University of Reading
Control of an Electromagnetic
Vehicle Suspension
A thesis submitted for the degree of Doctor of Philosophy
by
Neil Stuart McLagan
Department of Engineering
June 1992
Control of an Electromagnetic
Vehicle Suspension
i
Control of an electromagnetic vehicle suspension
This dissertation describes the analysis of an electromagnetic vehicle
suspension and the proposal and synthesis of a sophisticated suspension
control system. The difficulties associated with the control of
electromagnetic suspension providing both primary and secondary
suspension functionality are first discussed in the light of existing
research and development progress. The strengths and weaknesses of
existing control techniques are then identified, and a structured control
strategy is proposed. This in turn involves a new, nonlinear
electromagnet force control algorithm which employs air gap and
current feedback, and a sophisticated suspension control algorithm
which consists of an absolute position controller, with a position
reference supplied by a guideway following algorithm. Three feedback
states are measured for each electromagnet, namely air gap, current,
and acceleration. The suspension control algorithm is applied to the
vehicle heave, pitch, roll and torsion motions independently. The
resultant electromagnet force demands are fed into force controllers
which provide dominantly linear and independent electromagnet force
actuation. An experimental control system is developed using
transputers, and the control algorithms are implemented using the occam
parallel programming language. The proposed control theory is
validated by presenting both simulation results, and responses from the
experimental system. The results clearly show the efficacy of the
proposed control method.
ii
Acknowledgements
I am very grateful to my supervisor, Pradip Sinha, for igniting my interest in the
problems of controlling electromagnetic suspensions, and also for giving me the
opportunity to carry out the research described in this dissertation.
I would like to thank my friends and family for their encouragement and support. I am
particularly indebted to Greg Pye, Nigel Kneebone and Jeremy Hinton for the many
hours spent discussing various issues and reviewing this dissertation. Last, but not
least, I would like to thank Gail Tucker for her patience, tolerance and help.
The research and equipment described in this dissertation was funded by the Science
and Engineering Research Council.
iii
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Wheels and bearings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Controlled d.c. electromagnetic suspension . . . . . . . . . . . . . . . . . . . 2
1.3 Vehicle suspension configuration . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Review of electromagnetically suspended vehicles . . . . . . . . . . . . . 4
1.5 Anatomy of an electromagnetic vehicle suspension . . . . . . . . . . . . . 7
1.5.1 Suspension force actuation . . . . . . . . . . . . . . . . . . . . . . . 8
1.5.2 Decoupling the electromagnet motions . . . . . . . . . . . . . . . 9
1.5.3 Control of the vehicle mode motions . . . . . . . . . . . . . . . 10
1.6 Proposed vehicle suspension control strategy . . . . . . . . . . . . . . . . 12
1.7 Proposed implementation strategy . . . . . . . . . . . . . . . . . . . . . . . . 14
1.8 Direction and scope of this research . . . . . . . . . . . . . . . . . . . . . . 15
2 Electromagnet analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Electromagnet geometry and specification . . . . . . . . . . . . . . . . . . 18
2.3 Steady-state analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Magnetic force characteristic . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Lift and lateral force components . . . . . . . . . . . . . . . . . . 22
2.3.3 Air gap reluctance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.4 Iron path reluctance . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.5 Leakage path reluctance . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.6 Steady-state model equations . . . . . . . . . . . . . . . . . . . . . 26
2.3.7 Accuracy of the steady-state model . . . . . . . . . . . . . . . . 29
2.4 Dynamic analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.1 Magnetic force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.2 Magnetic flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4.3 Coil current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4.4 Eddy currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.5 Magnetic hysteresis . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4.6 Accuracy of the dynamic model . . . . . . . . . . . . . . . . . . . 38
2.5 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
iv
3 Electromagnet force control . . . . . . . . . . . . . . . . . . . . . . . . 423.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Operational envelope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.3 Electromagnet transfer function . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4 Force control strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4.1 Force feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4.2 Flux feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4.3 Acceleration feedback . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4.4 Air gap and current feedback . . . . . . . . . . . . . . . . . . . . . 52
3.4.5 Proposed force control strategy . . . . . . . . . . . . . . . . . . . 54
3.5 Force controller design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.6 Force controller performance . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4 Suspension mode control . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Functional requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3 Suspension control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.4 Synthesis of the suspension control system . . . . . . . . . . . . . . . . . 68
4.4.1 Position control transfer function . . . . . . . . . . . . . . . . . . 69
4.4.2 Position, velocity and acceleration feedback gain . . . . . . 71
4.4.3 Force actuation bandwidth . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.4 Position error integral time constant . . . . . . . . . . . . . . . . 73
4.4.5 Guideway following algorithm . . . . . . . . . . . . . . . . . . . . 73
4.4.6 State integration filters . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4.7 Suspension controller design specification . . . . . . . . . . . . 76
4.5 Lateral guidance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.6 Performance of the experimental mode suspension . . . . . . . . . . . . 79
4.6.1 Position controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.6.2 Full suspension system . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
v
5 Vehicle suspension control . . . . . . . . . . . . . . . . . . . . . . . . . 885.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.2 The experimental research vehicle and guideway . . . . . . . . . . . . . 88
5.3 Control strategy for the vehicle suspension . . . . . . . . . . . . . . . . . 90
5.4 Synthesis of the vehicle control system . . . . . . . . . . . . . . . . . . . . 92
5.4.1 Decoupling the electromagnet motions . . . . . . . . . . . . . . 93
5.4.2 Normalising the vehicle mode motions . . . . . . . . . . . . . . 94
5.4.3 Control of the vehicle mode motions . . . . . . . . . . . . . . . 95
5.4.4 Configuration of the vehicle mode suspension controllers . 95
5.5 Lateral vehicle guidance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.6 Performance of the experimental vehicle suspension . . . . . . . . . . . 99
5.6.1 Performance and stability of the mode position
controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.6.2 Decoupling of the vehicle modes and load sharing . . . . . . 104
5.6.3 Ride quality of the vehicle suspension . . . . . . . . . . . . . . 107
5.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6 Control system implementation . . . . . . . . . . . . . . . . . . . . . 1116.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.2 System requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.2.1 Bandwidths and sampling rates . . . . . . . . . . . . . . . . . . . 112
6.2.2 Range, resolution and accuracy . . . . . . . . . . . . . . . . . . . 115
6.3 Transducers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.3.1 Accelerometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.3.2 Air gap sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.3.3 Electromagnet current controller . . . . . . . . . . . . . . . . . . . 120
6.4 Signal processing, conversion and communication . . . . . . . . . . . . . 123
6.4.1 Transputers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.4.2 Control system hardware structure . . . . . . . . . . . . . . . . . 125
6.4.3 Analogue signal conditioning and conversion . . . . . . . . . 126
6.4.4 Digital signal processing and communication . . . . . . . . . . 128
6.5 Software design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.5.1 Occam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.5.2 Control system software structure . . . . . . . . . . . . . . . . . . 130
6.5.3 Discrete time domain integration . . . . . . . . . . . . . . . . . . 134
6.5.4 Numerical accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.5.5 Process configuration . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
vi
Appendices
A Electromagnet analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 151
B Electromagnet force control . . . . . . . . . . . . . . . . . . . . . . . 168
C Suspension mode control . . . . . . . . . . . . . . . . . . . . . . . . . 173
D Control system hardware . . . . . . . . . . . . . . . . . . . . . . . . . 178
E Control system software . . . . . . . . . . . . . . . . . . . . . . . . . . 220
F Operating instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
Introduction 1
1
Introduction
1.1 Wheels and bearings
The wheel is without doubt one of man’s most impressive early inventions. The
important difference between a wheeled cart and its predecessor, a sledge, lies in the
arrangement and quality of the bearing surfaces. The Concise Oxford English
Dictionary defines a bearing as: ‘a carrier or support for moving parts of any machine;
any part of the machine that bears the friction’. Even a crude wheeled cart has
relatively smooth and small bearing surfaces whereas a sledge has much larger bearings,
one of which (the ground) can be quite rough. The linear motion bearing between the
sledge and the ground, was thus transformed through the use of wheels, to a rotary
bearing between two controlled surfaces.
