Mechatronics 21 (2011) 1108–1122

Contents lists available at ScienceDirect

Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics

Integrated tilt with active lateral secondary suspension control for high speed
railway vehicles
Ronghui Zhou, Argyrios Zolotas ⇑, Roger Goodall
Control Systems Group, Department of Electronic and Electrical Engineering, Loughborough University, Loughborough LE11 3TU, UK

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a novel active suspension control configuration for high speed tilting railway vehicles
Received 19 July 2010 which integrates tilt with active lateral secondary suspension. The use of the active lateral secondary sus-
Accepted 3 July 2011 pension is to attenuate the vehicle body lateral vibration on straight track, while complementing tilt
Available online 4 August 2011
action during curving. Various control strategies are proposed to accommodate both tilt and active lateral
suspension multiple design requirements, whilst considering the strong interaction between vehicle
Keywords: body roll and lateral modes. Compared with the commercial solutions for tilt control, the proposed inte-
Railway vehicle dynamics
gration strategy improves the tilting control performance both on curved and straight track as illustrated
Active lateral suspension control
Tilt control system
by simulation tests and control assessments based on given track profiles.
Estimator-based control Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction
Tilting trains are now well accepted within the rail industry as a
means to operate at higher speeds compared to conventional
trains, without the need to upgrade rail infrastructure. The idea
is quite straightforward, based upon tilting the vehicle body in-
wards on the curve to compensate the larger lateral acceleration
that would otherwise be perceived by the passenger at higher
speed. Early passive tilting trains completely relied upon the natu-
ral pendulum motion laws which caused safety issues, i.e. vehicle
body over turning [1]. A tilt mechanism (tilting bolster in most
cases) in conjunction with an actuator to tilt the vehicle body
was introduced afterwards, which has become a standard technol-
ogy used in trains worldwide.
Fig. 1a shows a typical mechanical configuration, the example
type being the Swedish X2000 tilting train which is currently run-
ning between Stockholm and Gothenburg. In this configuration,
wheelsets are connected to the bogie via stiff primary suspension
elements, which are designed to meet vehicle stability and guid-
ance requirements. Each bogie has four wheels arranged in two
pairs, where each pair is rigidly connected via a common axle
(known as the solid-axle wheelset). A tilting bolster is connected
to the bogie by inclined swing links, and actuators fitted between
the bogie and the bolster create the tilting action below the sec-
ondary suspension. The tilting bolster is able to provide a maxi-
mum tilt of up to 10°, and the inclined link arrangement means
⇑ Corresponding author. Tel.: +44 (0)1509 227085.
E-mail addresses: r.zhou@lboro.ac.uk (R. Zhou), a.c.zolotas@lboro.ac.uk
(A. Zolotas), r.m.goodall@lboro.ac.uk (R. Goodall).
that the effective tilt center is above the vehicle body floor level.
A pair of airsprings are located between the body and the bolster,
and combined with dampers forms the soft passive secondary sus-
pension. This isolates the high frequency vibrations from the bogie
and provides an appropriate passenger ride comfort.
The curved sections on railway tracks are ‘‘canted’’ inward to-
wards the center of the curve, as shown in Fig. 1b, which reduces
the lateral (curving) acceleration experienced by the passengers.
The resultant lateral acceleration is known as cant deficiency (D,
m/s2), and is given by (in the case of non-tilt trains):
v2
D¼  cosðh0  jhv jÞ  g  sinðh0  jhv jÞ ð1Þ
R
where v is the vehicle forward speed, h0 is the track cant angle, hv
here is the outwards vehicle body roll angle, R is the track curve
radii and g is the constant of gravity. However, cant deficiency is
most often defined as an angle (hdm = D/g) presenting the difference
between the existing cant angle (h0 in (1)) and the angle required to
fully eliminate the effect of centrifugal force at maximum allowable
speed [1].
Higher speeds through curves with conventional tracks can only
be achieved by lengthening transitions and increasing curve radius,
which is very expensive and in some cases impossible. Alterna-
tively, tilting effectively reduces the cant deficiency by leaning
the vehicle bodies further towards the curve center. The passenger
curving acceleration in tilting trains can be calculated by following
equation,
v2
D¼  cosðh0 þ jhv jÞ  g  sinðh0 þ jhv jÞ ð2Þ
R

0957-4158/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechatronics.2011.07.001
 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122 1109

v 0
v
2
v
cos
Vehicle body R

v2
R g sin
Airspring

Tilt actuator
Tilting bolster
Primary
suspension
Vehicle
Wheelset
Track bogie 0

Fig. 1. Tilting trains.

where hv now is the inwards tilting action. Although the concept is introducing proportion of secondary suspension roll angle to pro-
relatively straightforward, Section 2 explains in detail that achiev- vide partial tilt compensation (60–70%). However, it is proved dif-
ing effective control of the tilting action is challenging, principally ficult to satisfy the trade-off between straight track ride comfort
because of the dynamic interaction between the lateral and roll/tilt and curving performance [4], largely due to the fundamental prob-
modes of the vehicle body. This paper studies rather novel concept lem of the interaction with the lateral suspension, because the roll
of (from the integrated point of view) the lateral actuator integrated and lateral modes of the vehicle body are strongly coupled in a dy-
into the system and installed between the vehicle body and tilting namic sense. If the tilting control loop bandwidth is low enough
bolster replacing the conventional passive lateral suspension not to interfere with the lateral suspension, it is then too slow act-
damper. This together with the airspring forms the active lateral ing on the curve transition.
secondary suspension. The control system design for the formed Various options for tilting have been studied, including the use
dual-actuator system is carried out using both classical design of inverse dynamics [5], but the most common commercial solu-
and modern control approaches. The set of research objectives is tion for the tilt actuator control is called ‘‘Precedence control’’, with
twofold: the scheme shown in Fig. 2b, which is based upon providing tilt
command from the vehicle body in front [3]. In this strategy, the
(1) to improve the straight track ride quality of the tilting train
(2) to attenuate the dynamics interaction between lateral and
roll modes of the vehicle body in order to provide high per- Track
formance local (that is per vehicle body) tilt control. Body lateral
0 (zero) acceleration
Controller Vehicle
2. Evolution of tilting control systems Dynamics
Suspension roll
The main performance requirements for the tilt control system
[2] are:
1/g
(i) to provide a comfortable response during curve transitions Equivalent cant Equivalent cant
(ii) to maintain the straight track ride quality (which may be deficiency deficiency angle
deteriorated by the tilting) within acceptable limits

