186

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 1, JANUARY 2015

Modeling and Control of a Full-Scale Roller-Rig

for the Analysis of Railway Braking Under
Degraded Adhesion Conditions
Benedetto Allotta, Member, IEEE, Roberto Conti, Enrico Meli, and Alessandro Ridolfi
Abstract Currently, braking on board subsystems such as
wheel slide protection (WSP) devices almost totally control the
longitudinal train dynamics. In particular, the vehicle safety
highly depends on the study and the development of these
systems, especially at high speeds and under degraded adhesion
conditions. Usually, to save time and to avoid expensive
on-track tests, the performances of braking subsystems are
tested on full-scale roller-rigs. Nevertheless, the analysis of the
subsystem behavior under degraded adhesion conditions is still
limited to a few applications on roller-rigs because large slidings
among the rollers and wheelsets produce severe wear of the
rolling surfaces. This circumstance is not acceptable due to
the effects on the maintenance costs (the rollers have to be
turned or substituted), on the system dynamical stability and
on the safety. In this paper, the modeling and control of an
innovative hardware in the loop (HIL) architecture to test
braking on board subsystems on full-scale roller-rigs is described.
The new approach permits to reproduce on the roller-rig a
generic wheel-rail adhesion pattern and, in particular, degraded
adhesion conditions. The presented strategy is also followed by
the innovative full-scale roller-rig of the Railway Research and
Approval Center of Firenze-Osmannoro (Italy); the new rollerrig has been built by Trenitalia and is owned by SIMPRO. At this
initial phase of the research activity, to effectively validate the
proposed approach, a complete model of the HIL system has been
developed. The complete numerical model is based on the real
characteristics of the components provided by Trenitalia. The
results coming from the simulation model have been compared
with the experimental data provided by Trenitalia and relative
to on-track tests performed in Velim, Czech Republic, with a
UIC-Z1 coach equipped with a fully working WSP system. The
preliminary validation performed with the HIL model highlights
the good performance of the HIL strategy in reproducing on
the roller-rig, the complex interaction between the degraded
adhesion conditions and railway vehicle dynamics during the
braking maneuver.
Index Terms Degraded adhesion conditions, hardware in the
loop (HIL), railway braking, roller-rigs.

I. I NTRODUCTION

OWADAYS, the longitudinal train dynamics is almost

totally controlled by on board subsystems, such as wheel
slide protection (WSP) braking devices. The study and the

Manuscript received January 4, 2014; accepted April 13, 2014. Date of

publication May 20, 2014; date of current version December 15, 2014.
Manuscript received in final form April 22, 2014. Recommended by Associate
Editor F. Caccavale.
The authors are with the Industrial Engineering Department, Florence
University, Florence 50139, Italy (e-mail: benedetto.allotta@unifi.it;
roberto.conti@unifi.it; enrico.meli@unifi.it; a.ridolfi@unifi.it).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TCST.2014.2320672

development of these systems are fundamental for the vehicle

safety, especially at high speeds and under degraded adhesion
conditions. On-track tests are currently quite expensive in
terms of infrastructure and vehicle management. Therefore,
to reduce these costs, full-scale roller-rigs are traditionally
employed to investigate the performances of braking subsystems [1][4]. However, in the presence of degraded adhesion,
the use of roller-rigs is still limited to few applications (see,
for example, full-scale roller-rigs for the study of the wear [5],
hardware in the loop (HIL) systems for WSP tests [8] and
full-scale roller-rigs for locomotive tests [8]) because the high
slidings between the rollers and wheelsets produce wear of the
rolling surfaces. This circumstance is very dangerous and not
acceptable: the flange wear can lead to the vehicle derailment
while the tread wear can produce hunting instability of the
vehicle [27]. Furthermore, the wheel flats may generate unsafe
vibrations of the vehicle on the roller-rig. Finally, the wear of
the rolling surfaces deeply affects the maintenance costs: the
rollers have to be frequently turned or substituted.
In this paper, the modeling and control of an innovative
HIL architecture to test braking on board subsystems on
full-scale roller-rigs is presented by the authors. The new
strategy permits to reproduce on the roller-rig a generic wheelrail adhesion pattern and, in particular, degraded adhesion
conditions (characterized by adhesion coefficient values equal
or less than 0.10 [31]). The control architecture, based on
a robust sliding mode approach, performs a simulation of
mechanical impedance by properly controlling the roller-rig
motors (all the motors are independently controlled). More
particularly, the roller motors are controlled to recreate, on
the wheelsets, the same angular velocities, applied torques, and
tangential efforts exchanged between the wheelsets and rails in
the reality and calculated by a reference virtual railway vehicle
model. The new control architecture allows the achievement
of this goal by only controlling the roller motors and without
having sliding (and consequently wear) between the wheelsets
and rollers. In fact, since the real adhesion coefficient between
the rollers and wheelsets surfaces is far higher than the
simulated one (greater than 0.40), negligible sliding occurs
and almost pure rolling conditions are always present between
them.
Some initial and partial results have been obtained in [9].
The major novelties of the presented work mainly concern the
two following aspects.
1) The Motor Controllers (Able to Reproduce on the RollerRig the Same Dynamical Behavior of the Virtual Train

1063-6536 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

ALLOTTA et al.: MODELING AND CONTROL OF A FULL-SCALE ROLLER-RIG

Fig. 1.

General architecture of the HIL system.