For the majority of applications the advantages of the mechanical transformation from
linear to rotary bearings have stood the test of time well. However, by harnessing
magnetic forces to support a moving body, a bearing with no physical contact between
its surfaces is possible. The lack of physical contact offers superior performance over
mechanical bearings in terms of friction and wear. For certain wheel-on-rail transport
applications, the benefits can be even greater, since magnetic linear bearings can be
used to replace both the wheel and its rotary bearing. For such applications, the wheel
may in future become as unusual as a horse-drawn carriage is today.
There are a number of different electromagnetic methods for supporting moving or
rotating masses1 (see Table 1.1). The attraction schemes are conventionally referred
to as electromagnetic suspension (EMS), whilst repulsion schemes are referred to as
electrodynamic levitation (EDL). A comprehensive review of the various EDL and
EMS schemes and their development potential can be found in reviews by Jayawant,2
Sinha3 and Weh.4 This dissertation describes the development of new, improved
techniques for controlling d.c. electromagnets for vehicle suspension applications.
Introduction 2
Table 1.1 Electromagnetic methods of supporting moving or rotating masses
Levitation using:
• forces of repulsion between permanent magnets.
• forces of repulsion between diamagnetic materials.
• superconducting magnets.
• forces of repulsion due to eddy currents induced in a conducting surface or body.
• force which acts on a current-carrying conductor in a magnetic field.
• mixed µ system, where µ is the permeability of the material.
Suspension using:
• a tuned LCR circuit and the magnetic force of attraction between an electromagnet
and a ferromagnetic body.
• controlled d.c. electromagnets and the force of attraction between magnetised bodies.
1.2 Controlled d.c. electromagnetic suspension
The force of attraction between two magnetised bodies is proportional to the inverse
square power of their separation, thus there is no point of equilibrium between two
magnetised bodies (Earnshaw’s theorem5). The force between an electromagnet and
its reaction rail is therefore open-loop unstable and closed-loop feedback control of the
electromagnet is necessary to stabilise the force and provide a satisfactory suspension
response. The essential elements of an EMS system, therefore, consist of an
electromagnet and its ferromagnetic reaction rail (see Figure 1.1), feedback sensor(s)
and associated control algorithm processing, and finally a current controller for the
electromagnet.
Graeminger appears to have been the first to propose a controlled electromagnetic
attraction system6 in 1912. He proposed a U-shaped electromagnet suspended beneath
an iron rail to carry letters. A measure of the air gap between the electromagnet and
the track was coupled mechanically to a rheostat which varied the electromagnet coil
current.
Twenty-five years later, Kemper built the first prototype EMS7 which supported 210 kg
at an air gap of 15 mm with a power consumption of 270 W. A capacitive
displacement sensor was used to measure the air gap. Thermionic valves were used to
amplify the air gap signal and a velocity signal, and also to drive the electromagnet
coil.
Introduction 3
However, the weight of the thermionic valve power controllers used to implement
Figure 1.1 Configuration of electromagnet and reaction
rail
Kemper’s EMS precluded their use in transport applications. It was after 1970, with
the advent of transistor technology capable of handling suitably high power levels, that
research into the use of EMS for transport applications flourished. Before reviewing
the more significant developments in vehicular EMS systems, it is instructive to
consider the configuration of the functional components of a conventional train
suspension.
1.3 Vehicle suspension configuration
Conventional trains are supported by a primary suspension which is coupled via a
secondary suspension to the vehicle chassis. The function of the primary suspension
(typically stiffly sprung wheels), is to maintain contact with the track, and hence avoid
derailment. The function of the secondary suspension is to provide a low bandwidth
coupling between the primary suspension and the vehicle, thus decoupling the vehicle
from high frequency track irregularities. The secondary suspension employs suitable
stiffness and damping components so that an acceptable passenger ride quality is
maintained for a given train speed and track profile. The secondary suspension also
reduces the wear and tear on the track and the primary and secondary suspensions.
This is because the decoupling of the vehicle mass from the wheels reduces the
dynamic forces generated by track irregularities.
Introduction 4
The two essential elements of a conventional train suspension are therefore the primary
suspension, which can be viewed as a high bandwidth track follower, and the secondary
suspension, which acts as a low pass filter with a low bandwidth.
Kemper’s approach of using air gap clearance and velocity feedback provides the
electrical equivalent of mechanical stiffness and damping respectively. The controlled
electromagnet, therefore, behaves in a similar manner to a conventional suspension with
the stiffness determined by the sum of the air gap feedback gain and the negative
electromagnet stiffness, and the damping is determined by the velocity feedback gain.
In principle, a conventional secondary suspension design could be utilised, but with an
electromagnet replacing the wheel. The feedback gains of the electromagnet controller
are designed to give the appropriate primary suspension stiffness and damping.