From a control point of view the objectives of the tilt control

system can be translated as, increasing the response of the system Scalling Vehicle 1 Actuator
at low frequencies (Curved track features) while reducing the high Bogie command
k/g LPF K(s)
frequency system response (Straight track features) and of course accel.
maintaining stability. The following paragraphs explain how tilt Tilt angle 1
Preview
control has developed over two or three decades. effect
Tilt control system based upon local vehicle measurements was Vehicle 2 Actuator
used in early tilting trains, branded at that time: ‘‘nulling control’’,
command
k/g LPF K(s)
and the scheme is shown in Fig. 2a (the cant deficiency is ex-
Tilt angle 2
pressed as angle here). The body lateral acceleration is measured
and employed to drive the tilt actuator in a direction which will Digitally .... Vehicle 3, etc.
bring it to zero (i.e. null acceleration level). However, full compen- transmitted
sation of the lateral acceleration experienced by the passenger has
been found to be one of the reasons to cause motion sickness [3].
Thus, the basic nulling strategy has been modified by simply Fig. 2. Tilt control schemes.
1110 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122
lateral accelerometer on a non-tilting part (vehicle bogie, to avoid
the coupled body modes interaction) of the previous vehicle indi-
cates the required tilting angle, with the body tilt angle feedback
controller locally ensuring that each vehicle tilts to the com-
manded angle, the carefully tuned advanced information enables
a sufficient level of filtering (Low Pass Filter (LPF) is used) to be ap-
plied to remove the effect of track irregularities (introduced by the
bogie-mounted accelerometer) on the tilt command signal.
More recently, ALSTOM developed a command driven controller
named ‘‘Anticipative Tilt Control’’ [6] which integrates the vehicle
measurement with information from an onboard track database
and provides the track information to the tilt controller. Therefore,
as long as the location of the train is determined, the system can
generate the tilt command using track data from the database. Also
the system is still fundamentally a precedence scheme, but aug-
mented by the track information when synchronized. However,
both the position of each vehicle along the track and the curve data
contained in the database need to be known accurately and reli-
ably for this approach to work effectively.
The research work described in [4,7] proposed advanced tilt
control based on local vehicle body signals that do not require pre-
cedence information, which makes the system simpler and more
straightforward in terms of failure detection, and also benefits
the leading vehicle which does not have the precedence informa-
tion. Various advanced controllers have been tested by means of
the system simulation, e.g. LQG control and H1 control. However
due to the dynamic interaction between roll and lateral modes of
the railway vehicle body, there is a limit to what can be achieved
in terms of improving the transition performance. Also, tilting
trains run at higher speed on the same rail infrastructure compared
with non-tilting train which deteriorates the ride-quality on
straight track.
Some commercial tilting trains nowadays combine the prece-
dence tilting with a hold-off device. It is a low bandwidth full ac-
tive lateral suspension which actively moves the vehicle body
laterally and is installed to compensate for quasi-static lateral force
due to residual lateral acceleration on curves. The control can be
fulfilled by introduced a displacement sensor fitted to the bogie
[8]. It has been used on the service vehicle, such as: Virgin Class
390 tilting trains for the UK West Coast Main Line, for which pneu-
matic cylinders are used [9]. The mainline test was also carried out
by Siemens SGP [10] which combined the semi-active damping, tilt
active suspension control and Hold-off device, also the lateral dis-
placement is measured to configure the hold-off device controller.
Moreover, the numerical model and the field test result of Italian
‘Pendolino’ tilting train combined with an active lateral suspension
(centering device) can be found in [11].
Although both tilt technology and active lateral secondary sus-
pension technology have been developed very well and the tilting
suspension has been combined with preliminary forms of lateral
centering device and semi-active damping, the full integration of
tilt and active lateral suspension has not been studied, and this is
the main contribution of the research in this paper.
3. System performance requirement and controller assessment
approach
System performance requirements and the control assessment
approach for active railway suspensions as well as tilt control de-
sign have been discussed in [2,12], and can be summarized below.
The active suspension design is a multi-objective optimization
process which needs to minimize the body acceleration on straight
track, consider the constraints for suspension deflection and sys-
tem stability, and optimize the curving performance. The controller
design for the dual-actuator system needs to meet both tilting and
active lateral suspension performance requirements. Two main de-
sign criteria are involved:
3.1. Deterministic performance criterion (respect to the curved track)
Maintaining appropriate curve transition comfort level for
standing and seated passengers, quantified by Pct value which pro-
vides the percentage of (both standing and seated) passengers who
feel uncomfortable during the curve transition, and can be calcu-
lated using the vehicle body lateral acceleration, lateral jerk and
roll rate [13], more details can be found in Appendix A. Moreover,
the tilt controller assessment also relies upon identifying how the
tilting vehicle would ideally perform on the transition from
straight to curved track and then quantifying the deviation of the
actual response compared with the identified ideal response. Asso-
ciated with performance improvement is the constraint on lateral
suspension deflection, which should not exceed the maximum al-
lowed requirement before bump stops are reached, i.e. ±60 mm
is used in this study.
3.2. Stochastic performance criterion (respect to the straight track)
Maximizing the ride quality in response to track irregularities
on straight track (stochastic criterion). Straight-track ride quality
of the tilting train should degrade no more than 7.5% compared
to the non-tilting train equivalent at the higher speed enabled by
the tilting action, but the aim should be to minimise passenger lat-
eral acceleration measurement (assessed by its R.M.S. value) by
means of the lateral actuator, i.e. in addition to improving the tilt
system response.
The vehicle models and control systems are tested with specific
track inputs including both deterministic (low frequency) and sto-
chastic (higher frequency) track features. The deterministic track
used was a curved track with a radius of 1200 m and a maximum
track cant angle (h0) of 5.84°, with a 150 m transition at the start
and end of the steady curve, as shown in Fig. 3. The track inputs re-
flect the design alignment  that is used in practice, in which the va-
lue of track curvature 1R and cant angle (h0) are varied (increase/
decrease) linearly during the transitions simultaneously, reaching
the maximum during the steady-state section. This simultaneous
variation introduces both lateral and roll inputs during the transi-
tion and produces a complex dynamic response from railway vehi-
cles, such as the coupled lateral and roll dynamics modes of the
vehicle body. The stochastic track inputs represent the irregulari-
ties in the track alignment on both straight track and curves, and
these were characterized by an approximate spatial spectrum
Transition Transition
Straight Steady- Straight
state
0
Cant
angle
1 Meter
R
Track
curvature
Meter
Fig. 3. Curved track section.
 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122 1111