Model): At this step of the research activity, the controllers of the motors are completely independent and the
robust control strategy, based on a robust sliding mode
approach, is totally new if compared with the previous
steps of the research project.
2) The Adhesion Model (Able to Accurately Reproduce
the Real Adhesion Behavior in Virtual Train Model):
At this research step, new and more accurate degraded
adhesion models have been used, considering the third
layer present at the wheelrail interface and the adhesion
recovery due to the energy dissipated at the contact by
the friction forces that destroy the contaminants.
At this phase of the research activity, the described strategy
has been completely simulated in the MATLAB-Simulink
environment through an accurate model of the whole HIL
architecture. Each model component is modeled according to
the real characteristics provided by Trenitalia. Furthermore,
the authors do not analyze the differences between wheelrail contact and wheel-roller contact. Such differences may
affect the roller-rig system in many ways: kinematics, stability
(bifurcation diagrams), dynamics, wear, and so on. At this
research step, the authors mainly focus on the crucial problems
of the system control and of the control ability to reproduce different adhesion patterns on the roller-rig (especially,
degraded adhesion conditions) [6], [7].
The proposed approach has been preliminarily validated
through a comparison with the experimental data provided
by Trenitalia and relative to on-track tests performed on a
straight railway track (in Velim, Czech Republic) with a
UIC-Z1 coach equipped with a fully working WSP system
[32][34]. This initial validation carried out through the HIL
model highlights the good performances of the HIL strategy in
reproducing on the roller-rig the complex interaction between
the degraded adhesion conditions and railway vehicle dynamics during the braking maneuver.
II. G ENERAL A RCHITECTURE OF THE HIL S YSTEM
In this section, the architecture of the Firenze-Osmannoro
HIL system is briefly described. Fig. 1 schematically shows
the main parts of the architecture (both for the hardware

187

components and the software ones). The models used to

simulate all these parts will be better explained in Section III.
The architecture comprises four main elements as
follows.
1) The test-rig (hardware), composed of two main parts:
the UIC-Z1 railway vehicle (equipped with the WSP
system) [32], [33] and the Firenze-Osmannoro rollerrig [with the innovative actuation system developed
in collaboration with SICME and based on interior
permanent magnets (IPM) synchronous motor with high
performance characteristics] [35], [36]. The inputs of the
test-rig are the roller control torques while the outputs
are the longitudinal reaction forces measured on the
roller supports and the measured angular velocities of
the rollers.
2) The virtual railway vehicle model (software), representing the model used to simulate the vehicle behavior
on the rails under different adhesion conditions [8]
and designed for a real-time implementation. This 2-D
multibody model simulates the longitudinal dynamics
of the vehicle while an innovative 2-D adhesion model
[28], [29] permits an accurate reproduction of the real
behavior of the adhesion coefficient during braking
phases under degraded adhesion conditions. At this
research step, new and more accurate degraded adhesion
models have been used, considering the third layer
present at the wheelrail interface and the adhesion
recovery due to the energy dissipated at the contact by
the friction forces that destroy the contaminants. The
inputs are the estimated torques on the wheelsets and
the outputs are the simulated wheelset angular velocities
and the tangential contact forces on the wheelsets.
3) The controllers (software) reproduce on the roller-rig
the same dynamical behavior of the virtual train model
(through the roller control torques) in terms of wheelset
angular velocities, applied torques and, consequently,
tangential forces [8]. The new control architecture allows
the achievement of this goal by only controlling the
roller motors and without having sliding (and consequently wear) between wheelsets and rollers. In fact,
since the real adhesion coefficient between the rollers
and wheelsets surfaces is far higher than the simulated
one, negligible sliding occurs and almost pure rolling
conditions are always present between them. Due to the
HIL system nonlinearities, a sliding mode approach has
been adopted for the controllers [37], [38]. At this step
of the research activity, the controllers of the motors are
completely independent and the robust control strategy,
based on the robust sliding mode approach, is totally
new if compared with the previous steps of the research
project.
4) The torque estimators (software): the data measured by
the sensors installed on the roller-rig are only the roller
angular velocities and the longitudinal reaction forces
on the roller supports. No sensors will be placed on
the vehicle to speed up the setup process. Starting from
these quantities, this block estimates the torques applied
on the wheelsets.

188

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 1, JANUARY 2015

Fig. 2.
Interactions among the models of the various HIL architecture
components.

Fig. 3.

UIC-Z1 coach.

Fig. 4.

WSP device and its logical scheme.

III. M ODELING OF THE F IRENZE -O SMANNORO

HIL S YSTEM
In this section, the models of the HIL system presented
in the previous section (both hardware and software parts)
and of all the components of the HIL architecture will be
explained in detail. The main elements of the architecture are
the same of Fig. 1, but, in this case, the test-rig is completely
simulated both for the UIC-Z1 railway vehicle (3-D vehicle
model and WSP model) and for the roller-rig (roller-rig 3-D
model). An innovative 3-D contact model especially developed
by the authors for this kind of application is used [24], [26].
The mechanical and electrical characteristics of the vehicle
[32], [33] and of the roller-rig [35], [36] are directly provided
by Trenitalia and RFI.
The flow of the data among the model parts is shown in
Fig. 2.
A. Test-Rig Model
The inputs of the whole test-rig model are the eight
roller control torques u l , u r (left and right) evaluated by the
controllers to reproduce on the test-rig the same dynamical
behavior of the virtual railway model. The outputs are the eight
roller angular velocities rl , rr and the longitudinal reaction
l , T r measured on the roller supports. The test-rig
forces Tmis
mis
model is composed of four parts (Fig. 2).
1) Vehicle Model: The considered railway vehicle is the
UIC-Z1 coach (shown in Fig. 3); its geometrical and physical
characteristics are provided by Trenitalia [32].
The wagon consists of one carbody, two bogie frames, eight
axleboxes, and four wheelsets. The UIC-Z1 coach has a twostage suspension system: the primary suspension, including
springs, dampers, and axlebox bushings, connects the bogie
frames to the four axleboxes while the secondary suspension,
including springs, dampers, lateral bump-stops, anti-roll bar,
and traction rod, connects the carbody to the bogie frames. The
multibody vehicle model considers all the degree of freedom
(DOFs) of system bodies. Both the primary suspensions and
the secondary ones have been modeled through 3-D viscoelastic force elements able to describe all the main nonlinearities
of the system.

