Bearing Capacity of Roads, Railways and Airfields – Loizos et al.

(Eds)
© 2017 Taylor & Francis Group, London, ISBN 978-1-138-29595-7

Track defects and the dynamic loads due to suspended (sprung)

masses of railway vehicles

K. Giannakos
Civil Engineer PhD, Fellow of ASCE, Member of TRB AR050 & 060 Committees, Piraeus, Greece

ABSTRACT:  The motion of a railway vehicle on a railway track is a forced oscillation with a forcing
excitation induced by the rail running table—which has the form of a “signal”–, expressed by a random,
non-periodic function. The motion is described by the second order differential equation of motion spe-
cialized for the system “vehicle-track”. The railway track, is simulated—with the observer situated on the
wheel—as an elastic means with springs and dashpots. The railway vehicle has the Suspended Masses
(SM) and the Non-Suspended Masses (NSM). The track defects/faults are the random excitation for the
rolling wheels of the vehicle. In the case of the Suspended Masses of the vehicles, the forces resulting from
the excitation, due to the track defects, when the wavelength of defects is short, are not large and have
small effect on the rolling of the wheel. In the case of the Non-Suspended Masses the forces resulting from
the excitation, due to the track defects of short wavelength, are large and have grate effect on the rolling of
the wheel. The present paper depicts that when the wavelength of defects is long, the Suspended Masses
perform a predominant role. The solution of the differential equation is presented for the Suspended
Masses of the vehicle. A sensitivity is also performed and a comparison between the theoretical results for
the vertical accelerations (directly proportional to the acting loads) due to the Suspended Masses vs the
measured ones, in a track under operation.

1  INTRODUCTION (Fig. 1 left) and the forces exerted on the vehicle’s

The motion of a railway vehicle on the rail running
table/surface or the motion of a road vehicle on the
road, the response of the structures to earthquakes,
etc., is a forced oscillation with a forcing excitation
(force), and damping expressed by a random, non-
periodic function. The motion is described by equa-
tions and, in railway engineering, it is illustrated
through the simplified form of a spring-mass-
damper system as depicted in Fig. 1, with a railway
vehicle running on a track with longitudinal defects
Figure 1.  A spring-mass-damper-system: (left) a vehicle
on a railway track with longitudinal defects and (right)
the forces exerted on the “car-body”.
“car-body” (Fig. 1 right). In this simplified model,
with the wheel rolling over a surface with defects,
but undeflected itself, the acting forces are:
a. the weight of the vehicle m⋅g;
b. the dynamic component of the Load Pdyn;
c. the motive force P(t) which is equal to the differ-
ence between the tractive force of the locomo-
tive minus the friction and it is positive in the
case of increase of speed (accelerated motion)
or negative in the case of braking (decrease of
speed) since it is equal to zero if the motive force
is equal to friction;
d. the reaction R1 provided by the system “vehicle-
track” equal to a spring constant or coefficient
of elasticity ρi (given in kN/mm) multiplied by
the subsidence u(x) of the center of gravity of
the vehicle;
e. the reaction R2 provided by the system “vehicle-
track” equal to a damping constant ci multiplied
by the first derivative of the subsidence u(x) of
the center of gravity of the vehicle.
In practice, the circulation of a railway vehicle
on a railway track differs significantly from this
simplified model, since the support/railway track
is not undeflected and the railway vehicle has the
Suspended Masses (SM) and the Non-Suspended

