2 Proc IMechE Part F: J Rail and Rapid Transit 0(0)

services during their operation. Using in-service vehicles pseudo-track geometry has been investigated by
to determine these parameters presents a potential Weston et al.9 Real et al.10 use frequency domain
saving for railway infrastructure managers. techniques to estimate track profile. A mixed acceler-
Track longitudinal profile can be considered as a ation data filtering approach is used by Lee et al.11 in
representation of the vertical profile of a track made the measured acceleration for stable displacement esti-
up of consecutive measurements of longitudinal level mation and waveband classification of the
as defined in EN13848.2 Rails are represented indi- irregularities.
vidually, i.e. a separate longitudinal profile exists for In this paper, the cross-entropy (CE) combinatorial
each rail. Rail longitudinal profile is comprised of a optimisation method, as described by de Boer et al.,12
combination of macro changes in track elevation in is adapted to determine track longitudinal profiles.
the longitudinal direction (e.g. track grades, vertical Harris et al.13 use the CE combinatorial optimisation
curves, etc.) and local rail irregularities. Rail irregula- technique to characterise vehicle model parameters
rities are geometrical deviations from the ideal rail and road surface profiles using measured vehicle
longitudinal profile. acceleration responses. However, to the authors’
Railway track longitudinal profile is an important knowledge, the CE method has not previously been
indicator of serviceability condition. It is desirable to used to determine longitudinal profiles for railway
have perfectly level track profiles so that dynamic track. In this paper, dynamic interaction models are
responses of the vehicle are minimised, thereby used in favour of moving load models so that dynamic
increasing passenger comfort, reducing wear on vehi- effects are not overestimated and more realistic accel-
cle components and reducing power consumption. eration histories can be generated.14
A reduction in vehicle dynamics also reduces the vehi- This paper reports the results of the numerical simu-
cle load on the track. Therefore, keeping a good lations carried out to test the concept of using CE opti-
vertical longitudinal profile helps maintain overall misation to determine track longitudinal profile. The
track condition through a reduction in vehicle next section explains the numerical models used for the
dynamic effects.3 However, it is inevitable that track simulations. Following this, the CE method is
irregularities will occur for a number of reasons described and the optimisation processes are discussed.
including rail head manufacturing defects, wear, This is followed by the results of the optimisation that
impact from wheel flats, track settlement and poor validates the methodology for a range of profiles.
maintenance.4 It is the passage of the vehicle across
the irregularities on the track profile that excites it and
Model description
invokes a dynamic response.
In numerical studies, track irregularities can be Numerical simulation of the vehicle–track interaction
generated using stationary random processes (VTI) is used in assessing the application of CE opti-
described by power spectral density (PSD) functions. misation to the estimation of track longitudinal pro-
PSD is used in several countries (including the US, file for given vehicle dynamic response. VTI
Germany, China and France) to classify track quality calculations are carried out in Matlab15 to generate
according to its irregularity spectrum.5 Definitions of the dynamic response of a vehicle travelling longitu-
PSD functions vary significantly due to their empirical dinally along a track. The vehicle models, track
nature and differences in measuring instruments and model, and PSD method for generating track irregu-
evaluation methods used in their formulation.6 Perrin larities are described in this section along with a brief
et al.7 propose an alternative method for generating description of the vehicle and track model coupling.
more realistic track geometry irregularities based on Two levels of model complexity are used. The
statistical properties of measured data. method is initially tested using Model A, a simple
Recently, the possibility of using inertial methods quarter-car vehicle model crossing a series of rigid
to estimate track profiles using acceleration data mea- rail profiles (i.e. deflection in the rail is not permitted).
sured on a vehicle has gained considerable interest. To demonstrate the capabilities of the algorithm, an
Weston et al.8 provide a review of the state-of-the- inverted bell-shaped irregularity added to an other-
art track monitoring using in-service vehicles. Most wise level track profile. Three rail profiles featuring
unattended geometry measurement systems (UGMS) randomly generated irregularities with varying
fixed to in-service trains require both inertial and opti- degrees of roughness are also generated and used in
cal sensors to return the parameters required for full the numerical tests. In this model, irregularities repre-
assessment of track geometry. A drawback of using sent variations from perfectly level track caused by
UGMS is that the in-service vehicles hosting the sys- deviation from design levels during construction, vari-
tems are required to stop at stations more regularly ation in track stiffness and from permanent deform-
than dedicated measuring trains. This has the effect of ation of track over time.
reducing the speed of the UGMS vehicles in certain Following this, Model B, a relatively complex vehicle
areas, compromising accuracy of measurement. and three-layer track model, is used to test the method.
The use of accelerometers and rate gyroscopes on When coupled together, these models return more real-
the bogie level of a railway vehicle to estimate a istic vehicle accelerations, partly as a result of permitting