The inputs of the model are the four wheelset torques Cs

modulated by the on board WSP and the contact forces calculated by the contact model, while the outputs are the kinematic
wheelset variables transmitted to the contact model, the four
original torques C (without the on board WSP modulation)
and the four wheelset angular velocities w . These last two
outputs are not accessible by the HIL system.
2) WSP System Model: The WSP device installed on the
UIC-Z1 coach [28], [33] allows the control of the torques
applied to the wheelsets, to prevent macrosliding during the
braking phase.
In Fig. 4, the logical scheme and an image of the WSP
device are shown. The inputs are the braking torques C and
the wheelset velocities w , while the outputs are the modulated
braking torques Cs . The WSP system working principle can
be divided into three different tasks: the evaluation of the
reference vehicle velocity Vref and acceleration aref based on
the wheelset angular velocities w and accelerations w ; the
computation of the logical sliding state stateWSP (equal to 1

ALLOTTA et al.: MODELING AND CONTROL OF A FULL-SCALE ROLLER-RIG

189

TABLE I
M AIN C HARACTERISTICS OF THE ROLLER -R IG S YSTEM
AND OF THE

Fig. 5. Right side of roller-rig system with the synchronous motors and the
rollers placed in the semianechoic room of the Research and Approval Center
of Firenze-Osmannoro.

Fig. 6.

W HEELSETS

Main roller-rig sensors.

TABLE II
ROLLER -R IG S ENSOR C HARACTERISTICS

if sliding occurs and 0 otherwise)and the consequent torque

modulation, through a speed and an accelerometric criterion
and by means of a suitable logical table [28]; and the periodic
braking release to bring back the perceived adhesion coefficient to the original value (often used when degraded adhesion
conditions are very persistent and the WSP logic tends to drift).
3) Roller-Rig Model: The 3-D multibody model of the
roller-rig (Fig. 5 and Table I) consists of eight independent
rollers with a particular roller profile able to exactly reproduce
the UIC60 rail pattern with different laying angles p [35]. The
railway vehicle is axially constrained on the rollers using two
axial links (front and rear) modeled using 3-D force elements
with linear stiffness and damping. The inputs of the test-rig
model are the eight torques u l , u r evaluated by the controllers
and the contact forces calculated by the contact model; the
outputs are the roller angular velocities rl , rr , the longitudinal
l , T r measured on the roller supports and
reaction forces Tmis
mis
the kinematic wheelset variables transmitted to the contact
model.
The roller-rig actuation system consists of eight synchronous motors, especially designed and developed in cooperation
with SICME [36] for this kind of application. The HIL
architecture includes a direct-drive connection between the
roller and the electrical machine. The synchronous motors
have high efficiency associated with high torque density and
flux weakening capability. Furthermore, to reach the dynamical

and robustness performances required by the railway

full-scale roller-rig, the motors are designed with a multilayerrotor characterized by a high saliency ratio and IPM. The
IPM motors are controlled in real-time through vector control
techniques; more particularly, the vector control is a torquecontrolled drive system in which the controller follows a
desired torque [10][15], [36].
The main sensors installed on the roller-rig (Fig. 6) are the
absolute encoders and the three-axial load cells on the roller
supports. These sensors are employed both in the torque
estimators and in the controllers and measure, respectively,
the roller angular velocities rl , rr and the longitudinal
l , T r on the roller supports. The sensor
reaction forces Tmis
mis
characteristics are reported in Table II [36].
4) Wheel-Roller Contact Model: The 3-D contact model
l/r
l/r
evaluates the contact forces Nc , Tc for all the eight
wheel-roller pairs starting from the kinematic variables of
l/r
the wheelsets and of the rollers: their positions Gw , Gr ,

190

Fig. 7.

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 1, JANUARY 2015

Fig. 8.

Virtual railway vehicle model.

Fig. 9.

Adhesion model.