1911

Figure 2.  Model of a Vehicle running the Rail Running
Table: the NSM and the SM [as well as the car-body], the
primary and secondary suspensions are depicted. Fur-
thermore the NSM of both the Track and the Vehicle are
depicted in two ellipses. The Rail Running Table has its
defects n and the track during the passage of the vehicle
presents a deflection y.
Masses (NSM) as described below. The behavior
of the NSM (the masses located under the primary
suspension of the vehicle) is approached with the
track simulated—with the observer situated on the
wheel—as an elastic mean with damping as illus-
trated in the simplified model in Fig.  2. For the
simulation of the SM of the vehicles (that is of
the masses located over the primary suspension of
the vehicle), the forces resulting from the forcing
excitation imposed by the rail running table—in the
most of the cases—have small effect on the rolling of
the wheel, but for defects of very long wavelength,
the oscillations of the SM become predominant
since the oscillations of the NSM decrease. The SM
are examined in the present paper.
2  SOLVING THE SECOND ORDER
DIFFERENTIAL EQUATION FOR A
RAILWAY TRACK UNDER A RUNNING
VEHICLE
In Fig. 2 a simplified model of the system “Rail-
way Track—Railway Vehicle” is depicted; in prac-
tice it is an ensemble of springs and dashpots. The
railway track is represented by the resultants of
springs and dashpots, as described in Giannakos
2016a; 2013. The railway vehicle has two levels
of suspensions the primary suspension between
the axles and the bogie’s frame and the secondary
suspension between the car-body and the bogie’s
frame. Under the primary suspension there are
the NSM (axle, wheels and a fraction of the semi-
suspended motive electromotor in the locomo-
tives) which act without any damping directly on
1912
the track panel. On the contrary the SM that are
cited above the primary suspension of the vehicle,
act through a combination of springs and dampers
on the track. A part of the track mass is also added
to the NSM, which participates in their motion
(Giannakos 2012; 2010).
The rail running table has the shape of a wave
that is not completely “rectilinear”, that is, it does
not form a perfectly straight line but contains
faults/defects, varying from a few fractions of a
millimeter to a few millimeters, and imposes forced
oscillation on the railway vehicles that move on it;
the faults/defects are represented by the ordinate n
in Fig. 2. Moreover during the rolling of the wheel,
a deflection y of the rail running table appears
(Giannakos 2004), since the support (track) is not
undeflected.
Forced oscillation is caused by the irregulari-
ties of the rail running table (like an input random
signal) –which are represented by n–, in a gravita-
tional field with acceleration g. Moreover, a section
of the mass of the railway track participates in the
motion of the NSM of the vehicle.
If the random excitation (track irregularities) is
given, it is difficult to derive the response, unless
the system is linear and invariable. In this case the
input signal can be defined by its spectral density
and from this we can calculate the spectral density
of the response. The theoretical results confirm
and explain the experimental verifications (Alias
1984, p.39, 71).
The second order differential equation of
motion in the case of a railway track on which a
vehicle is running, due to the interaction between
the vehicle’s axle and the track-panel becomes
(Giannakos 2004):
d 2y dy
(mNSM + mTRACK ) ⋅ dt 2 + Γ⋅
dt
+ hTRACK ⋅ y
d 2n (1)
= − mNSM ⋅ + ( mNSM + mSM ) ⋅ g
dt 2
where: mNSM = the NSM of the vehicle in tonnes-
mass, mTRACK the mass of the track that partici-
pates in the motion of the NSM, mSM = the SM in
tonnes-mass, Γ = damping coefficient of the track,
hTRACK = total dynamic stiffness coefficient of the
track, n the fault ordinate of the rail running table,
g the acceleration of gravity, y the total deflection
of the track,
1 ρ
hTRACK = ⋅ 4 E ⋅ J ⋅ total (2)
2⋅ 2 
with E, J the modulus of elasticity and the moment
of inertia of the rail (steel) and ℓ the distance
among the sleepers and
 1 1 1 1 1 1 d 2u d ( u − n)
= + + + + (3) ⇒ mNSM + Γ⋅
ρtotal ρrail ρ pad ρsleeper ρballast ρsubgrade dt 2 dt (8b)
+ hTRACK ⋅ (u − n) = 0
ρi the static stiffness coefficients of the constitutive
layers of the track, the quasi spring constants of where, u is the trajectory of the wheel over the ver-
the layers and ρtotal the resultant total static coef- tical fault (of ordinate n) in the longitudinal profile
ficient of the track. of the rail.
We hypothecate a cosine form defect on the rail If we apply the Fourier transform to Eqn (7a):
running table of the form:
Γ ⋅ (iω ) h
 V ⋅t  (iω )2 ⋅ Z (ω ) + ⋅ Z (ω ) + TRACK ⋅
η = a ⋅ cos ωt = a ⋅ cos  2π ⋅ mNSM mNSM
  (4)
λ  Z (ω ) = − (iω ) ⋅ N (ω ) ⇒
2

Z (ω )
⇒ H (ω ) = , H (ω )
2
where: η the ordinate of the defect along the track
(abscissa x), V the speed of the vehicle, t the time N (ω )
and λ the wavelength of the defect; so in the case mNSM
2
⋅ω 4
= (9)
(mNSM ⋅ ω 2 − hTRACK ) + Γ 2 ⋅ ω 2
2
of a defect of wavelength λ:

2π 2π
T= ⇒ ωt = t
ω T H(ω) is a complex transfer function, called fre-
2πVt mNSM + mSM quency response function, that makes it possible to
⇒ ωt = ⋅g (5)
λ hTRACK pass from the fault n to the subsidence Z. If we
apply the Fourier transform to equation (8a):
If we set a new variable z, as:
(iω )2 ⋅U (ω ) + Γ ⋅ (iω ) ⋅ Z (ω ) + hTRACK ⋅ (iω )0 ⋅
mSM + mNSM Z (ω ) = 0 ⇒
y=z+ ⋅
hTRACK U (ω ) hTRACK
2
+ Γ2 ⋅ω 2 (10)
G (ω ) = , G (ω ) =
2