equal to ((2p)2Xlv2/f(m2/(cycle/m)) with a lateral track roughness actuator technology the reader can refer to [15], however actuator
(Xl) of 0.33  108 [14]. technology is not a main concern of this paper contribution). The
numerical equations for the integrated tilting bolster and active
lateral secondary suspension are presented below, with system
4. Vehicle dynamics modeling parameters are explained and given values in Appendix B. Since
the roll/tilt angles are all relatively small, the equations below all
Fig. 4 shows the diagram of the end-view model for the tilting use the small angle approximation, i.e. sin(h) = h, cos(h) = 1.
train with tilting bolster. This model was originally presented in
[4] however without the lateral actuator. This model is extended Body lateral dynamics:
in this paper to include the lateral actuator and thus replace the
lateral secondary suspension damper. Both the lateral and roll €v ¼ 2ksy yv þ 2ksy h1 hv þ 2ksy yb þ 2ksy h2 hb
mv y
dynamics of the vehicle body and bogie are presented. The vertical mv v 2
dynamics are ignored but the vertical suspension is represented by  ð2hmt ksy  mv gÞhm þ mv gh0 
R
a pair of airsprings which contributes to the roll dynamics and this
 mv hg1 €h0 þ F a ð3Þ
effect is included in the modeling phase. An anti-roll bar system,
the end moment effect (due to the movement of the body center
Body roll dynamics:
of gravity), the translation and rotation of these reference axes
associated with curves are all depicted in the model. 2
iv r €hv ¼ ð2h1 ksy þ mv gÞyv  ½kv r þ 2h1 ksy
The tilt actuation system is modeled as a position servo in series 2
with mechanism which gives 3.5 Hz bandwidth and 50% damping. þ 2d1 ðkaz þ ksz Þhv  2h1 yb ðksy þ mv gÞ
2
It is a reasonable approximation because tilting is a low bandwidth þ ðkv r þ 2d kaz  2h1 h2 ksy Þhb  cv r h_ v
1
action. Since, in this study, the focus is on the behaviour of the roll 2 2
þ cv r h_ b þ 2ksz d1 hr þ ðkv r þ 2d1 kaz þ 2ksy h1 hmt Þhm
and lateral motion of the vehicle, thus, the effect of the wheelset
dynamics is incoprated simplly by using a 2rd order low pass filter þ cv r h_ m  iv r €h0  F a h1 ð4Þ
with 20 Hz frequency cutoff frequency and 20% damping (wheelset
Bogie lateral dynamics:
dynamics do not affect the tilt action and in general the main
behaviour of the secondary suspension, and thus the vehicle pas- €b ¼ 2ksy yv  2h1 ksy hv  2ðksy þ kpy Þyb
mb y
senger compartment). In this study, another actuator (F a ) is in-  2ðh2 ksy  h3 kpy Þhb  2cpy y_ b þ 2h3 cpy h_ b
stalled between the body and bolster in the lateral direction.
þ 2kpy yw þ 2cpy y_ w þ 2hmt ksy hm þ mb gh0
Ideal actuator is used for the controller design, but a variety of
technologies could be modelled, e.g. electro-mechanical actuator, mb v 2
  mb hg2 €h0  F a ð5Þ
hydraulic actuator and magnetic actuator (for more details on R

Vehicle body
d1

y+v
kvr
cvr
θv+ h1 h vt
Tilt center
k sy
k az k sz
Tilting k h mt
bolster rz c rz Fa bolster
h g1
+
θm
connection cpz y+b bogie h2
to bogie θb+
kpz
kpy
connection h3
to bolster Wheelset
h g2 c py d2
θ+o Rail level
y+
o

Fig. 4. End-view modelling.

1112 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122

Bogie roll dynamics: industrial-norm, damping is changed due to the damper coefficients
2
of the lateral secondary suspension being set to zero (replaced by a
ibr €hb ¼ 2h2 ksy yv þ ½kv r  2h2 h1 ksy þ 2d1 ðkaz þ ksz Þhv uncontrolled actuator during modeling). Active lateral secondary
2 2 suspension control will improve the lateral damping for these
 2ðh2 ksy  h3 kpy Þyb  ðkv r þ 2h2 ksy þ 2h3 kpy
2 2 modes, therefore the low lateral damping 0.75% is of no concern
þ 2d2 kpz þ 2d1 kaz Þhb þ cv r h_ v þ 2h3 cpy y_ b
as it stands in the uncontrolled case.
2 2 2
 ðcv r þ 2d cpz þ 2h cpy Þh_ b  2ksz d hr
2 3 1
2
 2h3 kpy yw  2h3 cpy y_ w  ðkv r þ 2d1 kaz  2ksy h2 hmt Þhm 5. Control system design
 cv r h_ m  ibr €h0  F a h2 ð6Þ
In this section, a set of control strategies controlling the tilting
Tilting actuation system effect: train with lateral actuator will be presented. A number of remarks
hm 483:6 are drawn for the particular cases.
ðsÞ ¼ 2 ð7Þ The first control strategy to be discussed is a Classical Decen-
hmi s þ 22s þ 483:6
tralized dual-actuator control (CD control) scheme with a tilt con-
Wheelset dynamics effect: trol loop and a control loop for the lateral actuator. In the first
stage, the coupling of the modes is not considered in the controller
yw 987
ðsÞ ¼ 2 ð8Þ design, while only local per vehicle sensors are used (i.e. no prece-
y0 s þ 12:57s þ 987
dence). Control parameters are fine-tuned with the Genetic Algo-
For the air spring state: rithms, also the limitation of this simple scheme is explained.
The next scheme refers to the Command-Driven Decentralized
h_ r ¼ c1 _ _
rz ðhr ðksz þ krz Þ þ ksz hv þ krz ðhb þ hm Þ þ crz ðhb þ hm ÞÞ dual-actuator control with precedence solution (CDD control)
ð9Þ which duplicates the commercial used approach but with the addi-
tion of the lateral actuator. The acceleration signal from the prece-
Tilting actuation system effect:
dent vehicle (mounted on the bogie) is used for tilt control while
hm 483:6 the lateral actuator control is provided locally. The discussion of
ðsÞ ¼ 2 ð10Þ two advanced tilt control schemes with local per vehicle schemes,
hmi s þ 22s þ 483:6
i.e. MIMO optimal control and Estimator-Based Decentralized
Wheelset dynamics effect: control (EBD control), follow to illustrate further performance
yw 987 improvement.
ðsÞ ¼ 2 ð11Þ The proposed schemes are compared to tilt-only (without lat-
y0 s þ 12:57s þ 987
eral actuator) controllers, i.e. tilt-only Precedence Tilt (PT) and
early Nulling Tilt (NT) control with classical PI controller. These
The system state space form is presented below, which has two two schemes provide benchmarks against which the benefits of
inputs (u) being the tilt angle command and the lateral actuator the second (lateral) actuator and the advanced local strategies
force command, states (x) and track disturbances (w). can be compared. The interested reader is referred to [4,16] regard-
ing details of tilt-only PT and early NT control.
x_ ¼ Ax þ Bu þ Cx ð12Þ
Three comparisons (as presented in Table 2) are addressed in
y ¼ Cx þ Du þ Hx ð13Þ this section:
where
(1) The comparison between CD control and NT control is to
x ¼ ½yv hv yb hb y_ v h_ v y_ b h_ b hr yw y_ w hm h_ m T g u ¼ ½hmi F a T show how the addition of the lateral actuator enables the
€h0 y T difficulties with local control to be overcome.
x ¼ ½R1 R_ 1 h0 h_ 0 0
(2) The comparison between CDD control and PT control is to
Disturbance parameters h0 ; h_ 0 ; y0 are included in the matrix A only show how much extra benefit is there from using a second
for simulation purposes. The system is dynamically complex with actuator (lateral actuator) with the commercial prece-
strong coupling between the lateral and roll mode which results dence-based solution.
in two sway modes combining both lateral and roll movement: an (3) The comparison between the advanced control (particularly
‘upper sway’ mode, with the node appears above the body c.o.g., the EBD control) and PT control is to demonstrate that
characterized predominately by a roll movement; and a ‘lower advanced control schemes (with the lateral actuator) are
sway’ mode, its node located below the body c.o.g., characterized able to enable the local per vehicle control to be as effective
predominantly by a lateral motion. The system modes are listed as precedence solution.
in Table 1, with the frequency of modes being close to the