Alternatively, the electromagnet could be used to replace the secondary suspension by
setting the feedback gains to give the stiffness and damping required for the secondary
suspension. A primary suspension is not required in this case due to the lack of
physical contact between the secondary suspension and the rail. This arrangement
eliminates all moving parts and associated maintenance requirements from the vehicle
suspension, plus all of the size and weight associated with wheels, axles, bogies, springs
and dampers. Whether this trade of a mechanical for an electromagnetic systems pays
off obviously depends on the cost, weight, size and maintenance requirements of the
electromagnet and its associated control circuitry. In addition, Hrovat8 has shown that
reducing a vehicle’s unsprung weight (that of the primary suspension) enables the ride
quality to be improved. Since an EMS incorporating both primary and secondary
suspension would have no unsprung weight, ride comfort quality above that of
mechanical suspensions is theoretically possible.
The two basic suspension configurations for EMS systems are therefore electromagnetic
primary with conventional secondary suspension, or electromagnetic primary plus
secondary suspension, the latter scheme offering a suspension with no moving parts.
1.4 Review of electromagnetically suspended vehicles
Table 1.2 lists some of the milestones in the development of electromagnetic vehicle
suspensions. The first vehicles developed incorporated both primary and secondary
Introduction 5
suspension in the EMS. Subsequently however, the high speed systems and some low
speed systems have reverted to using conventional secondary suspensions.
The Magnetmobil9 was the first full-scale system capable of carrying passengers. It
Table 1.2 Electromagnetically suspended vehicles
Organisation Vehicle DateWeight
/t
Speed
/kphCSS Propulsion
MBB, FRG Magnetmobil 1971 7 90 x DLIM
Krauss-Maffei, FRG Transrapid-02 1972 11 165 x LIM
Rohr Industries, USA Romag IMPS 1972 1 low x LIM *
University of Sussex, UK Sussex 1t 1974 1 low x LIM
Japan Airlines, Japan HSST-01 1975 1 300 x LIM
British Rail, UK BR 3t 1976 3 low x LIM
Transrapid Consort.,FRG Transrapid-06 1981 122 400 LSM *
PMG Consortium, UK PMG 8t 1984 8 48 x LIM
Timisoara Poly., ROM Magnibus-01 1988 4 72 LIM *
HSST Consortium, Japan HSST-05 1989 - 200 LIM
Key: CSS - Conventional Secondary Suspension
DLIM - Double-sided Linear Induction Motor
LIM - Single-sided Linear Induction Motor
LSM - Linear Synchronous Motor
* - Suspension is combined with propulsion system
used a pair of orthogonally orientated electromagnets at each corner to provide
independent lift and lateral suspension. Propulsion was provided by a double sided,
short stator linear induction motor mounted on the vehicle, with an aluminium reaction
rail on the guideway. The Transrapid-0210 vehicle, used two in-line electromagnets
at each corner which were slightly offset either side of the rail. Both electromagnets
contributed lift force, but a lateral force component was also generated by adjusting the
relative drive levels between the electromagnet pair. The propulsion configuration was
the same as the Magnetmobil except for the use of a single-sided linear induction
motor. The Romag system11 took a different approach, using one linear induction
motor at each corner to provide combined lift and propulsion, but there was no active
control of lateral motion. The University of Sussex12 vehicle, the BR13 and PMG14
vehicles and the HSST-0115 used essentially the same configuration as the
Introduction 6
Transrapid-02, incorporating the Krauss-Maffei offset electromagnet arrangement for
lateral guidance.
For practical reasons, the nominal operational air gap of EMS systems is limited to
about 15-20 mm, above which the weight and power consumption of the electromagnets
become excessive.16 For high speed systems, meeting ride comfort requirements with
such a small allowable air gap deviation would require excessively expensive track
construction and alignment maintenance. As a consequence, practical high speed EMS
systems need to use an electromagnetic primary suspension with a conventional
secondary suspension capable of a much larger suspension deflection.17
The Transrapid-0618,19 system uses a continuous array of ‘magnetic wheels’ down the
full length of both sides of each coach. Each magnetic wheel is coupled to the coach
via an air-spring secondary suspension and operates autonomously offering a highly
modular system with significant redundancy and hence robustness to individual module
failure. The lift and propulsion are both provided by a long stator (active track) linear
synchronous motor.20 The lift force is controlled by varying the effective resistance
of coils on the lift/propulsion ‘rotor’. Lateral guidance is provided independently of
lift/propulsion in each magnetic wheel by controlled d.c. electromagnets mounted
orthogonally with respect to the lift units. The problem of non-contacting power
collection for propulsion at high speed was overcome through the use of an active track.
The power required for the onboard controllers and general vehicle services is provided
by linear synchronous generators which are incorporated into each lift/propulsion unit.
All other vehicles use power rails with sliding shoes for power collection. The
Magnibus-0121, a low speed system, appears to be functionally equivalent to the
Romag system, but with the addition of conventional secondary suspension. The HSST-
0522,23 system is the latest development of the HSST-01 and now employs a
conventional mechanical secondary suspension.
In addition to passenger transport applications, EMS based materials transportation
systems for use in automated factory production lines have been researched in
Japan.24,25 These systems typically use a small vehicle weighing about 10 kg to carry
a load of just over 10 kg. Propulsion is provided by a number of short stator linear
induction motors distributed suitably along the guideway with an aluminium reaction
plate mounted on the vehicle. Such systems use onboard battery power with special
recharging stations to reduce power collection problems. To maximise the operational
time between recharges, hybrid magnets are used consisting of permanent magnets plus
control coils. The suspension controllers are designed to operate the hybrid magnets
Introduction 7
at an air gap which requires zero average current rather than at some fixed nominal air
gap.
Maximising the benefits of the non-contacting nature of electromagnetic suspension
requires the other vehicle systems to use a non-contacting technique. Whilst linear
induction or synchronous motors provide an appropriate propulsion technique, power
collection and route switching remain practical problems. Power collection from an
active track as employed by the Transrapid system is likely to prove too expensive for
low speed applications. Route switching is also a problem because a gap must be
introduced into a suspension rail at a junction to allow an electromagnet plus its support
structure to cross the rail. The suspension force is therefore lost as the electromagnet
traverses the rail gap. This problem can be overcome by using a duplicate set of
electromagnets with a suitable arrangement of duplicate rails at the junction. However,
such a solution is undesirable due to the resulting poor suspension utilisation and the
increase in weight. A novel solution was proposed by Jayawant and Wheeler26 where
two sets of magnet pole faces were connected to a single electromagnet, but this still
incurred a significant weight penalty over conventional electromagnets. An acceptable
configuration for low cost, non-contacting power collection, and route switching without
track movement or contact has yet to be established.
Having reviewed the general configuration of some representative electromagnetically
suspended vehicles, the detailed structure of an electromagnetic vehicle suspension
system is examined next.