g⇒
dy dz
=
d 2 y d 2z
⇒ 2 = 2 (6) Z (ω ) mNSM
2
⋅ω 4
dt dt dt dt
G(ω) is a complex transfer function, the fre-
where [[g⋅(mSM+mNSM)]/hTRACK] represents the sub- quency response function, that makes it possible to
sidence due to the static loads only, and z is the pass from Z to Z + n.
random subsidence due to the dynamic loads. Eqn If we name U the Fourier transform of u, N the
(1) becomes: Fourier transform of n, p = 2πiν = iω the variable
of frequency and Δ^Q the Fourier transform of ΔQ
d 2z dz d 2n and apply the Fourier transform at equation (8b):
mNSM + Γ ⋅ + hTRACK ⋅ z = − mNSM ⋅ 2 ⇒ (7a)
dt 2
dt dt
d 2u du dn
 d 2z d 2n  dz mNSM + Γ⋅ + hTRACK ⋅ u = Γ ⋅ + hTRACK ⋅ n ⇒
mNSM  2 + 2  + Γ ⋅ + hTRACK ⋅ z = 0 (7b) dt 2 dt dt
 dt dt  dt ( mNSM ⋅ p + Γ ⋅ p + hTRACK ) ⋅U = (Γ ⋅ p + hTRACK ) ⋅ N ⇒
2

Γ ⋅ p + hTRACK
U (ω ) = ⋅ N (ω ) (11a)
Since, in this case, we are examining the dynamic mNSM ⋅ p 2 + Γ ⋅ p + hTRACK
loads only (derived from the actions of the SM and 


B (ω )
the NSM), in order to approach their effect, we
could narrow the study of equation (7b), by chang-
ing again with a new variable u:
where:
d u d n d z
2 2 2
Γ 2 ⋅ ω 2 + hTRACK
2
u = n+z⇒ = + B (ω ) =
2
(11b)
dt 2 dt 2 dt 2 (mNSM ⋅ ω 2 − h)
2
+ Γ2 ⋅ω 2
Equation (7b) becomes:
B(ω) is a complex transfer function, the fre-
quency response function, that makes it possible to
d u 2
dz
mNSM + Γ ⋅ + hTRACK ⋅ z = 0 ⇒ (8a) pass from the fault n to the u = n+Z. Practically it is
dt 2 dt verified also by the following equation [11c]:

1913
 The excitation (rail irregularities), in reality,
B (ω ) = H (ω ) ⋅ G (ω )
2 2 2
is random and neither periodic nor analytically
hTRACK
2
+ Γ2 ⋅ω 2 defined, like the Eqn. (4). It can be defined by its
= (11c)
(mNSM ⋅ ω 2 − hTRACK ) + Γ 2 ⋅ ω 2
2
autocorrelation function in space and its spectral
density (Alias 1984, p.58; Gent & Janin, p.700;
passing from n to Z through H(ω) and afterwards Papoulis 1977). If f(x) is a signal with determined
from Z to n+Z through G(ω). This is a formula total energy and F(ν) its Fourier transform, from
that characterizes the transfer function between Parseval’s modulus theorem (Roddier 1971), the
the wheel trajectory and the fault in the longitu- total energy is (Wylie & Barrett 1995):
dinal level and enables, thereafter, the calculation +∞ +∞
of the transfer function between the dynamic load
f ( x ) ⋅ dx = ∫ F (ν )
2 2

and the track defect (fault). −∞

∫ −∞
⋅ dν (15a)
The transfer function B(ω) allows us to calcu-
late the effect of a spectrum of short wavelength
where, F(ν) = Α(ν) ⋅ eiϕ(ν) and power spectral density:
sinusoidal faults, like the undulatory wear. If we
replace ω/ωn = ρ, where ωn = the circular eigenfre-
S (ω ) = F (ν ) = A2 (ν )
2
quency (or natural cyclic frequency) of the oscilla- (16)
tion, and:
Wylie & Barrett (1995) solve Eqn (15a) as:
hTRACK 2πV Γ ω
ω n2 = ,ω = , 2ζω n = ,β = (11d)
mNSM λ mNSM ωn +∞
1
+∞

∫ f (t ) ∫ F (ω )
2 2
⋅ dt = ⋅ dω (15b)
2π
where ζ = damping coefficient. From Eqn (11b): −∞ −∞

1 + 4ζ 2 ⋅ β 2 The square of the modulus F(ω) is called the

B (ω ) = Bn ( β ) =
2 2
(11e) energy spectrum of the signal because F2(ω)⋅∆(ω)
(1 − β ) 2 2
+ 4ζ ⋅ β 2 2
represents the amount of energy in any ∆Ω segment
of the frequency spectrum, and the integral of F2(ω)
The transfer function C(ω) of the second over (- ⊥, + ∞) gives the total energy of the signal.
derivative of (Z+n) in relation to time: [d2(Z+n)/ An input signal—like the running rail table—creates
dt2], that is the acceleration γ, will be equal to through the vehicle an output signal: the wheel tra-
ω⋅Β(ω): jectory. The output spectral density and the input
spectral density of the excitation are related through
p 2 ⋅U (ω ) = p 2 B (ω ) ⋅ p 2 ⋅ N (ω ) ⇒ equation (Clough & Penzien 1993; Giannakos 2004):
⇒ U (ω ) = p 2 B (ω ) ⋅ N (ω ) (12a)
 
