 multiplied by ϕ2. The comfort criterion cannot be used for the
load model HSLM-A10. Further parameter studies can be found in
preliminary design of single-span beams with spans less than 45 m long,
[11,12].
as this criterion does not ensure that the acceleration and internal forces
To emphasise the influence of the beam stiffness, the Young’s
are within the design limits.
modulus of the described single-span beam was varied between
At the concept design stage it should be ascertained that the struc-
30,000 MPa and 40,000 MPa. Fig. 6 shows the calculation results for the
ture will not experience pronounced resonance under the operating
beam at mid-span. As expected, an increase in stiffness results in an
speeds. This is possible only if the resonance speeds are “shifted” out-
increase in the resonance speed (i.e. a different eigenfrequency) and
side the operating speed range as a result of high stiffness and eigen-
hence a shift of the resonance peaks on the x-axis. The maximum value
frequencies. No dynamic analysis needs to be carried out if the
of the bending moment and the acceleration due to resonance (η = 1)
minimum required eigenfrequency is determined using simple design
remain unchanged. The vertical deflection at beam mid-span is the only
tools, such as those found in [15], Appendix F, for some operating trains
parameter that can decrease due to an increase in stiffness.
and in [16] for all HSLM-A models. It should be noted, however, that
The influence of the superstructure mass is investigated in a manner
the design tools listed in [15], Appendix F, do not consider all trains
similar to the previous analysis. An increase in specific weight from
operating in Germany or other countries, such as the ICE-T, the ICE 1 or
25 kN/m3 to 50 kN/m3 causes the eigenfrequencies and hence the re-
the ICE 3, and therefore need to be used with discretion.
sonance speeds of the superstructure to decrease (Fig. 7). The bending
Even when using the tools for determining the eigenfrequencies, it is
moments and deflections at beam mid-span due to resonance remain
often helpful to check the internal forces, acceleration and deflections
unchanged, whereas the maximum acceleration decreases with in-
during the design of the bridge. To avoid having to carry out simula-
creasing superstructure mass.
tions for all HSLM-A models at the preliminary design stage, in [15],
The structural reactions due to resonance are influenced sig-
Appendix E.2, the method for calculating the dynamic aggressivity is
nificantly by the damping of the system [13]. This is illustrated in
presented, which can be used to graphically obtain the most critical
Fig. 8, which shows the calculation results for damping ratios of 0.5%,
train model. To do this, the so-called dynamic aggressivity A(L/λ) · G(λ)
1.0% and 2.0%. Subcritical damping, which is small in this case but
of each train needs to be determined based on the excitation wave-
typical for bridges, has a negligible influence on the eigenfrequencies of
length λ. The train simulation then only needs to be carried out for the
the superstructure. Therefore, no shift of the resonance peaks on the x-
train with the highest dynamic aggressivity; all the other trains are
axis occurs. However, increasing the damping causes a significant de-
covered by this analysis. This graphical method is based on the DER
crease in the maximum bending moments, acceleration and deflections.
(Décomposition de l’Excitation à la Résonance) method developed in
In the second parameter study single-span beams with span lengths
[17], with which the maximum deflections at mid-span and an ap-
between 5 m and 60 m were investigated. The single-span beams were
proximation of the acceleration of a single-span beam can be calculated.
discretised as simple beam models with a linearly elastic material law
The expression for the acceleration is split into three parts – a constant
including a Young’s modulus of 31,400 MPa (C40), and they were
Ct, an influence line A(L/λ) and a spectrum of the train G(λ).
subjected to a permanent superimposed dead load of 207 kN/m. For the
design of the cross-section it needed to be established whether the

Fig. 6. Influence of the stiffness to the maximum bending moment, the maximum acceleration and the maximum vertical deflection at midspan.

399
S. Schneider, S. Marx Engineering Structures 174 (2018) 396–406

Fig. 7. Influence of the specific weight to the maximum bending moment, the maximum acceleration and the maximum vertical deflection at midspan.