1.5 Anatomy of an electromagnetic vehicle suspension
The anatomy of an electromagnetic vehicle suspension is largely determined by the
functional requirements of the suspension system. The main functional requirement is
to decouple the passengers from guideway irregularities whilst following the general
guideway profile. In addition, external disturbance forces such as wind gusts must be
resisted, and passenger load variations must be accommodated. The forces generated
by the electromagnets suspending an EMS vehicle must therefore be controlled to meet
these requirements. The suspension parameters depend on factors such as guideway
profile, guideway stiffness and natural frequency, operational air gap range, disturbance
forces, passenger load variations and the required level of passenger comfort.
Introduction 8
An electromagnetic vehicle suspension consists of three important components. Firstly,
a set of electromagnets is required to provide force actuation to the vehicle body.
Secondly, a technique for decoupling the electromagnet motions is required, and finally,
suspension control algorithms are required for each of the decoupled motions. Each of
these components will now be considered.
1.5.1 Suspension force actuation
The number and configuration of the electromagnets required to suspend and guide a
vehicle depends mainly on the number of degrees of freedom to be controlled. Practical
factors such as vehicle shape and the required electromagnet redundancy (needed to
provide system availability under partial failure conditions) also contribute to the
vehicle configuration.
A vehicle assumed to behave like a perfectly rigid body in free space is capable of
linear motion and rotation with respect to three orthogonal axes. Convenient horizontal
reference axes for a tracked vehicle are the longitudinal and lateral axes of the
guideway, with the third orthogonal axis being vertical. The linear motion of the
vehicle along the guideway is controlled by the propulsion system, thus leaving five
degrees of freedom to be controlled by the vehicle suspension system. The vehicle
mode motions, which are conventionally referred to as heave and sway (vertical and
lateral motions), and pitch, roll and yaw (lateral, longitudinal and vertical axis
rotations), are illustrated in Figure 1.2.
Figure 1.2 Vehicle mode motions
Introduction 9
In reality, vehicle bodies are not perfectly rigid, and so additional degrees of freedom
exist. These correspond to the various linear and torsional bending motions which can
occur due to vehicle flexibility. Complete and independent control of the motion of a
vehicle body thus requires five independent electromagnet actuators for the rigid body
motions, with additional actuators needed if control of vehicle bending is required.
For practical reasons, most EMS vehicles have used eight electromagnets, with two
located at each corner of the vehicle to provide lift and lateral forces. Due to
redundancy in the configuration, the four lateral forces produce only two independent
vehicle mode forces/torques, namely sway and yaw. The four lift forces produce four
independent vehicle mode forces/torques, namely heave, pitch, roll and torsion (vehicle
body twist axial to its length).
1.5.2 Decoupling the electromagnet motions
The simplest vehicle suspension control strategy would be to use independent
suspension controllers, with identical parameters for each electromagnet. With this lift
control configuration, the resulting heave, pitch, roll and torsion motions of the vehicle
would all experience the same controller parameters. Unfortunately, for a vehicle with
electromagnets mounted directly on the chassis, the resultant stiff coupling between the
electromagnets results in the independent control configuration being generally
unacceptable. This is because the high controller stiffness and zero steady-state air gap
error required for the vehicle heave mode also applies to the other vehicle modes.
When zero torsion error cannot be achieved, for example due to normal track/vehicle
misalignment, the vehicle would be largely supported by a diagonal pair of
electromagnets, with the other electromagnet pair sitting virtually idle. This poor load
force distribution would cause two of the electromagnets to be overloaded. Independent
lateral electromagnet control would also produce equal suspension parameters for the
yaw and sway motions. This may be acceptable since the lateral motions are largely
decoupled. For general ride comfort considerations, it may also be desirable to have
different settings for the heave, pitch and roll mode suspension controllers.
If a conventional secondary suspension is used to couple the electromagnets to the
vehicle, then the low stiffness of the secondary suspension largely decouples the high
stiffness primary suspension electromagnets from the vehicle and hence from each
other. In this case, the electromagnets can be controlled independently, and autonomous
‘magnetic wheel’ modules can be employed as exemplified by the Transrapid-06
Introduction 10
vehicle. The ride comfort characteristics of the vehicle’s heave, pitch, roll, sway and
yaw motions are then determined by the conventional secondary suspension.
For vehicles employing electromagnets for secondary suspension, the direct attachment
of the electromagnets to the vehicle chassis results in a tightly coupled, multivariable
system. In this case, the independent electromagnet control scheme is unacceptable for
the reasons given earlier. A multivariable controller is therefore required which must
decouple the electromagnet motions (eg. by transforming them to independent vehicle
motions) and apply independent suspension controllers to each decoupled mode.
Multivariable control of the vehicle system is complicated due to the nonlinear and
unstable nature of the electromagnet force characteristic. As a consequence, all of the
vehicles listed in Table 1.2 that employed electromagnets for secondary suspension used
independent electromagnet force stabilisation controllers. The controllers significantly
reduced the force instability by using feedback of the derivative of the air gap flux for
each electromagnet. The multivariable control schemes used linear decoupling to
transform between the electromagnet motions and the vehicle mode motions.
Independent suspension controllers were then applied to each vehicle mode motion.
Results from Sinha and Jayawant27 showed that the multivariable control scheme could
achieve a superior control performance relative to an independent electromagnet
suspension control scheme. However, the nonlinear electromagnet force characteristic
resulted in these schemes suffering from significant cross-coupling between the heave,
pitch, roll and torsion modes which impaired the dynamic response of the vehicle
suspension. In addition, the elimination of steady-state pitch and roll offsets through
the use of error-integral feedback action could not be achieved on the Sussex and PMG
vehicles. This was due to low frequency cross-coupling problems between the vehicle
heave, pitch and roll modes.
1.5.3 Control of the vehicle mode motions
Having examined the configuration of the vehicle suspension control system, the
independent vehicle mode controllers that are applied to the decoupled motions can now
be considered. It is these controllers that must achieve the main functional requirements
of the vehicle suspension system outlined earlier.
The early EMS vehicles used air gap feedback to provide stiffness relative to the rail
and absolute velocity feedback to provide damping. The stiffness had to be high in
Introduction 11
order to counter load variations and disturbance forces within the small available
operational air gap range. The high stiffness resulted in a suspension bandwidth of
around 6 Hz, which was too high to meet ride comfort specifications for a cost effective
guideway. The PMG vehicle overcame this problem by using air gap stiffness at low
frequencies (below about 1.5 Hz), with vehicle position stiffness used for higher
frequencies. This was achieved through the use of a complementary pair of low and
high pass filters on the air gap and position feedback signals respectively. The
suspension thus provided a low frequency coupling to the guideway with a high
absolute stiffness to load variations and disturbance forces. Damping was provided by
applying phase-lead compensation to the complementary stiffness signal. The absolute
velocity and position signals were obtained by integration and double integration
respectively of the output from an accelerometer mounted near each electromagnet. The
integrators were given a low frequency cutoff to prevent drift problems. Acceleration
feedback was also employed in an attempt to reduce the nonlinearity of the
multivariable system.