SOUTPUT (ω ) = H (iω ) ⋅ SINPUT (ω )
2
C (ω ) (16a)

that is: In order to relate the temporal spectrum with the

spectrum in space we use the following equation:
ω 2
C (ω ) = ω 2 ⋅ B (ω ) = ω n2 ⋅ ⋅ B (ω ) ⇒
ω n2 2πVt 2π
(12b) ω ⋅t = ⇒ω = ⋅V ⇒ ω = Ω ⋅V (16b)
⇒ C (ω ) = ω n2 ⋅ β 2 ⋅ B (ω ) = ω 2 ⋅ B (ω ) λ λ
where λ is the wavelength of the defect; the circular
The increase of the vertical load on the track frequency in space Ω is the wave number k of the
due to the NSM, according to the principle equation of oscillation, and (Roddier 1971):
force  = mass × acceleration, is given by:
∞ ∞
d 2u d 2 (n + Z ) 1
∆Q = mNSM ⋅
dt 2
= mNSM ⋅
dt 2
(13) ∫ S (Ω) ⋅ d Ω = ∫ s (ω ) ⋅ dω ⇒ F  f (ax ) =
0 0
a
ν  ω  1
⋅ fˆ   ⇒⇒ S (ω ) = S   = ⋅ S ( Ω) (17a)
If we apply the Fourier transform to Eqn. (13):  a V  V

∆ˆ Q = mNSM ⋅ p 2 ⋅U (ω ) = mNSM ⋅ p 2 ⋅ fˆZ + n (ω ) ⇒ (14a) where is the symbol for the application of the
Fourier transform of f and f^ the function after
∆ˆ Q = mNSM ⋅ C (ω ) ⋅ N (ω ) = mNSM ⋅ p ⋅ B (ω ) ⋅ N (ω )
2

the transform. This is a property of the Fourier

∆ˆ Q= mNSM ⋅ β 2 ⋅ ω n2 ⋅ B (ω ) ⋅ N (ω ) (14b) transform.

1914
 Eqn (17a) applied in the case that the power From the above equations and the analytical
spectrum of the vertical defects along the track form of the spectrum of the longitudinal defects/
(for the NSM) in the space domain is S(Ω) then the faults of the track we could effectively calculate
power spectrum of the excitation of the wheel in the variance (mean square value) of the dynamic
the time domain will result after a replacement of component of the Loads on the track panel due
Ω by the ω/V. to the NSM. All the results of measurements on
track in the French railways network show that the
1 ω spectrum of defects in the longitudinal level has
sυ (ω ) = S  (17b)
V V  the form [SNCF (1981), Prud’Homme (1970)]:
A
The Variance or mean square value σ2(x) of the Sn ( Ω ) = , (22)
function is given by Papoulis (1991): ( B + Ω) 3

+∞
1 implying that the mean square value or variance of
σ 2 (x) = ⋅ ∫ S (ω ) ⋅ dω = x 2 (18)
2π −∞ the defects is given by the following Eqn (23):
+∞ +∞
where σ(x) the standard deviation of the function. 1 A A 1
σ 2 (z) =
π ∫0 ( B + Ω)3
⋅ ⋅ dΩ = ⋅ ∫ ⋅
From Eqn (18) we derive: π 0 ( x )3
+∞
+∞  A 1 
1
σ ( n) = ∫ Sn (ω ) ⋅ dω , σ 2 ( z ) =
2 dx = −  2
=
π 0  ( B + Ω)  0
2π
+∞ +∞ +∞
1 1 A  1  A  1 
( ) ( ) S (ω ) ⋅ dω
π ∫0 z π ∫0 ∆Q
S ω ⋅ d ω , σ 2
∆Q = (19) =− ⋅ 2 =− 0 − B 2 
2π  B + 2 BΩ + Ω2  0 2π
1 A
⇒ σ 2 (z) = ⋅
where n is the random variable of the defect 2π B 2 (23)
(input), z the subsidence of the wheel (output) and
∆Q the dynamic component of the Load that is If we examine only the much more severe case,
added to the Static Load of the wheel due to the for the case of the NSM, of the defects of short
NSM (output also). wavelength, consequently large Ω—like the undu-
From these equations and the analytic form of latory wear—then we can omit the term B, and
the spectrum of the defects/faults, we can calculate using Eqn. (16b):
the mean square value of the dynamic component of
the Load due to the NSM that is added to the rel- A A A ⋅V 3
evant dynamic component of the SM and the total Sn ( Ω ) = = = (24)
Ω 3 1 ω3
dynamic component of the load is added to the Static ⋅ ω 3
V3
Load of the wheel. From the power spectral density
and the variance functions and their definitions: The term B characterizes the defects with long
wavelengths, for which the maintenance of track is
S∆Q (ω ) = Sn (ω ) ⋅ B (ω )
2
(20a) effective, and when we examine this kind of defects
term B should be taken into account. SM should
∆Q = mNSM ⋅ ∆γ ⇒ σ 2 ( ∆Q ) = mNSM ⋅ σ 2 (γ ) ⇒ (20b) be examined for long wavelength defects.
σ 2 ( ∆Q ) During the research program performed by the
⇒ σ 2 (γ ) = (20c) Greek State Railways/OSE and the French State
mNSM Railways/SNCF for the cracking of sleepers/ties, the
measured values for the line “Les Aubrais—Vierzon”
and using the Eqns. (20) and (11–12b): in France were used; the parameters values, taken
into account, are: B = 0.36, A = 2.1⋅10–6 and S(Ω) is
1 1
+∞ calculated in m3 and σ(z) = 1.57 mm. The eigenfre-
σ 2 (γ ) = ⋅ σ 2 ( ∆Q ) = ∫ B (ω ) ⋅ Sn (ω ) ⋅ dω ⇒
2
⋅ quency of the NSM of the vehicles is approximately
mNSM mNSM
2
⋅π
+∞
0
30–40  Hz and even for speeds of 300  km/h there
mNSM
2