Wheel-roller contact model.

l/r

l/r

orientations w , r , velocities vw , vr , and angular

l/r
velocities w , r (Fig. 7).
The contact model comprises three different steps. First, all
l/r
the contact points Pc of each wheel-roller pair are detected.
Some innovative procedures have been recently developed in
[24], [26], and [42]; the new algorithms are based on the
reduction of the algebraic contact problem dimension through
exact analytical techniques [25]. Second, the normal contact
problem is solved through the Hertz theory [16] to evaluate
l/r
the normal contact forces Nc . Finally, the solution of the tangential contact problem is performed using the KalkerPolach
l/r
theory [16][20] to compute the tangential contact forces Tc .
The contact model guarantees high accuracy and numerical
efficiency; this way, the model can be implemented directly
online inside the whole test-rig model.
B. Virtual Railway Vehicle Model
The virtual railway vehicle model simulates the dynamical
behavior of the railway vehicle during a braking phase under
degraded adhesion conditions. At this research step, new and
more accurate degraded adhesion models have been used,
considering the third layer present at the wheelrail interface
and the adhesion recovery due to the energy dissipated at
the contact by the friction forces that destroy the contaminants. The model, designed for a real-time implementation, is
composed of two parts: the 2-D vehicle model and the 2-D
s
adhesion model. The inputs are the four estimated torques C
to be applied to the wheelsets while the outputs are the four
simulated tangential contact forces Tsim and the four simulated
wheel angular velocities ws .
1) Virtual Vehicle Model: The 2-D vehicle model of the
considered railway vehicle (UIC-Z1 coach) is a simplified 2-D
multibody model of the longitudinal train dynamics (only
3 DOFs for each body are considered) [32]. The model (Fig. 8)

consists of a carbody, two bogies, and four wheelsets, held

by the primary and secondary suspensions. Starting from
s , the model evaluates the kinematic
the estimated torques C
variables of the four wheelsets v ws , ws and the four normal
contact forces Ncs to be passed to the adhesion model and
receives the four tangential contact forces Tsim .
2) Adhesion Model: The adhesion model has been especially developed to describe degraded adhesion conditions
[21][23], [28], [43] and calculates, for all the four wheelsetrail pair, the tangential contact forces Tsim starting from the
wheelset kinematic variables v ws , ws and the normal contact
forces Ncs (Fig. 9).
The main phenomena characterizing the degraded adhesion are the large sliding occurring at the contact interface
and, therefore, the high energy dissipation. Such dissipation
causes a cleaning effect on the contact surfaces and finally
an adhesion recovery due to the removal of the external
contaminants. When the specific dissipated energy Wsp is low,
the cleaning effect is almost absent, the contaminant level
h does not change and the adhesion coefficient f is equal
to its original value f d in degraded adhesion conditions f d .
As the energy Wsp increases, the cleaning effect increases too,
the contaminant level h becomes thinner and the adhesion
coefficient f raises. In the end, for large values of Wsp , all the
contaminant is removed (h is null) and the adhesion coefficient
f reaches its maximum value fr ; the adhesion recovery due
to the removal of external contaminants is now completed.
At the same time, if the energy dissipation begins to decrease,
due, for example, to a lower sliding, the reverse process occurs
(Fig. 9).
Since the contaminant level h and its characteristics are
usually totally unknown, it is useful trying to experimentally

ALLOTTA et al.: MODELING AND CONTROL OF A FULL-SCALE ROLLER-RIG

191

correlate the adhesion coefficient f directly with the specific

dissipated energy Wsp

strategy, based on a robust sliding mode approach, is totally

new if compared with the previous steps of the research
project.
The inputs of the controller are the simulated tangential
forces Tsim , the simulated wheelset angular velocities ws , the
estimated wheel angular velocities 
w , the estimated motor
l/r
torques C S , and the roller angular velocities r . The outputs
l/r
are the eight roller control torques u .
The controller layout consists of eight independent controllers (one for each roller) and makes use of a sliding mode
strategy based on the dynamical equations of the roller-rig;
this way, it is possible to reduce the disturbance effects due
to the system nonlinearities and the parameter uncertainties
[37], [38]. The total control torques u l/r are defined as

Wsp = Tsim e = f Ncs e f =

Tsim
Ncs

(1)

where the creepage e is defined as

v ws rw ws
s
=
(2)
e=
v ws
v ws
where s is the sliding and rw is the wheel radius. This way, the
specific dissipated energy Wsp can also be interpreted as the
energy dissipated at the contact for unit of distance travelled
by the railway vehicle.
To reproduce the qualitative trend previously described and
to allow the adhesion coefficient to vary between the extreme
values fd and fr , the following expression for f is proposed:
f = [1 (Wsp )] f d + (Wsp ) fr

(3)

where (Wsp ) is an unknown transition function between the

degraded adhesion and adhesion recovery while the adhesion
levels f d , fr can be evaluated according to [17][20] as
a function of e, Ncs and the track friction coefficients d ,
r (corresponding to degraded adhesion and full adhesion
recovery, respectively). The function (Wsp ) has to be positive
and monotonous increasing; moreover, the following boundary conditions are supposed to be verified: (0) = 0 and
(+) = 1.
This way, the authors suppose that the transition between
degraded adhesion and adhesion recovery only depends
on Wsp . This hypothesis is obviously only an approximation
but, as it will be clearer in the next sections, it well describes
the adhesion behavior. Initially, to catch the physical essence
of the problem without introducing a large number of unmanageable and unmeasurable parameters, the authors have chosen
the following simple expression for (Wsp ):
(Wsp ) = 1 e Wsp