Table 1 5.1. Classical decentralized dual-actuator control based on ‘nulling

Tilt mechanism vehicle model dynamic modes. control’ configuration
Mode Damping (%) Frequency (Hz)
A sequential design process is used in the CD control because
Original system model [4]
the lateral actuator control loop is a high bandwidth strategy that
Lower sway 21.8 0.48
Upper sway 20.9 1.35 Table 2
Bogie lateral 9.95 16.7 Benchmarks and controller performance comparisons.
Bogie roll 28.3 7.26
Comparison Tilt actuator only Tilt + lateral
Enhanced model with lateral actuator
(Benchmarks) actuator
Lower sway 0.75 0.47
Upper sway 4.41 1.37 (1) NT <--> CD
Bogie lateral 6.35 16.55 (2) PT <--> CDD
Bogie roll 25.22 7.27 (3) PT <--> MIMO optimal EBD
 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122 1113

responds to the high frequency track irregularity (2–10 Hz), being
faster than the tilting action (for which the bandwidth is below
1 Hz). Controller parameters are tuned by Genetic Algorithm
(GA) to obtain the best performance.
5.1.1. Active lateral secondary suspension control design
The complementary filter skyhook damping control [17] com-
bined with the integral of the lateral secondary suspension deflec-
tion (xdefl) is designed for the lateral actuator control, which has the
function to improve the lateral ride quality whilst keeping the
deflection within the required limit to avoid the lateral bump stop.
It is seen as a better solution compared to the semi-damper com-
bined with hold-off device [18]. Note that the body lateral second-
ary suspension deflection (xdefl) is defined as:
xdefl ¼ yv  h1 hv þ hmt hm  yb  h2 hb
Fig. 5 shows the overall configuration. A pair of complementary sec-
ond order filters (High Pass + Low Pass = 1) with flat ‘‘Butterworth
response’’ are utilized. The LPF combined with a derivative of
suspension deflection, plus the High Pass Filter (HPF) combined
with an integrator processing the measured body lateral accelera-
€v m ) together generate a composite lateral damping command
tion (y
and feed into the skyhook damper coefficient, which in turn feeds
into the lateral actuator as the force command. The measured body
lateral acceleration is defined as:
€v m ¼
v2 €v
y  gðh0 þ hv Þ þ y ð15Þ
R
The LPF acts to minimize suspension excursion which is predomi-
nantly low frequency effect, the HPF provides the ride quality
improvement function. Furthermore, the integral of the lateral sec-
ondary suspension deflection is added to the lateral actuator control
loop to keep the lateral suspension deflection within the limit (i.e.
60 mm).
ð14Þ
5.1.2. Integrated tilt and active lateral secondary suspension control
The system configuration for the CD control is illustrated in
Fig. 6. Effective cant deficiency (e.c.d.):

Fig. 5. Complementary filter skyhook damping control.

Fig. 6. Classical decentralized control configuration.

1114 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122

Field of calculation of Field of calculation of

actual ideal actual ideal
RMS | y vm y vm | RMS | vm vm |

1s 3.6 s 1s 3.6 s
Acceleration

Absolute roll
velocity
Lateral

Ideal roll velocity

Actual acceleration

Ideal acceleration Actual roll velocity

Fig. 7. Calculation of dynamics effects on vehicle body lateral acceleration and roll rate for curved track assessments.

€v m
y R.M.S. value of the measured body lateral acceleration. Retuning
hdm ¼ k1  k2 hm
g the lateral actuator control parameters however is necessary
is used to drive the tilt actuator, which inherits the configuration of
the early NT control [3], where hm is the actuator roll angle, y€v m is
the measured body lateral acceleration. k1 and k2 are set to 0.75
and 0.25 respectively for 75% partial tilt compensation. The vehicle
forward speed used in the simulation is 58 m/s. An approximate PID
controller (PI with phase advance, N = 100) is used for the tilting
control.
 
ki kd s
K tilting ¼ kp þ þ ð17Þ
s s=N þ 1
5.1.3. Control parameter tuning
Parameter tuning for the control system can be regarded as
cumbersome task to proceed manually, as it has to meet all the
requirements presented in Section 3. There are three trade-offs be-
tween these criteria which need to be improved:
(1) For the tilting controller design, if the loop bandwidth is low
enough not to interfere with the lateral suspension, it is then
too slow acting on the curve transition [3], which is a critical
problem for the local tilting control system design and intro-
duces a trade-off for the tilting controller between the curv-
ing performance (Pct value) and suspension deflection.
(2) The tilting train operates at higher speed on the same rail
infrastructure compared to the non-tilting train, which dete-
riorates the ride-quality on straight track. It results in a
trade-off for tilting controller performance between the
curved track (Pct value) and the straight track responses
(R.M.S. value).
(3) The trade-off for the lateral actuator controller between the
ride quality on the straight track (R.M.S. value) and the sus-
pension deflection during curving.
The integration of tilt and active lateral secondary suspension
system can help to optimize the first two trade-offs. With appro-
priate system configuration, the third trade-off can be optimized
as well.
– Manually tuned approach The parameters for the lateral actu-
ator control (cut-off frequency (wc), skyhook damping coeffi-
cient (cs), and coefficient for the centering control (kdf)) can be
tuned first to improve the straight track ride-quality without
interfering with the curving suspension deflection. Then, the
tilting controller is designed to minimize the Pct value and
ð16Þ
due to the interaction between these two control loops.
– Genetic Algorithm optimization To further improve the per-
formance, the Non-dominated Sort Genetic Algorithm (NSGA_II)
[19] is used to tune the controller parameters, these being cut-
off frequency (wc), skyhook damping coefficient (cs), coefficient
for the centering control kdf and PID controller parameters.
Objective functions designed for NSGA_II are based on the per-
formance requirements presented in Section 3 and are listed
as follows:
(1) f1 = the Pct value for seated/standing passengers;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
(2) f2 ¼ 1n ni¼1 y €2v m (R.M.S. value of the measured body lateral
acceleration);
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P  actual 
(3) f3 ¼ 1n ni¼1 y €v m  y €ideal
vm (R.M.S. value of the deviation
between actual measured body lateral acceleration with
respect to ideal cases);
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
P 
(4) f4 ¼ 1n ni¼1 h_ actual vm  _ ideal (R.M.S. value of absolute body
h vm
roll rate deviation between measured and ideal responses,
where h_ v m ¼ h_ v þ h_ 0 );
The calculation of f3 and f4 is based in the time interval between
1s before the start of the curve transition and 3.6 s after the end of
the transition [2], as shown in Fig. 7.
Constraint functions for NSGA_II:
(1) w1 = f2 when f2 < 3.24%g (the value in the passive situation:
non tilt with passive suspension system), otherwise,
w1 = 1000. To give a better ride quality on straight track,
the value of w1 with active suspensions should be less than
the value in the passive situation.
(2) w2 < 60 mm (the lateral secondary suspension deflection
(xdefl) should be constrined within the limit);
(3) w3 > 0 dB (tilting controller Gain Margin (with closed lateral
actuator control loop);
(4) w4 > 30deg (tilting controller Phase Margin (with closed lat-
eral actuator control loop);
(5) Constraints on the internal stability of the overall system. It
is guaranteed by checking the poles for a set of internal
transfer functions. More details can be found in Appendix B.
Death penalty function is used for the constraints. The initial
boundaries for NSGA_II are set around the value based on the
results from manual tuning. The GA procedure was simulated for
300 generations with 40 populations. Fig. 8 shows the trade off
 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122 1115