For a guideway without gradients, a simple two pole filter design was satisfactory for
the complementary filters. However, the gradient entry and exit characteristic required
for the PMG guideway caused unacceptable air gap deviations when using two pole
filters. The final design used two and three pole filters to provide a compromise
between ride comfort and air gap deviation at guideway gradients.
The parameters of the vehicle guideway can have a critical influence on the steady-state
and dynamic behaviour of the vehicle suspension due to the coupling between the
vehicle and the guideway. If the static deflection of the guideway due to the weight
of the vehicle is to be accommodated without disturbing the passengers, then the
guideway deflection must be less than the operational range of the secondary
suspension. For vehicles with an electromagnetic secondary suspension, the small
operational air gap range requires a much stiffer guideway than that required for
vehicles employing a conventional secondary suspension. The high guideway stiffness
generally simplifies suspension design by enabling the track to be assumed to be rigid.
However, to avoid resonant oscillations, an adequate margin between the natural
frequencies of the various system components must be ensured. For example, the stiff
track required on the elevated concrete guideway for the PMG vehicle was required to
have a natural frequency above 10 Hz, to give adequate separation from the
suspension-guideway coupling bandwidth of about 1.5 Hz and the force rejection
bandwidth of about 6 Hz.
Introduction 12
1.6 Proposed vehicle suspension control strategy
The objective of the research described in this dissertation is to improve the
performance of electromagnetic secondary suspension for vehicles through the use of
improved control techniques.
The design of the electromagnetic suspension scheme for the PMG vehicle is the most
sophisticated of those employed for electromagnetic secondary suspension. However,
it has two areas of weakness. The first weakness is due to the nonlinear electromagnet
force actuation which causes cross-coupling between the independent vehicle mode
motions. This impairs the dynamic response of the suspension and prevents the use of
self-levelling roll and pitch mode controllers. The inaccurate nominal air gaps which
result from the lack of a self-levelling response reduce the allowable air gap deviation,
and hence give poor electromagnet utilisation.
To overcome this problem, a novel force control algorithm is proposed which is capable
of providing a sufficiently linear and stable force actuation. First, a detailed nonlinear
model of the electromagnet force characteristic is developed. The proposed control
algorithm then employs the model to determine the appropriate electromagnet excitation
for any required operating point.
The second weakness of the PMG vehicle is structural and stems from the use of a
single control block for the vehicle mode controllers. This enables the disturbance force
rejection characteristic to be freely chosen, but the guideway coupling for flat
guideways and guideways with gradients are both determined by the air gap feedback
filter. The implementation of the guideway interaction functions is thus tightly coupled
and the resultant performance of guideway following is therefore compromised. For
example, it may be possible to improve the vehicle guidance at the entry and exit of
curves by employing a matched filter technique. The matched filter could use the
functions defining the guideway curves to actively identify curves and guide the vehicle
appropriately.
The proposed solution to this structural problem is to partition the vehicle mode control
algorithm into two independent blocks. The first block is fed with the guideway
position which it processes using a suitable guideway following algorithm to produce
a vehicle position demand. This is then fed as the reference into a vehicle position
control algorithm which employs position error integral feedback to eliminate
steady-state position errors, and high stiffness in order to resist load variations and
Introduction 13
disturbance forces. The force demands from the vehicle position controller are then
sent to the electromagnet force controllers. Figure 1.3 outlines the structure of the
proposed vehicle control scheme.
If active guideway damping is also required,28 then assuming linear superposition, the
proposed system could be augmented with another independent control block. This
would receive the guideway velocity and, using a suitable algorithm, determine the
required damping force demand to be added to the force demands from the vehicle
position controller. Such a configuration would require a detailed analysis of the
coupling between the various control blocks since linear superposition of the vehicle
position control action and the track damping action is likely to be impaired by
nonlinearities within the vehicle suspension and guideway.
Having outlined the structure of the proposed vehicle control strategy, the design must
Figure 1.3 Structure of the proposed vehicle control
system
be developed and validated. For the vehicle suspension system, the features which are
difficult to model accurately are critical to the performance of the overall system. The
modelling difficulties are attributed to effects such as the nonlinearity and higher order
characteristics of the electromagnets, the vehicle chassis and the track. Proof of concept
using simulation as a validation tool is therefore considered to be inappropriate for the
Introduction 14
vehicle suspension system. The development of an experimental system (using
simulation as a design tool) is therefore considered to be necessary to facilitate
validation of the proposed vehicle control strategy.
1.7 Proposed implementation strategy
Three principle options are available for the hardware implementation of the suspension
controller for the experimental vehicle. These options are, analogue electronic circuitry,
programmable digital processors, or a combination of both analogue and digital
hardware. In order to determine the best implementation strategy, each of the control
system components must be considered.
For the electromagnet force control algorithm, analogue electronic circuitry is
impractical due to the complexity of the nonlinear electromagnet model which is
embodied within the algorithm. The use of a programmable digital processor is
therefore required for the force controller. The vehicle mode decoupling algorithms and
mode position controllers are linear and can thus be readily implemented using either
analogue or digital techniques. For the final control system component, the guideway
following algorithm, a programmable processor implementation is required to ensure
maximum flexibility in the choice of algorithm. In addition to the signal processing
requirements of the control system, executive control of the vehicle in terms of startup,
shutdown, fault detection and data monitoring is needed, and this is more readily and
flexibly achieved through the use of a programmable controller. The only system
functions that could feasibly use analogue signal processing are the vehicle mode
decoupling and the vehicle mode position control. Since these functions are logically
located between components that need digital signal processing, additional
analogue/digital conversions would be required if analogue signal processing were
employed. Therefore, to simplify the system hardware configuration, eliminate the
problems of drift and offsets, and maximise the flexibility of the implementation, the
use of digital processing for all system functions is proposed.
Since the control algorithms for the electromagnet forces and the vehicle mode motions
are independent of each other, they can be readily implemented using a coarsely grained
parallel processing29 approach. The prime benefit of this approach is that the vehicle
signal processing can be performed by a number of low cost microprocessors.
Additional benefits include easily scalable electromagnet configurations, and the
possibility of achieving fault tolerance at low cost through the use of spare processors.
Introduction 15
The main disadvantage of a parallel processing approach is due to the overhead
associated with the provision and use of the required inter-processor communication.
Overall, the benefits of parallel processing are considered to outweigh the disadvantages
for this particular application. The proposed implementation strategy can use one
processor per electromagnet force controller and vehicle mode motion controller. One
processor can conveniently run both of these functions because only one is active at any
instant in time.