(4)

where is now the only unknown parameter to be tuned on the

base of the experimental data (in this case, = 1.9 104 m/J)
[28], [34], [39]. Finally, the tangential forces Tsim = f Ncs
can be evaluated starting from the adhesion coefficient f
calculated by solving the nonlinear equation (3) [see (1) for
the definition of Wsp ] [40].
C. Controllers
The controllers have to reproduce on the roller-rig the
dynamical behavior of the virtual railway vehicle under
degraded adhesion conditions in terms of angular velocities w , applied torques Cs and, consequently, tangential
l/r
contact forces Tc . The new control architecture allows the
achievement of this goal by only controlling the roller motors
and without having sliding (and consequently wear) between
the wheelsets and rollers. In fact, since the real adhesion coefficient between the rollers and wheelsets surfaces is far higher
than the simulated one, negligible sliding occurs and almost
pure rolling conditions are always present between them.
At this step of the research activity, the controllers of the
motors are completely independent and the robust control

l/r

l/r

l/r

u l/r = u cont + u disc + u diff

(5)

l/r

where the continuous control part u cont is built starting from

the approximated 1-D models of wheelset and rollers and by
supposing the slidings between the contact surfaces negligible
(on the roller-rig the adhesion conditions are good, with a
friction coefficient roll equal to 0.3)
Cs = Jw w Tcl rw Tcr rw
u l = Jr rl Tcl rw u r = Jr rr Tcr rw
rw
rw
rl = w rl = w
rr
rr
rw
rw
r
r
r = w r = w
rr
rr

(6)

in which rr , rw are the roller and wheelset radii and Jr , Jw

are their inertias. By removing Tcl and Tcr in 6, the following
relation is obtained:
 2

rw  l
u + u r = Jtot w Jtot = Jw + 2 rrwr
Jr (7)
Cs
rr
where Jtot is the total inertia of the rollers and the wheelset
reduced to the wheelset rotation axis. Subsequently, the desired
wheelset dynamics is considered
C s = Jw ws + Tsim rw

(8)

together with the sliding surface S = ws w = 0 and

its time derivative S = ws w . If the torque estimation
C s  Cs is accurate enough, the sliding condition S = 0 can
be obtained, starting from (7) and (8), by taking


Jtot
Jtot
rr
1
Tsim rw
u=
Cs +
rw
Js
Js
u r
u
l
u = .
(9)
u =
2
2
l/r

On the other hand, u disc is the discontinuous control part

related to the rejection of the disturbances
u ldisc = u rdisc = k(ws 
w ).
l/r

(10)

The discontinuous controls u disc are characterized by the

gain k and the function  shown in Fig. 10 (the dead zone
amplitude and the slope are control parameters to be
tuned).

192

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 1, JANUARY 2015

TABLE III
M AIN C HARACTERISTICS OF THE B RAKING T EST

Fig. 10.

Discontinuous and auxiliary control characteristics.

l/r

Finally, u diff is an auxiliary control part aimed at synchronizing the roller angular velocities rl , rr




(11)
u ldiff = kd rl rr u rdiff = kd rl rr .
The function  is shown in Fig. 10, while the parameters kd ,
d and d have to be tuned. The controller performances will
be evaluated using the angular velocity error e = ws w
and the torque estimation error ec = C s Cs . Limited values
of the previous errors e , ec assure a good estimation of the
l/r
tangential contact forces Tc .
D. Torque Estimators
The estimators aim at evaluating the wheelset angular velocs starting
ities 
w and the torques applied to the wheelset C
l/r
from the roller angular velocities r and the longitudinal
l/r
reaction forces Tmis on the roller supports. Since the slidings
between the wheelset and rollers can be neglected, the following estimations approximately hold:
rr rl + rr
l + r
l/r
l/r
w = rrwr r 2 r Tc = Tmis . (12)
rw
2
Of course, the derivative operation d/dt has to be robust,
l/r
considering the numerical noise affecting r . At this point,
to estimate the motor torque applied to the wheelset, the
estimator employees the simplified dynamical model of the
wheelset
w =

C s = Jw w Tcl rw Tcr rw .

(13)

It is worth noting that, in this kind of applications, the

estimators have to be necessarily simple because they are
thought for a real-time implementation and, at the same time,
the physical characteristics of the railway vehicle on the rollerrig are generally unknown.
IV. E XPERIMENTAL DATA
The HIL model performance have been validated using
the comparison with the experimental data, provided by
Trenitalia S. p. A. [34] and coming from on-track braking
tests carried out in Velim (Czech Republic) with the coach
UIC-Z1 [32]. The considered vehicle is equipped with a fully
working WSP system [33]. These experimental tests have been
carried out on a straight railway track. The wheel profile is
the ORE S1002 (with a wheelset width dw equal to 1.5 m)
while the rail profile is the UIC60 (with a gauge dr equal to

Fig. 11.

sp

sp

Experimental vehicle and wheelset velocities v sp , v wi = rw wi .

1.435 m and a laying angle p equal to 1/20 rad). The main

characteristics of the braking test, considered as benchmark in
this paper, are summarized in Table III.
The value of the kinetic friction coefficient under degraded
adhesion conditions d depends on the test that has to
be performed on the track; the degraded adhesion conditions
are reproduced using a watery solution containing surfaceactive agents, e.g., a solution sprinkled by a specially provided nozzle on the first wheelset in the running direction.
The surface-active agent concentration in the solution varies
according to the type of test and the desired friction level.
The value of the kinetic friction coefficient under full adhesion recovery r corresponds to the classical kinetic friction
coefficient under dry conditions.
sp
sp
First, the vehicle and wheelset velocities v sp , v wi = rw wi
(i = 1, . . . , 4) are considered (Fig. 11). Both the WSP
intervention and the adhesion recovery in the second part of
the braking maneuver are clearly visible.
Second, the slidings among the wheelsets and the rails have
sp
sp
sp
been considered: si = v sp rw wi = v sp v wi (Figs. 1417).
However, these physical quantities cannot be locally compared
with each other because of the complexity and the chaoticity of
the system due, for instance, to the presence of discontinuous
and threshold elements like the WSP. To better evaluate the
sp
behavior of si from a global point of view, it is useful to
sp
introduce the statistical means si and the standard deviations
sp

i of the considered variables

sp
si
sp

 TF
1
sp
=
s dt
TF TI T I i

 TF 

1
sp
sp 2
=
dt
si si
TF TI T I

(14)

ALLOTTA et al.: MODELING AND CONTROL OF A FULL-SCALE ROLLER-RIG

193

TABLE IV
M AIN C ONTROL AND I NTEGRATION PARAMETERS

Fig. 13.