⇒ σ 2 (γ ) = ⋅ β 4 ⋅ ω n4 ⋅ B (ω ) ⋅ Sn (ω ) ⋅ dω ⇒
∫
2
are wavelengths less than 3 m (Prud’Homme 1970).
mNSM
2
⋅π 0 From equations (17) and (24):
1 + 4ζ 2 ⋅ β 2
+∞
1
⇒ σ 2 (γ ) = ∫
⋅ β 4 ⋅ ω n4 ⋅ ⋅ Sn (ω ) ⋅ dω
(1 − β ) 1 1 A ⋅V 3 A ⋅V 2
Sn (ω ) = ⋅ S ( Ω) = ⋅
2
π 2
+ 4ζ 2 ⋅ β 2
0
= 3 3 (25)
(21) V V ω3 ωn ⋅ β

1915
3  EXCITATION BY DEFECTS OF LONG
WAVELENGTHS
An analysis of the behavior of the NSM was pre-
sented in Giannakos (2016a; 2013), where diagrams
are cited for the relation between the wavelength of the
defect and the subsidence z(x) and the vertical accel-
eration z″(x) of the NSM (consequently, the dynamic
component of the load since force = mass × accelera-
tion). It is valid (Giannakos supra):
 ( mNSM + mTRACK ) z ′′ (t ) 
2
 2π  1
 ⋅  == − 
 m NSM α   n ⋅ Tn  g
 
 1
⋅ cos (2πµ ) − ( n) ⋅ cos (2 nπµ )   [% g ]
2
 2 ⋅ 1 − ( n)       
2
steady − state transient − part
where: n = ωn/ω1 = Τ1/Τn, Tn = 2π/ωn = the period of
the free oscillation of the wheel circulating on the
rail, T1 = 2π/ω1 = the necessary time for the wheel
to run over a defect of wavelength λ, ω1  =  λ/V,
V  =  speed, µ  =  a percentage of the wavelength
λ[0, 0.1, 0.2,...., 0.8, 0.9, 1], α = the amplitude of
the defects’ depth [derived from track recordings].
In the case of the Suspended Masses of the vehi-
cles, the forces resulting from the excitation, due
to the short wavelength track defects, are not large
and have small effect on the rolling of the wheel
(Giannakos 2016a; 2013). It is obvious that the
From Fig. 3 we observe that for wavelengths infer-
ring cyclic-frequencies higher than two times the
eigenfrequency of the wheel, the vertical acceleration
of the NSM becomes very small, almost negligible
(consequently the dynamic component of the load),
compared to the results presented in Giannakos
[2016a; 2013], for smaller values of ωn/ω1 (0.5 till 2).
In the cases of the long wavelength defects, the
influence of the SM should be examined.
If we assume that the defects of the two rails—
in a cross-section of a track—are quite the same
at the same time and presenting the same phase,
⋅ ⋅ or if we examine the trajectory of one wheel, then
the conclusion that will be derived can be used
 for more complicated cases of rolling of vehicles,
motion of car-bodies etc. Furthermore, we con-
 sider the simplified model of Fig. 1 with one-floor
mass-spring-damper system rolling on a rail’s run-
(26) ning surface.
In order to calculate the power spectrum density
of the excitation sE(ω) from the excitation spec-
trum of the wheel sυ(ω), we apply the Eqn.  16a
with the Eqns. 17, Eqn. 22 (the parameter B is not
omitted for the case of SM) and Eqns. 10 (in order
to pass from the defect n to n + Z):
sE (ω ) = B (ω ) ⋅ sυ (ω ) ⇒
2
ω2
1 + 4ζ 2 ⋅ 2
ωn AV 2 (27a)
⇒ sE (ω ) = ⋅
ω ( BV + ω )
2 3
 ω  2 2