(the R.M.S. value in the passive situation) which is found impossible

4
to be achieved in the Nulling Tilt-only situation [4]. The transition
R.M.S. value of the measured
body lateral acceleration (%g)

3.5 performance is also improved and closer to the tilt-only PT control

3
2.5
2 improvement of the controller performance is limited by the lateral
1.5
1
CD control
0.5
NT control
0
0 2 4 6
Pct value of seated passengers (%)
Fig. 8. Trade off plot from GA tuning.
for the tilting control between system curving performance (Pct va-
lue, f1) and straight track ride quality (R.M.S. value, f2). Note that,
the minimization of f3 and f4 in the optimization process helps to
improve the system curving performance, hence reducing the Pct
value. ‘⁄’ curve in Fig. 8 relates to the NT (tilt-only) control, while
‘o’ is for the proposed CD control. It can be clearly seen that the
trade-off is significantly improved by integrating the lateral actua-
tor control with the tilt control compared to the NT control. In fact,
a number of solutions in ‘o’ curve can meet all requirements pre-
sented in Section 3.
The chosen set of parameters from the GA procedure is:
kp : 0:15; ki : 1:46;
cs : 35720 Ns=m; wi : 3:18 rad=s; kdf : 8930 N=m;
Figs. 9, 10 and Table 3 present the set of simulation results for the
classical scheme. The Pct value for seated passengers is reduced to
8.6%g while keeping the straight track ride quality below 3.24%g
with the peak value of the lateral acceleration is 8.5%g. Fig. 10
shows the efficiency of the integration strategy on straight track,
which reduces the energy in the body acceleration spectrum com-
pared with NT control at frequenies 0  6 Hz. However, further
secondary suspension deflection constraint (60 mm) (referring to
Fig. 9d).
5.2. Command-Driven decentralized dual-actuator control
8 10 12 14 16 The command-driven control is used in the commercial pre-
cedence control, in this strategy, the tilt command is provided
by an accelerometer mounted on the bogie of the local vehicle,
with LPF used to remove unwanted high frequencies due to
the bogie harsh environment, while K = 0.75 to achieve the 75%
compensation of the body lateral acceleration. In this section,
the CDD control is studied for the integrated tilt with active lat-
eral secondary suspension. Local CDD control is designed for the
leading vehicle, while the precedence CDD control is employed
for the trailing vehicle. The complementary filter skyhook damp-
ing control is still employed for the lateral actuator control
which is based on local measurements, but just the leading vehi-
cle needs the centering control. The overall configuration is illus-
trated in Fig. 11.
The following remarks are noted:
(1) A prefileter is used to guarantee a tilt command for 75%
kd : 0:018; compensation.
(2) The cut-off frequency of the LPF2 is set to 0.45 Hz, which
introduces a 0.5 s delay. The preview time is chosen to match
the filter delay. It takes l = (58 m/s  0.5) = 29 m precedence,
approximately 1.5 vehicle length (from leading bogie in prec-
edent vehicle to the center of the trailing vehicle).

1.5 8
CD control
CD control
NT control
Ideal tilt
6 NT control
1 Ideal tilt
4

0.5 2
0
0 −2
−4
−0.5
−6
−1 −8
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200

10 0.06
CD control
NT control 0.04
8 Ideal tilt
0.02
6
0
4
−0.02
2
−0.04
0 −0.06 CD control
NT control
−2 −0.08
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200

Fig. 9. Simulation results for CD control.

1116 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122

Power Spectrum Density wi = 2p  0.3654 rad/s: Cut-off frequency for the complemen-
0.025 tary filters;
Passive (45m/s)
R.M.S. value: 3.24%g cs = 112,053 Ns/m: Skyhook damping coefficient;
0.02 CD control (58m/s)
R.M.S. value: 3.17%g kdf = 268.9 N/m: Centering control coefficient.
NT control (58m/s)
R.M.S. value: 3.39%g
0.015 Control parameters for the trailing vehicle:

0.01 wlpf2 = 2p  0.45 rad/s: Cut-off frequency for the LPF2;

n = 0.707: Damping ratio for the LPF2;
0.005 wi = 2p  0.3 rad/s: Cut-off frequency for the complementary
filters;
0 cs = 140,000 Ns/m: Skyhook damping coefficient.
0 1 2 3 4 5 6

Fig. 10. P.S.D. analysis for the measured body lateral acceleration.
(3) The precedence strategy helps to keep the secondary sus-
pension deflection within the limit. Thus, no centering con-
trol loop for the trailing vehicle is provided (but it can be
included into the system for further reduction of the suspen-
sion deflection).
Control parameters for the leading vehicle:
wlpf1 = 2p  0.7 rad/s: Cut-off frequency for the LPF1;
n = 0.707: Damping ratio for the LPF1;
The parameters above are optimized with the NSGA_II. The set-
up of the optimization is similar to the process in previous section.
The simulation results for both trailing vehicle and leading vehicle
on the specified track are illustrated in Figs. 12 and 13, while the
assessment is presented in Table 4. The performance of the trailing
vehicle has been improved due to the lateral actuator control, the
Pct value of seated passengers is 2.9% being less than the value of PT
control (3.7%). The scheme also provides 55% ride quality improve-
ment on straight track. P.S.D. analysis shows the lateral vibration of
the vehicle body has been significantly attenuated. However, the
performance of the leading vehicle is similar to the performance
of CD control due to the local delay in the signals used for control.
To further improve the performance of the leading vehicle, two ad-
vanced control strategies are studied next.

Table 3
Control system assessment for CD control a 58 (m/s).

Dual-actuator CD Tilt-only NT Tilt-only PT

Deterministic (curved track)
Lateral acceleration. – Steady-state (%g) 4.6 n/a 4.6
– R.M.S. deviation error (%g) 1.9 4.7 0.73
– Peak value (%g) 8.5 13.0 5
Roll gyroscope – R.M.S. deviation (rad/s) 0.021 0.03 0.015
– Peak value (rad/s) 0.110 0.093 0.11
– Peak jerk level (%g/s) 5.1 6.95 3.02
Pct(P-factor) – Standing (% of passengers) 34.2 47.9 20.6
– Seated (% of passengers) 8.63 14 3.7
Stochastic (straight track)
Passenger comfort – R.M.S. passive (%g) 3.24 3.24 3.24
– R.M.S. active (%g) 3.17 3.39 2.29
– Degradation (%) 1.85 5.94 28.6

Fig. 11. Precedence command-driven decentralized control.