Experimental and simulated train velocities v sp , v s .

TABLE V
sp

sp

E XPERIMENTAL si ,
i AND S IMULATED ssi ,
si
S TATISTICAL I NDICES

Fig. 14.
wheelset.

Fig. 12.

Simulated vehicle and wheelset velocities v s , v wsi = rw wsi .

where TI and TF are, respectively, the initial and final times

of the simulation (Table V).
V. M ODEL VALIDATION
In this section. the whole HIL architecture model is simulated and validated. More in detail, both the dynamical and
the control performances of the system will be analyzed. The
main control and integration parameters are summarized in
Table IV (see Section III-C) [41].
The simulated vehicle and wheelset velocities v s , v wsi =
rw wsi are reported in Fig. 12). Figs. 11 and 12 highlight
a good qualitative matching between the experimental and

sp

Experimental and simulated train slidings s1 , ss1 for the first

simulated data, both concerning the WSP intervention and the

adhesion recovery in the second part of the braking maneuver.
The direct comparison between the experimental and simulated train velocities v sp , v s is shown in Fig. 13 and shows
also a good quantitative agreement between the considered
quantities.
Subsequently, according to Section IV, the simulated slidings among the wheelsets and rails ssi = v s rw wsi =
v s v wsi are considered and compared with the experimental
sp
ones si (Figs. 1417).
The matching between the experimental and simulated
slidings is qualitatively good. However, since these physical quantities cannot be locally compared with each other
because of the complexity and the chaoticity of the system,
the statistical means ssi and the standard deviations
si of
the simulated slidings ssi are introduced (according to 14) to
better evaluate the global behavior of analyzed variables. The
sp
sp
comparison between the experimental si ,
i and simulated
ssi ,
si statistical indices is reported in Table V and highlights
also a good quantitative match between the studied quantities.
The controller performances are evaluated in terms of
angular velocity error e = ws w and the torque estimation error ec = C s Cs . Small values of the errors e , ec
l/r
assure a good estimation of the tangential contact forces Tc .
The time history of the angular velocity error e is plotted

194

Fig. 15.
wheelset.

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 1, JANUARY 2015

Fig. 18.

Angular velocity error e = ws w .

Fig. 19.

Torque estimation error ec = C s Cs .

Fig. 20.

Torque applied to the wheelset Cs .

sp
Experimental and simulated train slidings s2 , ss2 for the second

sp

Fig. 16. Experimental and simulated train slidings s3 , ss3 for the third
wheelset.

sp

Fig. 17. Experimental and simulated train slidings s4 , ss4 for the fourth
wheelset.

in Fig. 18 and shows the control capability of stabilizing the

system and rejecting the disturbances produced by the initial
transient and the adhesion recovery in the second phase of the
braking maneuver.
The torque estimation error ec = C s Cs and the real
torques applied to the wheelset Cs are reported, respectively,
in Figs. 19 and 20. Also in this case, the controllers turn out
to be effective in reproducing the real torques applied to the
wheelsets of the vehicle.

Finally, the result analysis highlights the control capability

of reproducing on the roller-rig a generic wheel-rail degraded
adhesion pattern calculated by the reference virtual railway
vehicle model (in terms of angular velocities w , applied
torques on the wheelsets Cs and, consequently, in terms
l/r
of tangential efforts Tc exchanged between the wheelsets
and the rails). In particular, if compared with the previous
steps of the research activity, the new HIL architecture model
shows the following encouraging features:
1) better results accuracy (especially, in the second part of
the braking maneuver where adhesion recovery occurs
and system dynamics becomes more critical) due to the
effectiveness of the new independent controllers and
to the accuracy and smoothness of the new adhesion
models;

ALLOTTA et al.: MODELING AND CONTROL OF A FULL-SCALE ROLLER-RIG

195

2) better controller performances (especially, in the second

part of the braking maneuver) because of the independent control approach used in the HIL architecture;
3) minor numerical noise (especially, in the second part
of the braking maneuver) due to the accuracy and
smoothness of the new adhesion models.