1 − 2  + 4ζ ⋅ 2
2
main effort should be addressed to the measure-  ω n  ωn
ments of defects, in practice, and their reliability,
due to the defects’ wavelengths and the instrumen- where: sE(ω) is the power spectrum density of the
tation (measuring bases, techniques, vehicles). excitation, ωn is always the eigenfrequency of
In Fig. 3 the results of a sensitivity analysis are the NSM, sυ(ω) the spectrum of the excitation of
depicted, specifically for the case of long wave- the wheel due to the track defects/faults, |B(ω)| the
length defects (ωn/ω1  =  2, 2.5, 3, 3.5; horizontal modulus of the transfer function of the motion of
axis represents the values of µ) in relation to the the wheel and ζ the damping coefficient of the track:
vertical acceleration z″(t) (as a percentage of the
gravitational acceleration g) multiplied by a (con- 1 Γ
stant) factor [(mNSM + mTRACK)/(mNSM ⋅ a)]: ζ= ⋅ (SNCF 1981, p.28).
2 mNSM ⋅ hTRACK

The sυ(ω) can be calculated from Eqn.  22. We

have to take into account Eqn. (17a) also:

1 1 A 1 A
sυ (ω ) = ⋅ S (Ω) = ⋅ = ⋅ ⇒
V V ( B + Ω)3 V  ω
3

 B + 
V

AV 2
⇒ sυ (ω ) = (27b)
( BV + ω )3
Figure 3.  Vertical acceleration of the NSM in relation The value of sυ(ω) derived from Eqn. 27b is used
to the wavelength [for large wavelengths]. in Eqn. 27a.

1916
 From the Eqns. 11 and the Eqns. 12, 13 and 14, 30–40 Hz (see relevantly Prud’Homme 1970). It is
with the analysis cited above, we keep that, C(ω) is approximately (for ω = ωn→β = 1):
the transfer function of the second derivative of
(Z + n) in relation to time: [d2(Z + n)/dt2], that is the AV 2  1 
acceleration γ and it is equal to ω⋅B(ω). sγ (ω ) = ω n4 ⋅ ⋅
3 
+ 1 ⋅
In the case of the SM, |C(ω)| is the modulus of ( BV + ω n)
 4ζ 2

2
the transfer function of the accelerations of the ω 
car-body. Consequently, for the spectrum of the 1 + 4ζ ′ 2 ⋅  n 
 ω n′  (28c)
accelerations of the car-body we will use Eqn. 15a ⋅ 2
substituting the parameters ζ and ωn of the track  ω   2
 ωn 
2

1 −    + 4ζ ′ 2 ⋅  
n
with the relevant parameters ζ′ (damping coeffi-
  ω n′    ω n′ 
cient) and ω′n (eigenfrequency) of the car-body.
 
For the railway passenger-vehicles the eigen- Term − B
frequencies ω′n of the car-body are in the area of
1 Hz, since with the development of high-speeds it In this case, of ω = ωn ≈ 30–40 Hz, we have the
could arrive 10 Hz. For the damping coefficient of case of the NSM and consequently the case of
the car-body of the railway vehicles two character- relatively short wavelengths; thus the term B can
istic values of ζ′ could be used with reliability: 0.15 be omitted (see Giannakos [2016a; 2013]). From
and 0.20 (see relevantly Prud’Homme 1970; Alias Eqn. 26b:
1984). From Eqn. 26a (Giannakos 2016b):
AV 2 AV 2 AV 2
sυ (ω ) = = = 3 3 (27d)
sγ (ω ) = C (ω ) ⋅ s E (ω ) = ω ⋅ ω ωn ⋅ β
2 3 3
4
 
AV 2 1 + 4ζ 2 ⋅ β 2  + ω 
 BV
⋅ ⋅ ≈0

( BV + ω ) (1 − β 2 )2 + 4ζ 2 ⋅ β 2
3

(27c)
1 + 4ζ ′ 2 ⋅ β ′ 2 which replaces the relevant term (BVAV+ ω ) , in Eqn. 28b, 2

⋅ c above.
n

(1 − β ′ 2 )2 + 4ζ ′ 2 ⋅ β ′ 2 Now, we try to simplify the Term-B:

There are two (2) maximums in the curve of 2

sγ(ω): The first maximum corresponds—in prac-  ωn 
1 + 4ζ ′ 2 ⋅ 
tice—to the eigenfrequency of the car-body of the  ω n′ 
vehicle. It is approximately (for ω = ω′n→β′ = 1): Term − B = 2
  ωn  2   ωn 
2