The microprocessor family selected to implement the parallel processor control system
was the Inmos transputer.30 This was chosen because it provides a range of processor
powers, parallel language support using occam, a parallel processor development
system, a low cost/performance ratio, and also because Transputers have inter-processor
communication interfaces included on-chip.
1.8 Direction and scope of this research
This dissertation describes the theoretical and practical research work involved in the
analysis, design, implementation and validation of the proposed new electromagnetic
vehicle suspension control strategy.
In order to validate the proposed control strategy, a small experimental vehicle chassis
(capable of carrying one person) was constructed and equipped with an implementation
of the proposed suspension control system. Since the lateral motions of the vehicle do
not suffer from problematic cross-coupling, electromagnets to provide lateral force
actuation were not employed. This results in only four electromagnets rather than eight
being required, which significantly reduced the cost and complexity of the experimental
vehicle. The electromagnets, track and linear induction motor from an earlier research
project were provided for use with the new experimental vehicle. A single
electromagnet experimental rig was also constructed to facilitate algorithm testing on
an independent suspension configuration.
This account of the research work is partitioned into seven chapters as listed in
Table 1.3. In each chapter the theoretical basis is described and then results from the
experimental systems are discussed and conclusions are drawn.
Introduction 16
In Chapter 2 the steady-state and dynamic behaviour of the electromagnets used on the
Table 1.3 Summary of the research work
1 General literature review and proposed research strategy.
2 Analysis of the electromagnet force characteristic.
3 Synthesis of the electromagnet force control algorithm.
4 Analysis of an independent mode suspension and synthesis of the control algorithm.
5 Analysis of the vehicle motions and synthesis of the vehicle control system.
6 Selection/development of the hardware and software for the control system.
7 Overall conclusions and identification of areas for further research.
experimental research vehicle is analysed. Equations are developed which model the
steady-state lift and lateral forces in terms of core dimensions, air gap, coil current and
lateral displacement. The models include the effects of air gap flux fringing, leakage
flux between the electromagnet poles pieces, and the variability of core permeability
due to saturation effects. Dynamic model equations are developed for the flux lag time
constant due to the electromagnet coil circuit and the eddy current circuits within the
electromagnet and rail cores. The model for the flux lag time constant is a function of
the core dimensions, air gap, number of coil turns, coil resistance, core construction and
core material resistivity.
Chapter 3 describes the analysis of the performance of various feedback control
strategies in terms of their capability to reduce the instability and nonlinearity of the
electromagnet force characteristic. The strategies considered include feedback of force,
flux, rate of change of flux, acceleration, air gap and current. A novel feedback control
algorithm is then proposed using only air gap and current feedback, and embodying the
detailed electromagnet model developed in Chapter 2.
In Chapter 4 the requirements for an independent suspension are analysed. A
suspension control scheme is then proposed which consists of two separate functions.
Firstly, a track coupling algorithm uses the measured track position to determine a
required suspension position. Suspension positioning with disturbance force rejection
is then performed by a state feedback mode position controller which incorporates
acceleration, velocity, position error and position error integral feedback. All feedback
signals for the proposed scheme are derived from measured values of acceleration and
air gap. The validity of the proposed strategy is then tested using simulations and an
experimental single electromagnet suspension rig. Since the experimental vehicle track
Introduction 17
has no significant gradients, a second-order, low pass filter is used for the track to
suspension coupling algorithm to obtain experimental results.
In Chapter 5 the suspension requirements for the complete multivariable vehicle
suspension are analysed. The transformations required to decouple the electromagnet
motions are identified so that independent vehicle mode control loops can be realised.
Extensive experimental suspension tests covering stability, mode decoupling, ride
comfort over simulated rail steps, disturbance force rejection and linear motor force
coupling are then presented and evaluated.
Chapter 6 describes the implementation of the experimental vehicle control system.
First, the development of the electronic hardware for the control system and the
selection criteria for the feedback sensors are described. Hardware development
includes closed-loop electromagnet current controllers, a transputer module motherboard,
transputer based analogue to digital and digital to analogue converter cards, and fibre
optic interface cards for the transputer communication links. The structure of the
concurrent software which implements the vehicle suspension control algorithms is
described next. Topics include selection of sampling rate, discrete digital
implementation of the continuous time design, required numerical accuracy of the
digital processors and finally it proposes a scalable multi-processor configuration
strategy which can efficiently utilise one processor per electromagnet. Practical
implementation features such as real-time data monitoring and logging, control of
vehicle suspension startup and shutdown, and system fault detection are also included
in the experimental system design.
The last chapter draws overall conclusions about the success and limitations of the
results of this research and suggests some areas for further work.
Electromagnet analysis 18
2
Electromagnet analysis
2.1 Introduction
The characteristic behaviour of the electromagnets used to suspend the experimental
research vehicle must be analysed and modelled before an electromagnet control law
can be synthesised. The behaviour of suspension electromagnets is complicated by their
nonlinear and unstable nature, and the dynamic geometry changes associated with the
electromagnet moving along an uneven track.
Kortüm and Utzt31 have shown that a linearised model of a suspension electromagnet
is inadequate for effective simulation of the full operational envelope of a suspension
electromagnet. This chapter therefore presents a detailed nonlinear analysis and a set
of model equations which give good accuracy over the full operational envelope of the
experimental electromagnets.
The geometry and specification of the experimental electromagnet are outlined first.
Then the steady-state behaviour of the electromagnet is modelled. This is followed by
modelling of the dynamic behaviour of the electromagnet flux. The chapter is
concluded with a summary of the electromagnet model equations and their accuracy.
2.2 Electromagnet geometry and specification
Figure 2.1 shows the physical arrangement of the experimental U-shaped electromagnet
and the inverted U-shaped track. Magnetic flux passes through the air gap between the
poles of the electromagnet and track and this generates a force of attraction which
suspends the electromagnet beneath the track.
Electromagnet analysis 19
The electromagnets used for this research were previously used on a research vehicle
Figure 2.1 Physical arrangement of the electromagnet
and track (cross-section perpendicular to track axis)
at the University of Warwick.32,33 They consist of insulated copper windings wound
on steel cores and were designed to lift a maximum load of 50 kg at a nominal
operating air gap of 3-4 mm. Table 2.1 lists the electromagnet dimension indices along
with the relevant values for the experimental electromagnet.
2.3 Steady-state analysis
The role of an electromagnet in a vehicle suspension application is that of a controlled
force actuator. Excitation of the electromagnet coils generates a magneto-motive force
which causes a magnetic flux to flow through the electromagnet, air gaps and track.
The interaction of the air gap flux and field strength generates a force of attraction
between the electromagnet and the reaction rail. The steady-state force characteristic
of the experimental electromagnet will now be analysed and model equations developed.