 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122 1117

Z Z
T
5.3. MIMO optimal dual-actuator tilt control
x_ ¼ yv hv yb hb y_ v h_v y_ b h_b hr yw y_ w hm h_ m hdm xdefl
Compared with the decentralized control, optimal control al-
lows for direct use of the MIMO state space model thus allowing The weighted outputs selected are body roll rate ðh_ v þ h_ 0 Þ, body tilt
R R
for any couplings in the states during the design process in a cen- angle ðhv Þ; hdm and xdefl . The first two selected outputs are
tralized control design manner. The theory on optimal control is weighted for the fast tilting action. Final value for the output weight
well explained in [20], while it is discussed in the context of tilt- Q and control weight R are optimized with GA and listed below:
 
only control in [21]. For the dual-actuator case, two extra states 1 1 1 1
are included in the state space for disturbance rejection, reference Q ¼ diag ; ; ; ;
0:14192 0:64082 0:00932 0:00662
tracking and decreasing the lateral suspension deflection. These  
1 1
are the integral of the effective cant deficiency and the R ¼ diag ;
R R  integral 4:22992 139602
of lateral secondary suspension deflection hdm ; xdefl . Thus the
augmented system is now of the form: The simulation results are presented in Table 5 and Fig. 14. Optimal
       control has a similar curving performance to the local CDD control,
x_ A 0 x B
¼ þ u however provides a significant ride quality improvement on
x_ 0 C0 0 x0 0 straight track, with R.M.S. value of the measured body lateral accel-
R R T
where x0 ¼ hdm ; xdefl ; u ¼ ½da F a . In particular, the state space eration equal to 1.93%g. The P.S.D. analysis is shown in Fig. 14.
matrices for the vehicle model includes the following states:
5.4. Estimator-based decoupling control

Power Spectrum Density An alternative way to improve the performance of the tilt con-
0.018 troller, rather than utilizing optimal control, is to use the EBD con-
Trailing vehicle
0.016 trol which employs a more effective set of signals. In this context,
Leading vehicle
0.014 R.M.S. value: 3.39%g the signal used for tilt action refers to cant deficiency unaffected by
0.012 the suspension interaction. This could be regarded as mimicking a
0.01 command driven signal, to derive the track only signal
0.008 R.M.S. value: 3.18%g components.
With regards to the signal used for the lateral actuator, this re-
0.006
lates to the estimation of the body lateral acceleration (y €v ) unaf-
0.004
R.M.S. value: 1.45%g fected by the curved track components. Therefore, the lateral
0.002
active suspension can purely deal with the lateral irregularity
0 without increasing the lateral curving deflection, while the tilt ac-
0 1 2 3 4 5 6
tion will be based on true cant deficiency (htdm), unaffected by the
suspension dynamic [4]:
Fig. 12. P.S.D. analysis for the measured body lateral acceleration.

0.8 8
Trailing vehicle Trailing vehicle
Leading vehicle
6 Leading vehicle
0.6 Ideal
Ideal 4
0.4 2
0
0.2
−2
0 −4

−0.2 −6
−8
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200

10 0.04
Trailing vehicle Trailing vehicle
8 Leading vehicle Leading vehicle
Ideal 0.02
6
0
4
−0.02
2

0 −0.04

−2 −0.06
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200

Fig. 13. Simulation results for CDD control.

1118 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122

Table 4
Control system assessment for CDD control a 58 (m/s).

Leading dual-actuator Trailing dual-actuator Trailing tilt-only PT

Deterministic (curved track)
Lateral acceleration. – Steady-state (%g) n/a 4.6 4.6
– R.M.S. deviation error (%g) 4.65 0.4 0.73
– Peak value (%g) 6.8 5.2 5
Roll gyroscope – R.M.S. deviation (rad/s) 0.022 0.018 0.015
– Peak value (rad/s) 0.110 0.106 0.11
– Peak jerk level (%g/s) 5.384 2.42 3.02
Pct(P-factor) – Standing (% of passengers) 31.3 18.85 20.6
– Seated (% of passengers) 7.6 2.94 3.7
Stochastic (straight track)
passenger comfort – R.M.S. passive (%g) 3.24 3.24 3.24
– R.M.S. active (%g) 3.18 1.45 2.29
– Degradation (%) 1.85 55.45 28.6

Table 5
information), body lateral accelerometer (for cant deficiency infor-
Control system assessment for Optimal control a 58 (m/s). mation) and vehicle body yaw gyroscope (required only for extra
information on the curvature R1). The body roll gyroscope mea-
MIMO optimal
sures absolute roll rate ðh_ v þ h_ 0 Þ, thus h_0 must also be included in
Deterministic (curved track) the vector of state estimates. The system state space can be refor-
Lateral acceleration. – Steady-state (%g) 4.6
mulated for the design of the KBF in order to treat parts of distur-
– R.M.S. deviation error (%g) 1.7
– Peak value (%g) 8.3 bance ðh0 h_ 0 R1 Þ as states, and it is given by:
Roll gyroscope – R.M.S. deviation (rad/s) 0.018
– Peak value (rad/s) 0.099 x_ ¼ ½yv hv yb hb y_ v h_ v y_ b h_ b hr yw y_ w hm h_ m h0 h_ 0 R1 T
– Peak jerk level (%g/s) 4.92
Pct (P-factor) – Standing (% of passengers) 31.8
In the design, the process noise is characterized by wk ¼ ½R_ 1 h_ 0 T ,
– Seated (% of passengers) 8.1 while the two inputs are [da Fa]T. The KBF can be designed offline
Stochastic (straight track)
using (19) and (20).
passenger comfort – R.M.S. passive (%g) 3.24
x_ kf ¼ Akf xkf þ Bkf u þ Ck wk ð19Þ
– R.M.S. active (%g) 1.93
– Degradation (%) 40.43 ykf ¼ C kf xkf þ Dkf u þ v ð20Þ

while the state estimates will be calculated by solving the following

differential equation:
Power Spectrum Density
0.018 x_ ¼ Akf b
b x þ Bkf u þ K f ðykf  C kf b
x  Dkf uÞ ð21Þ
Passive
0.016 Optimal control with b x the vector of the state estimates and Kf is the KBF gain ma-
0.014 R.M.S. value:3.24%g
trix. The sensor noise levels are characterized by a covariance ma-
0.012 trix with each diagonal value is set to 1% of the expected
0.01 R.M.S. value 1.93%g maximum value, taken as 3 times the true R.M.S. value of the sensor
0.008 output signal on straight track with irregularities, plus the peak
0.006 value on the pure curved track. Sensor noise covariance and process
0.004 noise covariance are set as,
0.002 R ¼ diagð1:62  103 ; 1:88  106 ; 1  106 Þ
0  
0 1 2 3 4 5 6 1 1
Q ¼ diag 2
; 2
0:1655 12958

Fig. 14. P.S.D. analysis for the measured body lateral acceleration.
The estimation results for the true cant deficiency and estimated
body lateral acceleration are illustrated in Fig. 15. The errors are
small, and mainly due to the noise in the measurements, and the
v2 estimator successfully estimates all necessary quantities even if
htdm ¼  ðh0 þ hv Þ ð18Þ
gR the state vector included disturbance input components.