4
Ct =
m·π (2)

πL
cos( λ
)
A (L/λ) = 2L 2
( λ ) −1 (3)

i 2 i 2
1 ⎛ 2πxk ⎞ ⎛ 2πxk ⎞ ⎛
⎜ ∑ P k cos ⎝ λ ⎠ ⎟ + ⎜ ∑ P k sin ⎝ λ ⎠ ⎟ 1
G (λ) = max ⎛ ⎞ ⎛ ⎞
i = 0 to M − 1 ζXi
⎝ k=0 ⎠ ⎝ k=0 ⎠ ⎝
Xi
−e−2πζ λ ⎞
⎠ (4)

In Eqs. 2–4 L is the length of the single-span beam, m the mass

distribution, λ the excitation wavelength, Pk the k-th axle load and xk
the distance of axle k to the first axle. Parameter M represents the total
number of the single loads, Xi the length of a single train with i axles
and ζ the damping ratio. The spectrum G(λ) of a train depends mainly Fig. 9. Maximum accelerations at midspan for all investigated cross-section
on the size and number of its axle loads, the uniformity of the axle types.
configurations and the damping of the structure. The spectrum G(λ) of a
train with damping ζ = 0 is called the dynamic train signature S0(λ).
For a train with damping ζ = 0 only the axle loads and the excitation
wavelength are taken into account. Fig. 11 shows the dynamic train
signatures of all HSLM-A models.
By examining the maximum values of the dynamic aggressivity in
the range from λ = 0 to the excitation wavelength λv at 1.2 times the
speed limit (Eq. (5)), the governing load model can be identified. For a
single-span beam with L = 20 m, the first eigenfrequency n0 = 2.6 Hz
and a crossing speed v = 350 km/h the governing load model is HSLM-
A10 (Fig. 12).
λ v = 1, 2·v ö / n 0 (5)

3.3. Continuous beam with constant height and uniform span lengths

Under static loads the internal forces in a continuous beam do not

exceed those in a single-span beam of equal span length. Whether this
Fig. 10. Maximum bending moments at midspan for all investigated cross-
also holds true for dynamic loading is discussed in this section. Single-
section types.
span and continuous beams with equal span lengths and cross sections

Fig. 8. Influence of the damping to the maximum bending moment, the maximum acceleration and the maximum vertical deflection at midspan.

400
S. Schneider, S. Marx Engineering Structures 174 (2018) 396–406

Fig. 11. Dynamic signature S0(λ) of all HSLM-A. Fig. 13. Frequency ratio of continuous beams [19]

eigenfrequency of continuous beams with two, three and four spans is

Fig. 12. Dyn. aggressivity of all HSLM-A for a single span bridge with L = 20 m
and ζ = 0.01.
have the same first eigenfrequency. Because the number of flexural
eigenmodes of continuous beams increases with the number of spans, it
is necessary to find a distinct and systematic designation for the ei-
genmodes. To prevent confusion, the eigenmodes without nodes in the
span are called first-order eigenmodes, and the eigenmodes with one
node in each span are called second-order eigenmodes (see also [18]). A
node is in this context a point which does not deflect and remains on the
beam axis under eigenvibrations and only experiences rotation. Con-
tinuous beams with n spans generally have n first-order eigenmodes
(Table 2).
Fig. 13 shows the frequency ratios of all first-order eigenfrequencies
with respect to the basic flexural eigenfrequency n0 for the continuous
beams. The basic flexural eigenfrequency n0 and hence the first critical
speed at which resonance can occur at i = 1 are the same no matter
whether the structure has one or four spans. The n-th first-order flexural
higher by a factor of 1.56, 1.87 and 2.02, respectively, than that of a
single-span beam of the same span and cross section. The highest first-
order flexural eigenfrequency of continuous beams with five or more
spans is very similar, which means that continuous beams have several
resonance speeds that differ little from each other.
Continuous beams generally behave in a similar way to statically
determinate single-span beams if their structural parameters such as
stiffness, mass and damping are changed (see Figs. 6–8). However, it
still needs to be established whether continuous beams exhibit more or
less favourable dynamic behaviour than single-span beams. For two-
span beams with a short total length (less than 30 m) the dynamic train
simulation yields very similar internal forces and acceleration values for
the first first-order resonance speeds as for single-span beams of the
same span length and cross section. Hence, two-span beams with span
lengths of less than 15 m offer no advantage with respect to single-span
beams of the same dimensions, since the first first-order resonance
speed of both are the same. This is due to the fact that there is always
only one bogie on the superstructure of such short two-span bridges,
which excites the first eigenmode of the beam in a similar manner to a
single-span bridge. Continuous beams with total lengths exceeding
30 m exhibit significantly improved dynamic behaviour with respect to
single-span beams of the same dimensions, because there are always
two neighbouring bogies that are simultaneously positioned on dif-
ferent spans of the bridge and the excited vibrations hence overlap
(Fig. 14). The maximum structural reactions in the spans occur at those
resonance speeds in whose eigenmodes all spans deflect in the same
direction (Fig. 14 middle). This eigenmode is always the n-th first-order
flexural eigenmode, where n is the number of spans. Dynamic increases
in the negative moments at the supports only result from flexural ei-
genmodes which exhibit a curvature in the vicinity of the supports
(Fig. 14 right). Therefore, two-span beams with span lengths exceeding
15 m can be used by trains with significantly higher speeds and exhibit
a less severe dynamic response than multi-span bridges composed of
simply supported spans of the same dimensions.