The analysis is performed by considering the following:
• fundamental magnetic force characteristic.
• lift and lateral force components.
• air gap reluctance.
• iron path reluctance.
• leakage path reluctance.
Electromagnet analysis 20
2.3.1 Magnetic force characteristic
Table 2.1 Experimental electromagnet dimension indices and values
Index Value Dimension
l 200 mm Length of the electromagnet
w 33 mm Width between the electromagnet pole pieces
h 63 mm Height of the pole pieces above the yoke
p 9.5 mm Width of the pole pieces
t 30 mm Width between the track pole pieces
g 0-7 mm Suspension air gap length
N 274
turns
Total number of coils
Rcoils 0.8 Ω Resistance of the coils
ρ 100 nΩm Resistivity of the steel cores (estimated value)
m 7.3 kg Mass of the electromagnet
By assuming that the poles of the electromagnet and the track have equal magnetic
potential over their working faces, the force of attraction across each air gap, Fairgap,
between the electromagnet and track is given by:34
where Hairgap is the magneto-motive force gradient across each air gap and Φairgap is the
2.1Fairgap
1
2H
airgapΦ
airgapnewtons
air gap magnetic flux. As force is generated across two air gaps, the total
electromagnet lift force, Fmagnet, is given by:
Equation 2.2 can be more conveniently expressed by considering it in terms of the
2.2Fmagnet
2 Fairgap
Hairgap
Φairgap
newtons
magneto-motive force across each air gap, Mairgap, and the reluctance of each air gap,
Rairgap.
Electromagnet analysis 21
The magneto-motive force gradient and air gap flux can be expressed as:
where g is the length of each air gap. Substituting Equations 2.3 and 2.4 into Equation
2.3Hairgap
Mairgap
gamperes/metre
2.4Φairgap
Mairgap
Rairgap
webers
2.2 gives:
2.5Fmagnet
M2
airgap
g Rairgap
newtons
A first order approximation for the electromagnet lift force may be made by assuming
the iron paths to have zero reluctance (i.e. infinite permeability), and that the air gap
flux density between the poles, is uniformly distributed over an area equal to the pole
face area. This gives the first order approximations for the air gap magneto-motive
force, M airgap, and reluctance, R airgap, as:
where N is the total number of coil turns, I is the current flowing through the coils, µo
2.6Mairgap
NI
2amperes
2.7Rairgap
g
µo
Apole
amperes/weber
is the permeability of free space and Apole is the pole face area. These two expressions
may be substituted into Equation 2.5 to give a first order approximation for the lift
force, F magnet, as:
Equation 2.8 shows that the magnetic force is a nonlinear function of both current and
2.8Fmagnet
µo(NI)2 A
pole
4g2newtons
air gap length. Also, it shows that for a constant current the force decreases with
increasing air gap, hence it has a negative stiffness coefficient. There is therefore no
point of equilibrium between two magnetised bodies,35 and so the open-loop force-air
gap characteristic of an electromagnet is unstable. Figure 2.2 shows a graph of the
Electromagnet analysis 22
electromagnet force characteristic predicted by the first order approximation given by
Equation 2.8.
The assumption of uniform air gap flux distribution is valid only for air gaps that are
Figure 2.2 First-order model of electromagnet lift force (Equation 2.8)
much smaller than the pole width. At larger air gaps, flux fringing increases the
effective air gap flux area and hence decreases the air gap reluctance. Since the first
order approximation neglects flux fringing and cannot determine the effects of lateral
displacement of the electromagnet relative to the track, a more detailed analysis is
required.
2.3.2 Lift and lateral force components
The experimental suspension electromagnet is relatively long and thin. Therefore,
end-effects can be neglected and a 2-dimensional analysis can be performed by
considering the electromagnet cross-section. This assumption is not strictly true with
regard to eddy-currents when the electromagnet is moving along its rail, so they are
analysed independently in Section 2.4.4. By assuming the pole surfaces to be magnetic
equipotentials, the electromagnet force can be determined using conformal mapping
Electromagnet analysis 23
techniques, but an exact analysis considering all four corners is difficult because the
expressions involve complex elliptic integrals and require the solution of implicit
equations containing elliptic functions.36 However, by assuming that an interval of
uniform magnetic field exists in the air gap, the problem may be reduced to the sum
of 2 two-corner problems, which produce the following simpler formulas for lift and
lateral forces:37
where F magnet is the first order approximation for electromagnet force (Equation 2.8),
2.9Flift
Fmagnet
12g
πp
1y
gtan 1 y
gnewtons
2.10Flateral
Fmagnet
2g
πptan 1 y
gnewtons
p is the pole width and y is the lateral offset between the electromagnet and track poles.
The uniform magnetic field assumption limits the useful range of these expressions to
a maximum air gap and lateral offset of about two-thirds of the pole width.
Figure 2.3 shows a graph of the electromagnet lift force characteristic predicted by
Figure 2.3 Electromagnet lift force model with air gap fringe flux (Equation 2.9)
Equation 2.9. This graph indicates a maximum increase in force relative to the first
Electromagnet analysis 24
order approximation (Equation 2.8) of 40% at an air gap of 7 mm, falling to a decrease
of 7½% at an air gap of 1 mm. The decrease in force at 1 mm is due to the slight
difference between the track pole separation and the electromagnet pole separation.
2.3.3 Air gap reluctance
The flux fringing correction factor in Equation 2.9 cannot be applied directly to the air
gap reluctance (Equation 2.7) because of the inclusion of the orthogonal force
components. However, by considering the case of zero lateral offset, the effects of flux
fringing can be modelled simply and to a good degree of accuracy. This does not
prejudice the application of the full accuracy model for orthogonal forces later on. The
air gap reluctance incorporating lateral fringe flux is given by:
Equations 2.9, 2.10 and 2.11 model the lift and lateral force components and the air gap
2.11R
airgap
g
µo
l
p2g
π
amperes/weber
reluctance, but the magneto-motive force across the air gap (Equation 2.6) still neglects
the magneto-motive force needed to overcome the reluctance of the iron paths due to
their finite permeability.
2.3.4 Iron path reluctance
The reluctance of the iron paths is a function of their geometry and the relative
permeability (µr) of the core material. Since the permeability is a non-linear function
of past and present flux density, incorporating the effects of hysteresis and saturation,
it is very difficult to quantify exactly. It is presumably for this reason that most
researchers in this field choose to neglect its effect by assuming infinite permeability.