 1 −  ′   + 4ζ ′ ⋅  ′ 
2

sγ (ω ) = ω n′ 4 ⋅
AV 2
⋅
1 + 4ζ 2 ⋅ β 2
⋅   ωn    ωn 
( BV + ω ′ ) (
1 − β ) + 4ζ ⋅ β
 
n
3


2 2 2 2
=
1
+
(28a),   ωn   2 2 2
Term − A
 ωn 
1 + 4ζ ′ ⋅ 1
2 2
AV 2
 1  1 −
  ′  + 4ζ ′ 2
⋅  ω ′ 
⋅ ≈ ω n′ 4 ⋅ ⋅ +1
(1 − 1 )
2 2
+ 4ζ ′ 2 ⋅ 12 ( BV + ω ′ ) 3
 4ζ ′ 2    ωn   n
n 
→0
ω n2 ω n2
where Term-A→1 (for ω′n = 1 Hz, ωn = 35 Hz): +4ζ ′ ω n′
2 2
≈ 4ζ ′ 2ω n′ 2
(ω ′n
2
− ω n2 ) + 4ζ ′ 2ω n2ω n′ 2
2
ω n4 (27e)
ω n′2
1 + 4ζ 2 ⋅ Eqn (27b) is transformed (using Eqn. 27d, 27e):
ω n2
Term − A = 2
=
 ω n′ 2  ω n′ 2
1 − + 4ζ 2
⋅ AV 2  1  1
 ω 2  ω2 sγ (ω ) = ω n4 ⋅ ⋅  2 + 1 ⋅ 4ζ ′ 2ω n′ 2 2 ⇒
(28b) ωn ⋅ β
  4ζ  ωn
n n 3
12 (27f)
1 + 4ζ 2 ⋅ =1

= 352 →1  1  ω ′2
12 
2
12 ⇒ sγ (ω ) = AV ⋅  2 + 1 ⋅ 4ζ ′ 2 n
2
  4ζ  ωn
1 − 2  + 4ζ ⋅ 2
2
35 35
Which is the second maximum of sγ(ω), corre-
The second maximum corresponds—in prac- sponding—in practice—to the eigenfrequency of the
tice—to the eigenfrequency of the system “vehicle- system “vehicle-track”, as it was mentioned above.
track”, that is the eigenfrequency of the system of The variance of the accelerations of the car-body
the NSM and the track, which is approximately of the railway vehicles is given by (Eqns. 18, 19):