The required signals cannot be measured by physical devices, and 5.4.2. Controller design
thus a Kalman-Bucy Filter (estimator) is utilized to provide the esti- The estimated body lateral acceleration and true cant deficiency
mated quantities. can facilitate the control design process very effectively. The sky-
hook damping control is adopted for the lateral actuator with the
5.4.1. Kalman-Bucy Filter design estimated body lateral acceleration.
The use of a Kalman-Bucy Filter (KBF) to estimate the true cant s2 1
deficiency only is presented in [21], while it is extended here to cs  HP  1=s ¼ 58000   ð22Þ
s2 þ 2  wi  ns þ w2i s
estimate the signal for lateral actuator control. The design process
is as follows: The inputs to the KBF are the two system inputs and where cs is the skyhook damping coefficient, wi is the cut-off fre-
the three measurements: vehicle body roll gyroscope (cant quency with the value (2p  0.13 rad/s), damping ratio n = 0.707.
 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122 1119

0.3 0.08
Estimated
0.25 Simulated 0.06
Estimation error
0.04
0.2
0.02
0.15
0
0.1
−0.02
0.05
−0.04

0 −0.06 Estimated
Simulated
−0.05 −0.08
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200

Fig. 15. Estimation results on curved track.

Fig. 16. Estimator-based decoupling control.

An approximate PID controller (kp = 0.001, ki = 4.8, kd = 0.07, Note that, the true cant deficiency is configured as (24) for provid-
N = 100) is designed for tilting control with the true cant defi- b b
ing 75% lateral acceleration compensation. R; h0 ; b
h v are estimated
ciency, the overall configuration is shown in Fig. 16. by the KBF.
 
4:8 0:07s v2
fPID ¼ 0:001 þ
s
þ
s=1000 þ 1
ð23Þ htdm ¼ 0:78  ð0:78b
h0 þ b
hv Þ ð24Þ
b
gR

The simulation results are illustrated in Table 6, Figs. 17 and 18. The
Table 6
Control system assessment for EBD control a 58 (m/s). performance of the EBD control is close to the PT control as well as
that of the ideal tilt control case. The R.M.S. deviation error of EBD
Dual-actuator Tilt-only
control for the lateral acceleration is 0.9%g, which is close to the va-
EBD PT
lue of the PT control (0.725%g). Also, the Pct value for the seated pas-
Deterministic (curved track)
sengers (4.91%g) of the EBD control is close to the value for the PT
Lateral – Steady-state (%g) 4.6 4.6
acceleration. – R.M.S. deviation error (%g) 0.9 0.73 control (3.7%g), and 50% ride quality improvement is provided.
– Peak value (%g) 5.6 5 The R.M.S. value of measured body lateral acceleration is redueced
Roll gyroscope – R.M.S. deviation (rad/s) 0.014 0.015 to 1.64%g. Fig. 19 shows the much improved stability margins
– Peak value (rad/s) 0.107 0.11 achieved with EBD control.
– Peak jerk level (%g/s) 3.79 3.02
Pct (P-factor) – Standing (% of passengers) 23.8 20.6
– Seated (% of passengers) 4.91 3.7 6. Discussion
Stochastic (straight track)
Passenger comfort – R.M.S. passive (%g) 3.24 3.24 Fig. 20 gives the curving performance and straight track ride
– R.M.S. active (%g) 1.64 2.29
quality comparison between the 4 proposed new controllers for
– Degradation (%) 49.38 28.6
the dual-actuator system, also for the comparison with the tilt-
1120 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122

8
EBD control EBD control
0.6 PT control 6 PT control
Ideal
0.5 4 Ideal

0.4 2
0.3 0
0.2 −2
0.1 −4
0 −6
−0.1 −8
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200

9 0.02
EBD control
8 PT control 0.01
7 Ideal
0
6
5 −0.01
4 −0.02
3 −0.03
2
−0.04
1 EBD control
0 −0.05
PT control
−1 −0.06
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200

Fig. 17. Simulation results for EBD control.

Power Spectrum Density only PT and NT control approaches. The CDD control provides the
0.018 best performance in trailing vehicles both on curved and straight
EBD control
0.016 Passive track. It shows the potential benefit to use the active lateral sec-
ondary suspension in improving the commercial PT control system
0.014
R.M.S. value: 3.24%g performance. EBD control gives a good ride quality on straight
0.012 track and has a closer Pct value for the seated passengers to the
0.01
PT control. Hence, the EBD approach is a candidate strategy to be
used in the case of leading vehicle performance improvement. CD
0.008 control also provides the improvement to the problem of tilt-only
0.006 R.M.S. value: 1.64%g
NT control performance, this is a simple solution for the implemen-
0.004
tation on the real train however the improvement is constrained.
The simulation based on the end-view model shows the effi-
0.002 ciency of the integration strategy. The controllers can be directly
0 implemented into the full vehicle model (including the yaw dynam-
0 1 2 3 4 5 6
ics) with the same parameters, i.e. separate control system for the
front lateral actuator and rear lateral actuator, but with symmetric
Fig. 18. P.S.D. analysis for the measured body lateral acceleration.
tilting control [22]. Also, modal control approach can be applied to
the two lateral actuators to control the vehicle yaw and lateral
dynamics. It can further improve the straight track ride quality [23].
Nichols Chart
40
7. Conclusion
30

This paper introduced an active suspension integration strategy

Open−Loop Gain (dB)

20 CD control
that combines tilt and active lateral secondary suspension control
10 aiming to enhance the tilt control system performance. A set of
control design approaches for the aforementioned structure is pre-
0 sented and assessed. The simulation results show the efficiency of
CD control the integration strategy as well as that of the designed controllers.
−10 GM: 5.18 dB
PM: 33.8 deg The study demonstrates that, when an active lateral suspension is
***********************
−20 EBD control integrated into a tilting railway vehicle, not only can significant
GM: 14.3 dB ride quality improvement be provided (as has been repeated many
PM: 60 deg
−30 EBD control times before), but also effective tilt control based only upon local
vehicle measurements can be achieved (i.e. no precedence).
−40
−450 −360 −270 −180 −90 0 Further work has included the effects of actuator dynamics, and
Open−Loop Phase (deg) also demonstrated an FPGA-based implementation of the control-
lers via a Hardware-In-the-Loop (HIL) testing with a full vehicle
Fig. 19. Nichols plot for the tilting control. model.
 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122 1121

Pct value forseated passengers (%)

Nulling Tilting (NT) control
16 14
14 PrecedenceTilting (PT) control
12 Classical Decentralised (CD)
10 8.63 control
7.6 8.1
Command-Driven Decentralised
8
(CDD) control (Leading vehicle)
6 4.91
3.7 CDD control (Trailing vehicle)
4 2.94
2 MIMO Optimal control
0 Estimator-Based Decentralised
(EBD) control

R.M.S. value of measured body lateral acceleration

Required to be < 3.24%g
NT control
4 PT control
3.39 3.18 3.17
CD control
3
2.29 CDD control (Leading
1.93 vehicle)
2 1.64
1.45 CDD control (Trailing
vehicle)
1
MIMO Optimal control
0 EBD control

Fig. 20. Controller performance comparison.

Table A.1
Acknowledgement
Pct constants.