Table 2
First order eigenmodes.
Eigenmode Single-span Two-span Three-span Four-span

1. eigenmode
1. order
2. eigenmode –
1. order
3. eigenmode – –
1. order
4. eigenmode – – –
1. order

401
S. Schneider, S. Marx
Fig. 14. Maximum bending moments of a beam with a T cross section (left: at midpan of a single-span beam with L = 25 m; middle: in the second field of a two-span
beam with L = 25 m; right: at the support of a two-span beam with L = 25 m).
Fig. 15. Maximum bending moments at midspan of a single-span beam with
L = 25 m and double T cross section.
Fig. 16. Maximum bending moments in the second field of a two-span beam
with L = 25 m and double T cross section.
In the first first-order eigenmode of continuous beams, neigh-
bouring spans deflect in different directions. A crossing train, despite its
nearly harmonic excitation force, hence causes an unloading of neigh-
bouring spans and thus a decrease or even partial cancellation of re-
sonance effects due to interference. In two-span beams a particularly
large amount of vibration of the basic eigenmode is cancelled out if the
span length L corresponds to the coach length Lc of a train. This be-
haviour is visible in Figs. 15 and 16, which show the maximum internal
forces of HSLM-A08 and HSLM-A10 versus train speed for single-span
and two-span beams with span lengths of 25 m and a first eigen-
frequency of n0 = 2.44 Hz. As expected, in a single-span bridge a
402
Engineering Structures 174 (2018) 396–406
resonance peak occurs at a speed of vres,1,1,k = Lc,k · n0, which translates
into LüP,A08 = 25 m at 220 km/h for HSLM-A08 and Lc,A10 = 27 m at
237 km/h for HSLM-A10. In a two-span bridge of the same dimensions
the maximum span moments do not occur at the first but at the second
first-order resonance speed vres,1,2,k = Lc,k · n1. With HSLM-A08 the re-
sonance peak disappears at the first eigenfrequency because of the
concordance of span length and Lc, whereas a small excitation remains
for HSLM-A10 at the first eigenfrequency. Furthermore, in two-span
beams vibrations are cancelled out due to the second first-order ei-
genmode for L = 1.25 · Lc (see Fig. 14 middle). For more information on
vibration cancellation in single-span beams please refer to [16,20,21].
Complete cancellation of the vibration by interference in the first
eigenmode can be achieved for trains with Lc = L only with an even
number of spans, because otherwise the first and last spans do not have
“interference partners”. In a three-span beam a resonance peak can
therefore be seen at the first resonance speed (Figs. 17 and 18).
The first resonance speed has a significantly smaller amplitude than
that for a single-span beam, because the vibration energy of the addi-
tional span is transferred to the entire bridge and thus dissipated more
easily. In contrast, the vibrations of the second eigenmode for trains
with Lc = L in the first and third spans cancel each other out, as these
spans act as their respective interference partners (Fig. 18). In three-
span beams, the strongest dynamic response is caused by the third first-
order resonance speed. As mentioned previously, this resonance speed
is 1.87 times higher than the first resonance speeds of the single-span
beam. Using these observations, the most unfavourable resonance
speeds can be shifted outside the investigated range of speeds by in-
creasing the number of spans.
The most important thing that can be learned from the previous
observations is that longer continuous beams exhibit dynamically stable
behaviour under significantly higher crossing speeds than single-span
Fig. 17. Maximum bending moments in the second field of a three-span beam
with L = 25 m and double T cross section.
S. Schneider, S. Marx
Fig. 18. Maximum bending moments in the third field of a three-span beam
with L = 25 m and double T cross section.
beams or multi-span bridges with simply supported spans of the same
dimensions. The dynamic reactions are considerably smaller for the
lower resonance speeds, and the high critical first-order resonance
speeds can be shifted outside the range of operating speeds. For the
design, however, these positive effects have to be weighed against other
influencing parameters such as rail stresses and construction tech-
nology.
3.4. Haunched continuous beams with uniform span lengths
Haunched continuous beams offer more options for the design of
bridges. Because of the increased beam stiffness at the supports the
beams can be more slender in the rest of the span, offering aesthetic
advantages [1]. With haunches at the support the height of the super-
structure can also be varied according to architectural requirements in
bridges with spans of different lengths. It is further easier to fulfil
specific requirements regarding structure gauges, such as a higher
vertical clearance over the road than over the pavements.