1917
 ∞ presented by Alias (1984; 1987) and Prud’Homme
1
σ (γ ) = ⋅ s (ω ) ⋅ dω
π ∫0 γ
2
(29)
which converges for ω infinite and can be calculated
in a computer. Consequently, the variance of the
part of the dynamic component of the load due to
the SM of the vehicle is given by (see Eqn. 20b):
∆QSM = mSM ⋅ ∆γ ⇒ σ 2 ( ∆QSM ) = mSM
2
⋅ σ 2 (γ ) (30)
It is obvious that the variance sγ(ω) and the stand-
ard deviation σ(γ) are functions of the inverse of the
cyclic eigenfrequency ωn of the track, which (Eqns.
11d, 2) is a function of the static (ρTRACK) and/or
dynamic (hTRACK) stiffness coefficients of the track.
4  MEASURED VALUES OF THE
STANDARD DEVIATION OF THE
VERTICAL ACCELERATIONS DUE TO
THE SM
In real conditions, according to measurements of
the French State Railways (SNCF), the standard
deviation σ(γ) of the vertical accelerations [as a
percentage of the gravitational acceleration g] due
to the Suspended Masses of the vehicles, in rela-
tion to the running speed is depicted in Fig. 4, as
Figure 4.  Standard deviation σ(γ) of the measured ver-
tical accelerations, due to the Suspended Masses of the
railway vehicles, in relation to the running speed as pub-
lished [Alias (1984; 1987) and Prud’Homme (1970)] and
Eqn. (30) for NL = 0.5 and NL = 1.0. The figure has been
designed by the author of the present paper.
1918
(1970). The measurements were performed in
tracks under operation in the French network
(SNCF) and obviously, the measured values of
σ(γ) have been influenced by the variation of the
real values of ρsubgrade and ρtrack, as it existed along
the measured tracks (consequently, the values of
ρ were not constant), since the variability of the
track stiffness e.g. due to imperfect sleeper support
and inhomogeneities of the track structure is an
inherent property. Fig. 4 is the final product of the
measurements performed along the tracks of the
French network, and it is given in the publications
above, as of general validity.
In Fig. 4, six curves are presented for eigenfre-
quencies of the car-body 1 to 6  Hz. More ana-
lytically, the coupled system car-body-bogie-axles
(a two-floor system) presents an eigen-frequency,
that of the axle on the track, approximately
30–40  Hz, in an attenuated form, correspond-
ing to the track defects which will become much
more important when the speed in consideration
will provoke a frequency very close to the coupled
frequency of the car-body with the axles. This
coupled low frequency of the car-body-bogie-
axles, which mainly affects the car-body and it
is specifically interesting to us in this analysis, is
approximately 1 Hz for the passenger wagons and
higher for the freight wagons (see relevantly Alias
1987, p. 47). This implies that the curve for 1 Hz
eigenfrequency represents the passenger wagons
and the higher frequencies the freight wagons,
running at lower speeds.
According to the analysis cited above—about
the coupled system car-body-bogie-axles—it is
clear that finally in this coupled motion (of 1 Hz
frequency) all the parts of the vehicle participate,
so we can approach the equations using the total
Qwheel instead of mSM.
Consequently, an approximation could be used
for the calculation of the variance of this part of
the dynamic component of the load (Giannakos
2004; SNCF 1981):
V − 40
σ ( ∆QSM ) = ⋅ N L ⋅ Qwheel (31)
1000
where: Qwheel is the static wheel load, V is the oper-
ational speed, and the coefficient NL is the mean
standard deviation of the longitudinal level condi-
tion of the track, on a 300 m length approximately,
for both rails is the mean standard deviation of
the longitudinal level condition of the track, on a
300  m length fluctuating between 0.7–1.5  mm or
more (see SNCF 1981; Janin 1982, p.335–336); for
the Greek network NL is estimated to fluctuate—
mainly—between 1 and 1,5 (Giannakos 2004).
 In more details, NL, the average of the brutal
signal on a basis of approximately 300  m for the
vertical and horizontal defects of the two rails, is
the convolution:
+∞  x − x0 
1 papazisi.gr, Athens.
N L ( x0 ) = ∫ ηl ( x )e 
 300 
dx (31)
300 −∞
where ηl(x) is the value of the primary signal; in
practice a weighted average index which “crashes”
less the isolated defects than one classic average
and simulates roughly the “memory” of the vehicle
(see relevantly Janin 1982).
In Fig. 5 the curves of the eigenfrequencies of
the vehicles as measured by the SNCF are depicted
and, also, the Eqn (30) for two values of NL: 0.5
and 1. The approximation is close enough, but a
further research and sensitivity analysis should be
performed in the future.
5  CONCLUDING REMARKS
In this paper the second order differential equa-
tion of motion of the Suspended (Sprung) Masses
(SM) of a railway vehicle rolling on a railway track
with defects/faults is investigated and its solution
is approached through a sensitivity analysis of the
speed (from 50 till 300 km/h). The investigation is
performed through the Fourier transform and the
solution is verified from findings from a research
program performed by the Greek railways in col-
laboration with the French state railways (SNCF)
–with the author as coordinator—and a multi-year
further research performed by the author as visit-
ing professor in the University.
The formula, given by Eqn (30), as it was pro-
posed after this research program, gives results in a
satisfactory approximation to the measured values
in France. In any case a further research and sensi-
tivity analysis in the future is also necessary.
REFERENCES
Alias, J. 1984. La Voie Ferree—Techniques de Construc-
tion et Entretien, deuxieme edition, Eyrolles, Paris.
1919
Alias, J. 1987. Le Rail, Eyrolles, Paris.
Clough, R.W. & Penzien, J. 1993. Dynamics of Struc-
tures, second edition, McGraw-Hill Intl., Singapore.
Gent, I. & Janin, G. 1969. La Qualite de la Voie Ferree,
RGCF, Dec., Dunod, reprint, 1–29.
Giannakos, K. 2004. Actions on the Railway Track, www.
Giannakos, K. 2010. Theoretical calculation of the track-
mass in the motion of unsprung masses in relation to
track dynamic stiffness and damping, Intl J. Pavem.
Eng. (IJPE), Special Rail Issue “High-Speed Railway
Infrastructure: Recent Developments and Perform-
ance”, Vol.11, Is. 4, p. 319–330.
Giannakos, K. 2012. Influence of the track’s damping on
the track mass participating in the motion of the Non
Suspended Masses of railway vehicles—theoretical
calculation and comparison to measurements, volume
published in honour of (fs) professor G. Giannopou-
los, Aristotle Univ. of Thessaloniki.
Giannakos, K. 2013. Track Defects and the Dynamic
Loads due to Non-Suspended Masses of Railway
Vehicles, NAUN Intl J. Mechanics, Vol.7, Is.3, p.
180–191.
Giannakos, K. 2016a. Modeling the Influence of Short
Wavelength Defects in a Railway Track on the
Dynamic Behavior of the Non-Suspended Masses, J.
of Mech. Syst. and Signal Proc., Elsevier, Vol. 68–69,
p.68–83.
Giannakos, K. 2016b. Second Order Differential Equa-
tion of Motion in Railways: the Variance of the
Dynamic Component of Actions due to the Sprung
Masses of the Vehicles, Intl J. of Theoretical &
Applied Mechanics, http://www.iaras.org/iaras/jour-
nals/ijtam, p.30–37.
Janin, G. 1982. La Maintenance de la Geometrie de la
Voie, RGCF, Juin, p.331–346.
Papoulis, A. 1977. Signal Analysis, McGraw-Hill Inc.,
USA.
Papoulis, A. 1991. Probability, Random Variables and
Stochastic Processes, third edition, McGraw-Hill,
Singapore.
Prud’Homme, A. 1970. La Voie, RGCF, Janvier, reprint.
Roddier, F. 1971. Distributions et Transformation de
Fourier, Edisciense, Paris.
SNCF/Direction de l’ Equipement. 1981. Mecanique de
la Voie, reprint, Paris.
Wylie, C.R. & Barrett, L.C. 1995. Advanced Engineering
Mathematics, sixth edition, McGraw-Hill, Inc., USA.