The authors would like to thank the anonymous reviewers for Condition A B C D E

their constructive comments and suggestions on the manuscript. Standing passenger 2.8 2.03 11.1 0.185 2.283
Seated passenger 0.88 0.95 5.9 0.120 1.626

Appendix A. Curving transition performance assessment

Pct factor calculation: €v m = maximum vehicle body lateral acceleration calculated

v v
Pct ¼ ðAyv m þ Byv m  CÞ P 0 þ Dðh_ v m ÞE
where A–E are constants defined below Table A.1:
Pct = passenger comfor index on curve transition, representing
the percentage of passenger that will feel discomfort.
y
between the start of curve transition and 1.6 s after the end of
ðA:1Þ
the transition (expressed in %g, shown in Fig. A.1)
v
yv m = maximum lateral jerk level, calculated as the maximum
difference between two subsequent values of y €v m no closer than
1 s, in time interval between 1 s before start of the curve tran-
sition and the end of the transition (%g/s)

Fieldof calculation of y vm

1s 1.6 s
Absolute roll
Acceleration

Field of calculation of
velocity

vm
Lateral

Transition
Transition

Field of calculation of y
v
€v m ; yv m ; h_ v m for Pct factor.
Fig. A.1. Calculation of quantities y
1122 R. Zhou et al. / Mechatronics 21 (2011) 1108–1122

du u ¼ ðI þ KGÞ1 du  KðI þ GKÞ1 dy

y ðC:1Þ
y ¼ GðI þ KGÞ1 du þ ðI þ GKÞ1 dy
K The decentralized control has the controller:

K tilting 0 0
K¼ ðC:2Þ
0 HP  s  cs LP  1=s  cs
System has three outputs: measured body lateral acceleration, sus-
pension deflection and effective cant deficiency. Two inputs are tilt
angle command and lateral actuator force command.
G
References
dy u
[1] Persson R, Goodall RM, Sasak K. Carbody tilting – technologies and benefits.
Fig. C.1. Block diagram used to check internal stability of feedback system. Vehicle Syst Dyn 2009;47:949–81.
[2] Goodall RM, Zolotas AC, Evans J. Assessment of the performance of tilt system
controllers. In: The railway conference at Railtex’ 00, NEC Birmingham (UK);
November 2000.
[3] Goodall RM. Tilting trains and beyond – the future for active railway
h_ v m = maximum absolute body roll velocity calculated from the suspensions: Part 1 Improving passenger comfort. Comput Control Eng J
beginning of the curve transition to the end of curve transition 1990;10(4):153–60.
(deg/s, shown in Fig. A.1). [4] Zolotas AC, Goodall RM, Halikias GD. Recent results in tilt control design and
assessment of high speed railway vehicles. IMECHE J Rail Rapid Trans
2007;221(F2):291–312.
[5] Suescun A. Use of inverse dynamics in the development of tilt control
Appendix B. Notation of the model strategies for rail vehicles. Vehicle Syst Dyn Suppl 1996;25:655–67.
[6] Hauser G. TILTRONIXTM, Anticipative Tilt Control. In: Le rail 4th conference –
yv, yb, y0 Lateral displacement of body, bogie and track Paris; 2006.
hv, hb, da Roll displacement of body, bogie and actuator [7] Zolotas AC, Goodall RM, D Halikias G. New control strategies for tilting trains.
Vehicle Syst Dyn 2002;37:171–82.
h0, R Track cant, curve radius
[8] Goodall RM. Active railway suspensions, Implementation status and
hr Airspring reservoir roll defection technological trends. Vehicle Syst Dyn 1997;28(2):87–117.
yw Lateral displacement of wheel set [9] O’Neill HR, Wale GD. Semi-active suspension improves rail vehicle ride.
hmi Ideal mechanism roll input Comput Control Eng J 1994;5(4):183–8.
[10] Stribersky A, Muller H, Rath B. The development of an integrated suspension
hm ; h_ m Actuator mechanism roll position and rate control technology for passenger trains. IMECHE J Rail Rapid Transit
v Vehicle forward speed 1998;212(1):33–42.
[11] Cheli F, Diana G, Resta F. Numerical model of a tilting body railway vehicle
mv Half body mass, 16,000 kg
compared with rig and on track tests. Vehicle Syst Dyn 2001;35(6):417–42.
mb Bogie mass, 3680 kg [12] Goodall RM, Mei TX. Active suspensions. Handbook of railway vehicle
ivr Half body roll inertia, 20,000 kgm2 dynamics. Taylor and Francis; 2006 [chapter 11].
[13] CEN: 2007, Railway applications-ride comfort for passengers- Measurements
ibr Bogie roll inertia, 2500 kgm2
and evaluation, enquiry version pr EN12299. CEN. Brussels.
kaz Airspring area stiff., 210,500 N/m [14] Pratt I. Active suspension applied to railway trains. Ph.D thesis, Loughborough
ksz Airspring series stiff., 300,000 N/m University of Technology; 1996.
krz Airspring reserv. stiff., 201,000 N/m [15] Md Yusof H, Goodall RM, Dixon R. Assessment of actuator requirements for
active railway suspensions. In: Proceedings of the 5th IFAC symposium on
crz Airspring reserv. damp., 20,000 (Ns)/m mechatronic systems, marriott Boston, Cambridge Cambridge, MA, USA;
ksy Secondary lateral stiff., 100,000 N/m September 13–15, 2010.
kvr Anti-roll bar stiff./bogie, 1,500,000 (Nm)/rad [16] Zolotas AC. Advanced control strategies for tilting trains. PhD thesis,
Loughborough University of Technology; 2002.
cvr Anti-roll bar damp./bogie, 18,200 (Ns)/m [17] Li H, Goodall RM. Linear and non-linear skyhook damping control laws for
kpz Primary vertical stiff., 1,600,000 N/m active railway suspensions. Control Eng Practice 1999;7(7):843–50.
cpz Primary vertical damp., 20,000 (Ns)/m [18] Streiter R, Boller M, Schneider R. Active lateral suspension for high speed trains
– a step towards the mechatronic Bogie; 2001.
kpy Primary lateral stiff., 18,000,000 N/m [19] Srinivas N, Deb Kalyanmoy. Multiobjective optimization using nondominated
cpy Primary lateral damp., 20,000 (Ns)/m sorting in genetic algorithms. Evol Comput 1994;2:221–48.
d1 Airspring semi-spacing, 0.835 m [20] Skogestad S, Postlethwaite I. Multivariable feedback control: analysis and
design. Wiley; 2000 [Reprinted Version].
d2 Prim. vert. suspen. semi-spacing, 1.00 m [21] Zolotas AC, Goodall RM. Improving the tilt control performance of high speed
h1 2ndary later. suspen. height (body cog), 0.844 m railway vehicles: an LQG approach. In: Proceedings of the 16th IFAC world
h2 2ndary later. suspen. height (bogie cog), 0.252 m congress. Prague; 2005.
[22] Pearson JT, Goodall RM, Pratt I. Control system studies of an active anti-roll
h3 Primary later. suspen. height (bogie cog),
bar tilt system for railway vehicles. IMECHE J Rail Rapid Transit (F1)
0.194 m 1998:43–60.
hg1 Bogie cog height (rail level), 1.696 m [23] Zhou R, Zolotas AC, Goodall RM. 9 DOF Railway vehicle modeling and control
hg2 Body cog height (rail level), 0.60 m for the integrated tilting bolster with active lateral secondary suspension.
UKACC International Conference on Control, Coventry, UK, September; 2010.
hmt Mechanism c.o.g. vertical separation from p. 1280–1285.
effective tilt center, 0.60 m

Appendix C. Internal stability analysis

The feedback system in Fig. C.1 is internally stable if and only if

all four closed-loop transfer matrices in (C.1) are stable [18].