The dynamic behaviour is influenced strongly by the haunch ar-
rangement. The presence of a haunch does not influence the first ei-
genfrequency very much compared to a continuous beam with constant
height, since the curvature of the respective eigenmode at the haunch is
very small (Tables 3 and 4). The first eigenfrequency increases notice-
ably only if longer haunches are present. The increase depends on the
haunch height and does not exceed approximately 20%. Activating the
bending stiffness of the columns by providing rigid connections with
the beam can also have a positive impact on these eigenfrequencies.
Eigenfrequencies whose eigenmodes exhibit large curvatures in the
vicinity of the columns increase significantly even with relatively short
haunches.
Fig. 20 shows the influence of the haunch height on the bending
moment in the first span of a two-span beam with a span length of
L = 20 m and a cross section consisting of a slab-with-cantilever. For
Table 3
Eigenfrequencies of investigated two-span beam with L = 20 m and slab with cantilevers cross section.
a – 0.25
b 1.0 1.5
1. eigenfrequency [Hz] 2.93 2.96
1. order
Increase – 1.1%
2. eigenfrequency [Hz] 4.58 5.36
1. order
Increase – 17.0%
403
Engineering Structures 174 (2018) 396–406
the investigation the beam height parameter b was varied between 1.5
and 2.5. The length parameter a, however, was kept constant at 0.25
(Fig. 19).
Under dynamic loading the span moments are reduced due to the
presence of haunches, just as they are under static loading. The re-
duction, however, is dependent on the excited vibration mode. The
presence of a haunch has only a negligible effect on the magnitude of
the span moments and the location of the resonance peaks due to a train
crossing at the first first-order resonance speed (Fig. 20). However, the
internal forces are reduced due to the second first-order eigenmode and
the second first-order resonance speeds increase so that they are shifted
outside the range of operating speeds.
As expected, the negative moments over the supports increase if
there are haunches (Fig. 21), but only for eigenmodes exhibiting cur-
vatures at the supports. Hence, the first first-order eigenmode does not
have any impact on the dynamic increase of the negative support mo-
ment. In the second first-order eigenmode the support moments can
become larger than the span moments of the single-span beam (de-
pending on the height and length of the haunch); the resonance peaks,
however, appear at significantly higher speeds. The results in Figs. 20
and 21 show that by using a haunch with a large height (b = 2.5) the
resonance peaks of the second first-order resonance speeds can be
shifted almost entirely outside the range of operating speeds.
The influence of the haunch length was investigated by examining
the bending moments in the third span of a three-span beam. The height
parameter b was kept constant at 1.5 and the length parameter a was
varied between 0.25 and 0.5 (Fig. 22). A length parameter a of 0.5 is
generally not used in practice and is included here merely as a
boundary value. From the results it can clearly be seen that the ad-
vantage of increasing the haunch length is the increase in the second
and third first-order eigenfrequencies. Only with very long haunches
(a = 0.5) does the first eigenfrequency also increase slightly. Therefore,
adding haunches is not a suitable solution for reducing the internal
forces resulting from a train crossing at the first resonance speed.
The question of whether adding haunches to a beam has a positive
or negative impact on the dynamic stability cannot be answered in a
general manner. Even though a beam with haunches can resist larger
internal forces, adding haunches should only be considered if the
second or higher first-order eigenmode is excited by a crossing train.
These eigenmodes exhibit curvatures at the supports, which means that
the internal forces in that area are increased; however, they only occur
at significantly higher speeds compared to beams without haunches.
Furthermore, the haunched parts of the beam are also capable of re-
sisting larger internal forces. Adding haunches to reduce the structural
reactions due to the vibrations of the first first-order eigenmode does
not work, as this will lead neither to a significant decrease of the in-
ternal forces, acceleration or deflections nor to a considerable increase
in the first first-order resonance speed.
3.5. Continuous beam with unequal span lengths
In the previous sections it was shown that continuous beams exhibit
more dynamically stable behaviour than multi-span simply supported
0.5
2.0 2.5 1.5 2.0 2.5
2.98 2.99 3.15 3.28 3.36
1.6% 2.0% 7.4% 11.7% 14.5%
5.95 6.38 5.59 6.56 7.49
29.8% 39.3% 22.1% 43.2% 63.6%
S. Schneider, S. Marx Engineering Structures 174 (2018) 396–406