[4] R. Dukkipati, A parametric study of the lateral stability of a rail bogie

on a roller rig, Proc. Inst. Mech. Eng. F, vol. 213, no. 1, pp. 3947,
1999.
[5] W. Zhang, J. Chen, X. Wu, and X. Jin, Wheel/rail adhesion and analysis
by using full scale roller rig, Wear, vol. 253, nos. 12, pp. 8288, 2002.
[6] N. Bosso, A. Gugliotta, and A. Som, Introduction of a wheel-rail
and wheel-roller contact model for independent wheels in a multibody
code, in Proc. ASME/IEEE Joint Rail Conf., Washington, DC, USA,
Apr. 2002, pp. 151159.
[7] N. Bosso, A. Gugliotta, and A. Som, Comparison of different scaling
techniques for the dynamics of a bogie on roller-rig, in Proc. 17th
IAVSD Symp., Copenaghen, Denmark, 2001.
[8] M. Malvezzi, B. Allotta, and L. Pugi, Feasibility of degraded adhesion
tests in a locomotive roller rig, Proc. Inst. Mech. Eng. F, vol. 222,
no. 1, pp. 2743, 2008.
[9] B. Allotta, R. Conti, E. Meli, L. Pugi, and A. Ridolfi, Development
of a HIL railway roller rig model for the traction and braking testing
activities under degraded adhesion conditions, Int. J. Non-Linear Mech.,
vol. 57, pp. 5064, Dec. 2013.
[10] N. Bianchi, M. D. Pr, and S. Bolognani, Design of a fault-tolerant IPM
motor for electric power steering, IEEE Trans. Veh. Technol., vol. 55,
no. 4, pp. 11021111, Jul. 2006.
[11] L. Parsa and H. A. Toliyat, Fault-tolerant interior-permanent-magnet
machines for hybrid electric vehicle applications, IEEE Trans. Veh.
Technol., vol. 56, no. 4, pp. 15461552, Jul. 2007.
[12] K. Gulez, A. A. Adam, and H. Pastaci, A novel direct torque
control algorithm for IPMSM with minimum harmonics and torque
ripples, IEEE/ASME Trans. Mechatronics, vol. 12, no. 2, pp. 223227,
Apr. 2007.
[13] Y. Okada, S. Miyamoto, and T. Ohishi, Levitation and torque control of
internal permanent magnet type bearingless motor, IEEE Trans. Control
Syst. Technol., vol. 4, no. 5, pp. 565571, Sep. 1996.
[14] R. Krishnan, Permanent Magnet Synchronous and Brushless DC Motor
Drives. Boca Raton, FL, USA: CRC Press, 2010.
[15] A. Vagati, G. Pellegrino, and P. Guglielmi, Design tradeoff between
constant power speed range, uncontrolled generator operation and rated
current of IPM motor drives, IEEE Trans. Ind. Appl., vol. 47, no. 5,
pp. 19952003, Sep./Oct. 2011.
[16] J. J. Kalker, Three-Dimensional Elastic Bodies in Rolling Contact.
Norwell, MA, USA: Kluwer, 1990.
[17] O. Polach, Creep forces in simulations of traction vehicles running on
adhesion limit, Wear, vol. 258, nos. 78, pp. 9921000, 2005.
[18] O. Polach, A fast wheel-rail forces calculation computer code, Veh.
Syst. Dyn., Int. J. Veh. Mech. Mobility, vol. 33, pp. 728739, Aug. 1999.
[19] J. Pombo and J. Ambrosio, Application of a wheel-rail contact model to
railway dynamics in small radius curved tracks, Multibody Syst. Dyn.,
vol. 19, nos. 12, pp. 91114, 2008.
[20] J. Pombo and A. J. Silva, A new wheel-rail contact model for railway
dynamics, Veh. Syst. Dyn., Int. J. Veh. Mech. Mobility, vol. 45, no. 2,
pp. 165189, 2007.
[21] P. J. Blau, Embedding wear models into friction models, Tribol. Lett.,
vol. 34, no. 1, pp. 7579, 2009.
[22] M. Boiteux, Le probleme de ladherence en freinage, Revue Generale
des Chemins de fer, vol. 105, no. 6, pp. 5972, Apr. 1986.
[23] P. Voltr, M. Lata, and O. Cerny, Measuring of wheel-rail adhesion
characteristics at a test stand, in Proc. 18th Conf. Eng. Mech., 2012.
[24] M. Malvezzi, E. Meli, and S. Falomi, Multibody modeling of railway
vehicles: Innovative algorithms for the detection of wheel-rail contact
points, Wear, vol. 271, nos. 12, pp. 453461, 2011.
[25] S. Magheri, M. Malvezzi, E. Meli, S. Falomi, and A. Rindi, An
innovative wheel-rail contact model for multibody applications, Wear,
vol. 271, nos. 12, pp. 462471, 2011.
[26] S. Falomi, M. Malvezzi, E. Meli, and A. Rindi, Determination of
wheelRail contact points: Comparison between classical and neural
network based procedures, Meccanica, vol. 44, no. 6, pp. 661686,
2009.
[27] A. Innocenti, L. Marini, E. Meli, G. Pallini, and A. Rindi, Development
of a wear model for the analysis of complex railway networks, Wear,
vol. 309, nos. 12, pp. 174191, 2013.
[28] B. Allotta, E. Meli, A. Ridolfi, and A. Rindi, Development of an innovative wheel-rail contact model for the analysis of degraded adhesion
in railway systems, Tribol. Int., vol. 69, pp. 128140, Jan. 2013.
[29] E. Meli, A. Ridolfi, and A. Rindi, An innovative degraded adhesion
model for railway vehicles: Development and experimental validation,
Meccanica, vol. 49, no. 4, pp. 919937, 2014.