Table 4
Eigenfrequencies of investigated three-span beam with L = 20 m and slab with cantilevers cross section.
a – 0.25 0.5

b 1.0 1.5 2.0 2.5 1.5 2.0 2.5

1. eigenfrequency [Hz] 2.93 2.97 3.00 3.01 3.24 3.45 3.61

1. order
Increase – 1.5% 2.2% 2.7% 10.4% 17.7% 23.2%

2. eigenfrequency [Hz] 3.76 4.20 4.54 4.81 4.46 5.14 5.83

1. order
Increase – 11.7% 21.0% 28.1% 18.7% 36.8% 55.1%

3. eigenfrequency [Hz] 5.49 6.74 7.67 8.32 7.24 8.88 10.37

1. order
Increase – 22.8% 39.7% 51.6% 32.0% 61.8% 89.1%

Fig. 19. Haunched two-span and three-span beam.

Fig. 20. Maximum bending moments of a two span beam in field 1 due to HSLM-A01 to A10 and L = 20 m (left: without haunches; middle: a = 0.25 and b = 1.5;
right: a = 0.25 and b = 2.5).

Fig. 21. Maximum bending moments of a two span beam at the support due to HSLM-A01 to A10 and L = 20 m (left: without haunches; middle: a = 0.25 and
b = 1.5; right: a = 0.25 and b = 2.5).

beams. If the local conditions allow it, it makes sense to reduce the end
spans of continuous beams under static uniformly distributed loads by a
factor of 0.8. This results in nearly uniform internal force maxima in
side and midspan under quasi-static loads. The effect of the shortening
of the end spans on the eigenfrequencies and the internal forces of a
continuous beam is investigated by examining two beams with three
and four spans, span lengths of L = 25 m and shortened end spans with
a length of 0.8 · L (Fig. 23).
One advantage of shortening the end spans is the resulting increase
in the superstructure eigenfrequencies. However, this increase is
smaller or larger depending on the eigenfrequency; for example, the
second first-order eigenfrequency of the three-span beam and the
second and third first-order eigenfrequency of the four-span beam are
increased more significantly than the other eigenfrequencies (Table 5).

404
S. Schneider, S. Marx Engineering Structures 174 (2018) 396–406

Fig. 22. Maximum bending moments of a three span beam in field 3 due to HSLM-A01 to A10 and L = 20 m (top: without haunches; middle: a = 0.25 and b = 1.5;
bottom: a = 0.5 and b = 1.5).

Fig. 23. Continuous beam with 0.8-times shortened

side spans.

The comparison of two three-span beams, one with uniform span
length and one with end spans shortened by a factor of 0.8 (Figs. 24 and
25), shows that the shortening of the end spans leads to an increase in
the internal forces in the first span due to the second first-order ei-
genmode, but also that the resonance speeds are increased con-
siderably. The internal forces of the first eigenmode are decreased
significantly, and those of the third first-order eigenmode are shifted
outside the range of operating speeds. The bending moments in span 2
and at the supports is impacted positively by the shortening of the end
spans because of the resulting increase in the resonance speeds. The
large increase in the second eigenfrequency is insignificant for the in-
ternal span, as this span experiences only minimal dynamic increases
due to the second first-order eigenmode. The reverse is true for the
support moment, which experiences a large increase due to the second
first-order eigenmode. However, the support moment is affected posi-
tively by an increase in the second first-order eigenfrequency. It can
hence be concluded that a shortening of the end spans of the in-
vestigated three-span beam also offers advantages for dynamic train
crossings.
4. Conclusion
The main aim in the design of railway bridges for high-speed traffic
is the optimisation of the dynamic stability and robustness, while
creating a slender and aesthetic structure. The system geometry and
stiffness have to be designed for the extreme dynamic forces caused by
the different European high-speed trains so as to avoid resonance
during train crossings.
Single-span bridges do not offer much room for optimisation, as
they experience the maximum possible structural reactions and do not
have any neighbouring spans for interactions. The only effective para-
meter for optimisation of a single-span beam is the beam height.
However, to achieve the necessary stiffness for the high design speeds
the beam height is generally large, which makes the structure look
massive and chunky. If load models HSLM-A01 to HSLM-A10 need to be
considered, the tool presented in [16] can be used to determine the
minimum required eigenfrequencies, obviating the need for a dynamic
analysis. This tool is particularly useful for the preliminary design, but
it can only be used for single-span beams. The results of the presented
parameter study also show that the deflection criterion presented in
[10] for the cross-section design of single-span beams for crossing
speeds of 350 km/h yields reliable results only for span lengths
L ≥ 40 m.
Statically indeterminate systems such as continuous beams have
much greater system stiffness and neighbouring spans strongly interact
during a train crossing. Therefore, bridges exceeding a length of 30 m
experience lower structural reactions. The vibrations caused by a train
crossing at the first first-order resonance speed, in particular, are re-
duced significantly, which means that a train can cross a continuous
beam at considerably higher speeds than a multi-span simply supported
beam of the same dimensions.
Even small changes in span length configuration or stiffness

Table 5
Eigenfrequencies of investigated three and four-span beam with L = 25 m and side spans with 0.8 · L.
25–25–25 20–25–20 25–25–25–25 20–25–25–20