VI. C ONCLUSION
In this paper, the authors described the modeling and control
of an innovative HIL architecture to test braking on board
subsystems on full-scale roller-rigs. The new strategy permits
to reproduce on the roller-rig, a generic wheel-rail adhesion
pattern and, in particular, degraded adhesion conditions. The
proposed approach has been preliminarily validated through
the experimental data provided by Trenitalia and highlighted a
good performance in reproducing on the roller-rig the complex
interaction between the degraded adhesion conditions and
railway vehicle dynamics during the braking maneuver.
From the railway industry point of view, the innovative
full-scale roller-rig of the Firenze-Osmannoro research facility
represents an important improvement if compared with the
current state-of-the-art. The main advantages of the new rollerrig concern the reduction of the expensive on-track tests (rent
of vehicle and railway lines) and the possibility to test several
on board subsystems on a roller-rig both with good and
degraded adhesion conditions.
Many further developments of this research activity are
scheduled for the future. They will regard the implementation
of the control strategy and the virtual vehicle model on
the Firenze-Osmannoro roller-rig. This way, first, a further
validation of the proposed HIL approach will be possible
through experimental tests performed directly on the rollerrig. Subsequently, the whole HIL system will be employed
to design and test new on board subsystems such as WSP,
antiskid, and hence forth.
Moreover, from a theoretical viewpoint, the differences
between wheel-rail contact and wheel-roller contact will be
investigated. Such differences may affect the roller-rig system
in many ways (kinematics, stability, bifurcation diagrams,
dynamics, wear, and hence forth) and have to be accurately
studied.
ACKNOWLEDGMENT
The authors would like to thank Trenitalia, SIMPRO, and
SICME Motors for supplying the technical and experimental
data relative to the UIC-Z1 vehicle, the Firenze-Osmannoro
roller-rig and the braking tests under degraded adhesion
conditions.
R EFERENCES
[1] K. Ahn, J. Park, and S. Ryew, The construction of a full-scale wheel/rail
roller rig in Korea, in Proc. IEEE Int. CASE, Aug. 2012, pp. 802803.
[2] N. Lee, C. Kang, W. Lee, and T. Dongen, Roller rig tests of a semiactive suspension system for a railway vehicle, in Proc. IEEE ICCAS,
Oct. 2012, pp. 117122.
[3] A. Jaschinki, H. Chollet, and S. Iwnicki, The application of the roller
rigs to railway vehicle dynamics, Veh. Syst. Dyn., Int. J. Veh. Mech.
Mobility, vol. 31, nos. 56, pp. 325344, 1999.

196

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 1, JANUARY 2015

[30] M. Ignesti, M. Malvezzi, L. Marini, E. Meli, and A. Rindi, Development of a wear model for the prediction of wheel and rail profile
evolution in railway systems, Wear, vols. 284285, pp. 117, Apr. 2012.
[31] Railway Applications, Braking-Wheel Sliding Protection, UNI-EN
Standard 15595, 2009.
[32] Trenitalia SpA, UIC-Z1 coach, Trenitalia, Rome, Italy, Internal Rep.,
2000.
[33] Trenitalia SpA, WSP system, Trenitalia, Rome, Italy, Internal Rep.,
2005.
[34] Trenitalia SpA, On-track braking tests, Trenitalia, Rome, Italy, Internal
Rep., 2006.
[35] Trenitalia SpA, Full-scale roller-rig: Technical documentation,
Trenitalia, Rome, Italy, Internal Rep., 2011.
[36] SICME Motori, IPM synchronous motor datasheet, SICME Motori,
Turin, Italy, Internal Rep., 2010.
[37] E. D. Sontag, Mathematical Control Theory. New York, NY, USA:
Springer-Verlag, 1998.
[38] H. K. Khalil, Nonlinear Systems. Englewood Cliffs, NJ, USA:
Prentice-Hall, 2002.
[39] J. Nocedal and S. Wright, Numerical Optimization (Operation Research).
Berlin, Germany: Springer-Verlag, 1999.
[40] C. Kelley, Iterative Methods for Linear and Nonlinear Equations.
Philadelphia, PA, USA: SIAM, 1995.
[41] L. Shampine and M. Reichelt, The MATLAB ODE suite, SIAM J. Sci.
Comput., vol. 18, no. 1, pp. 122, 1997.
[42] L. Governi, M. Carfagni, R. Furferi, L. Puggelli, and Y. Volpe, Digital
bas-relief design: A novel shape from shading-based method, Comput.Aided Design Appl., vol. 11, no. 2, pp. 153164, 2014.
[43] W. Cai, W. Liao, D. Li, and Y. Song, Observer based traction/braking
control design for high speed trains considering adhesion nonlinearity,
Abstract Appl. Anal., 2014, doi: 10.1155/2014/968017.

Benedetto Allotta (M12) received the Ph.D. degree

in robotics from the Scuola Superiore S. Anna, Pisa,
Italy, in 1992.
He is a Full Professor of Robotics with the School
of Engineering, University of Florence, Florence,
Italy, where he is the Co-Founder and Co-Ordinator
with the Laboratory of Mechatronics and Dynamic
Modelling. His current research interests include
visual control of robots, mechatronics, underwater
robotics, sensor-based navigation of vehicles, hardware in the loop simulation, and automation in
transport systems.

Roberto Conti is a Ph.D. Researcher and Assistant

Professor of Robotics with the School of Engineering, University of Florence, Florence, Italy. His
current research interests include robotics, wearable
robotics, and underwater robotics.

Enrico Meli is a Ph.D. Researcher and Assistant

Professor of Machine Theory and Robotics with
the School of Engineering, University of Florence,
Florence, Italy. His current research interests include
vehicle dynamics, autonomous vehicles, and control
theory.

Alessandro Ridolfi is a Ph.D. Researcher and Assistant Professor of Machine Theory and Robotics with
the School of Engineering, University of Florence,
Florence, Italy. His current research interests include
vehicle dynamics, robotics, and underwater robotics.