1. eigenfrequency [Hz] 2.44 3.09 2.44 2.82

1. order
Increase – 26.6% – 15.6%

2. eigenfrequency [Hz] 3.13 4.71 2.85 3.81

1. order
Increase – 50.5% – 33.7%

3. eigenfrequency [Hz] 4.57 5.80 3.81 5.12

1. order
Increase – 26.9% – 34.4%

4. eigenfrequency [Hz] – 4.93 5.74

1. order
Increase – – 16.4%

405
S. Schneider, S. Marx
Fig. 24. Maximum bending moments of a three span beam with L = 20 m and constant span length due to HSLM-A01 to –A10 (left: field 1; middle: field 2; right:
support).
Fig. 25. Maximum bending moments of a three span beam with L = 20 m and 0.8-times shortened side spans due to HSLM-A01 to –A10 (left: field 1; middle: field 2;
right: support).
distribution can lead to distinct increases or decreases in the dynamic
response of the system and can therefore not be predicted accurately
without carrying out a dynamic train simulation. Even so, in this con-
tribution the qualitative influence of haunch configuration and of a
shortening of the end spans was investigated, as these parameters can
be used to optimise the dynamic behaviour of the superstructure. By
adding haunches those eigenfrequencies whose eigenmodes exhibit
curvatures at the supports can be increased. This also leads to a de-
crease in the span moments following the excitation of these eigen-
modes, whereas the corresponding support moments are increased
considerably. If the end spans are shortened, the eigenfrequencies of all
eigenmodes are increased, which results in an increase in all resonance
speeds, some of which are thereby shifted outside the range of in-
vestigated speeds. By using the described structural modifications and
taking advantage of interference, specific types of resonance can be
decreased during the design phase.
The results obtained from the presented parameter studies can
hence be used to adjust the dynamic stability of a structure subjected to
high-speed trains.
References
[1] Marx S, Seidl G. Integral railway bridges in Germany. Struct Eng Int
2011;21(3):332–40.
[2] Yan B, Dai Gong-Lian HuN. Recent development of design and construction of short
span high-speed railway bridges in China. Eng Struct 2015;100:707–17.
[3] Mao L, Lu Y. Critical speed and resonance criteria of railway bridge response to
moving trains. J Bridge Eng 2013;18(2):131–41.
[4] Frýba L. Vibration of solids and structures under moving loads. Springer Science &
Business Media; 2013.
[5] Richtlinie 804 – Eisenbahnbrücken (und sonstige Ingenieurbauwerke) planen,
bauen, instand halten. [Guideline 804 – Railway bridges (and other engineering
structures) design, construct, maintain]. DB Netz AG; 2013 [in German].
406
Engineering Structures 174 (2018) 396–406
[6] Mähr T, Fink J. Comparison and assessment of various load models to calculate train
crossings of bridges at high speed. Stahlbau 2009;78(10):733–41. [in German].
[7] Dehestani M, Mofid M, Vafai A. Investigation of critical influential speed for moving
mass problems on beams. Appl Math Model 2009;33(10):3885–95.
[8] Yau JD, Yang YB, Kuo SR. Impact response of high speed rail bridges and riding
comfort of rail cars. Eng Struct 1999;21(9):836–44.
[9] Frýba L. Dynamic behavior of bridges due to high speed trains. In: Calcada R,
Delgado R, Matos AC, editors. Bridges for high-speed railways: revised papers from
the workshop, Porto, Portugal; 3–4 June 2004.
[10] DIN EN 1990 – Eurocode: basis of structural design; German version EN 1990:2002
+ A1:2005/AC:2010; December 2010.
[11] ERRI D214/RP9 – Rail bridges for speeds > 200 km/h – final report. European Rail
Research Institute (ERRI), Utrecht; December 1999.
[12] Mellier C. Optimal design of bridges for high-speed trains. Master of science thesis.
Stockholm: Royal Institute of Technology (KTH); 2010.
[13] Kwark JW, Choi ES, Kim YJ, Kim BS, Kim SI. Dynamic behavior of two-span con-
tinuous concrete bridges under moving high-speed train. Comput Struct
2004;82(4–5):463–74.
[14] Frýba L. A rough assessment of railway bridges for high speed trains. Eng Struct
2001;23(5):548–56.
[15] DIN EN 1991-2:2010-12 – Eurocode 1: actions on structures – Part 2: Traffic loads
on bridges; German version EN 1991-2:2003 + AC:2010; December 2010.
[16] Spengler M, Graubner C-A. Proposal of a response spectrum for application to dy-
namics of railway bridges. Bauingenieur 2009;84(1):1–9. [in German].
[17] ERRI D 214/RP 6. Rail bridges for speeds > 200 km/h – calculation for bridges with
simply supported beams during the passage of a train December Utrecht: European
Rail Research Institute (ERRI); 1999.
[18] Richtlinie für die dynamische Berechnung von Eisenbahnbrücken. [Guideline for
the dynamic calculation of railway bridges]. ÖBB-Infrastruktur AG; February 2011
[in German].
[19] Yau JD. Resonance of continuous bridges due to high speed trains. J Mar Sci
Technol 2001;9(1):14–20.
[20] Sudheesh Kumar CP, Sujatha C, Shankar K. Vibration of simply supported beams
under a single moving load: a detailed study of cancellation phenomenon. Int J
Mech Sci 2015;99:40–7.
[21] Museros P, Moliner E, Martínez-Rodrigo MD. Free vibrations of simply-supported
beam bridges under moving loads: maximum resonance, cancellation and resonant
vertical acceleration. J Sound Vib 2013;332(2):326–45.