The experimental suspension electromagnets are assumed to be made of mild steel for
which the typical maximum permeability is about 2000-3000,38 and the variation of
permeability with flux density39 is outlined in Table 2.2. To accommodate the fact
that the cores of the experimental electromagnet and track have been machined and
welded without subsequent heat treatment, a maximum value for µr of 2000 has been
assumed. With air gaps ranging from about 1% to 7% of the iron path length, the iron
Electromagnet analysis 25
path reluctance is significant, causing a reduction in force at small air gaps and/or high
Table 2.2 Typical variation of µr for mild steel
Flux density µr / maximum µr
0 T 10 %
0.15 T 50 %
0.4 - 1.0 T 100 %
1.3 T 50 %
1.5 T 10 %
2.1 T µr = 1 (saturation)
flux densities.
The reluctance of the track, Rtrack, and electromagnet core, Rmagnet, are given by:
where t is the width between the track poles, w is the width between the electromagnet
2.12Rtrack
(t 4p)
µT
µo
lpamperes/weber
2.13Rmagnet
(2h w 2p)
µM
µo
lpamperes/weber
poles, p is the pole width (and rail pole height), h is the pole height above the yoke, l
is the length of the electromagnet and µT, µM are the relative permeabilities of the track
and electromagnet respectively.
Consideration of the iron path reluctance is required to determine the operational limits
of the electromagnet (due to the onset of saturation) and also to enable it to be
controlled over its full operational envelope. To evaluate the reluctance of the iron
paths, the permeability must be known, and this in turn requires knowledge of the flux
density. Therefore, the leakage flux between the electromagnet poles must be
investigated since the electromagnet core carries both useful suspension flux and the
parasitic leakage flux.
Electromagnet analysis 26
2.3.5 Leakage path reluctance
The magneto-motive force between the poles of the electromagnet causes a parasitic
leakage flux to flow between them in addition to the useful suspension flux which flows
through the air gaps and the track (see Figure 2.4). Finite element analysis has shown
that the leakage flux can exceed the suspension flux at large operational air gaps.40
The leakage flux path reluctance must therefore be modelled. By assuming the
magneto-motive force to be generated linearly over the length of the vertical pole
pieces, the effective height of the poles is halved. A first order approximation to
leakage flux reluctance can now be made by neglecting fringe flux. This gives:
where w is the width between the poles, l is the length of the electromagnet and h is
2.14Rleakage
w
µo
l
h
2
amperes/weber
the height of the poles. Using the expression obtained earlier for the air gap path
reluctance (Equation 2.11), the above expression for the leakage reluctance can be
modified to include leakage fringe flux at the ends and on top of the electromagnet.
This gives:
The incorporation of flux fringing above the top of the electromagnet poles is bound
2.15R
leakage
w
µo
l2w
π
h
2
w
π
amperes/weber
to overestimate the leakage flux at small air gaps due to the close proximity of the track
poles and yoke. However, since the effect of leakage flux on the steady-state force is
significant only at larger air gaps, this fact is not a problem.
Flux leakage around the outer faces of the electromagnet poles has not been modelled
because observation of flux plots obtained using finite element techniques41 shows that
such leakage is not significant for the U-shaped electromagnets.
2.3.6 Steady-state model equations
The model equations for the reluctance of the air gap, leakage and core flux paths are
now combined to build model equations for the electromagnet lift and lateral forces.
Electromagnet analysis 27
The magnetic flux paths of the electromagnet and rail are modelled by the equivalent
circuit shown in Figure 2.4. The coil windings generate a magneto-motive force, Mcoils,
given by:
This drives flux through the electromagnet (reluctance = Rmagnet), where it splits between
2.16Mcoils
NI amperes
the useful path carrying suspension flux (reluctance = 2Rairgap + Rtrack), and a parasitic
path carrying leakage flux (reluctance = Rleakage).
The expression for electromagnet force given in Equation 2.5 is repeated here for
Figure 2.4 Magnet flux paths and flux model circuit diagram
convenience:
The above equation requires an expression for Mairgap in terms of Mcoils. Network
2.17Fmagnet
M2
airgap
g Rairgap
newtons
analysis of the flux circuit diagram is detailed in Appendix A and produces the
following result:
2.18Mairgap
Rairgap
Rleakage
(Rtrack
2Rairgap
)(Rleakage
Rmagnet
) Rmagnet
Rleakage
Mcoils
Substituting for Mairgap (Equation 2.18) into Equation 2.17 gives:
Electromagnet analysis 28
where Rairgap, Rleakage, Rtrack and Rmagnet are defined by Equations 2.11, 2.15, 2.12 and 2.13
2.19F
magnet
Rairgap
R2
leakage
g (Rtrack
2Rairgap
)(Rleakage
Rmagnet
) Rmagnet
Rleakage
2M
2
coils newtons
respectively. The lift and lateral forces can now be expressed by using the factors in
Equations 2.9 and 2.10 and compensating these for the fact that the air gap reluctance
already incorporates flux fringing at zero lateral offset, this gives:
These equations are needed to model the difference between the pole spacings of the
2.20Flift
Fmagnet
1y/g tan 1 y/g
1 πp/2gnewtons
2.21Flateral
Fmagnet
tan 1 y/g
1 πp/2gnewtons
electromagnet and track in addition to modelling lateral offsets between the track and
the electromagnet.
An expression for the air gap flux can now be obtained by substituting for Mairgap
(Equation 2.18) into Equation 2.4 giving:
2.22Φ
airgap
Rleakage
(Rtrack
2Rairgap
)(Rleakage
Rmagnet
) Rmagnet
Rleakage
Mcoils
webers
An expression for the leakage flux is obtained similarly giving:
The flux which flows through the yoke of the electromagnet is the sum of the air gap
2.23Φ
leakage
Rtrack
2Rairgap
(Rtrack
2Rairgap
)(Rleakage
Rmagnet
) Rmagnet
Rleakage
Mcoils
webers
and leakage fluxes above, giving:
2.24Φ
magnet
Rtrack
2Rairgap
Rleakage
(Rtrack
2Rairgap
)(Rleakage
Rmagnet
) Rmagnet
Rleakage
Mcoils
webers
Figure 2.5 shows a graph of the electromagnet lift force characteristic given by
Equation 2.20 with a fixed value for the permeability of the iron paths. This graph
predicts a decrease in lift force relative to the previous graph (Figure 2.3) of
Electromagnet analysis 29
Equation 2.9 due to the effects of the iron path reluctance, and leakage flux. The lift
Figure 2.5 Electromagnet lift force model with fixed µiron (Equation 2.20)
force is reduced by 4% at an air gap of 1 mm and by 11% at 7 mm.
Figure 2.6 shows a graph of the full model of the electromagnet lift force predicted by
Equation 2.20 with the iron path permeability dependent on the iron flux density
according to the values given in Table 2.2. This graph illustrates the effect of the fall
in the permeability of the electromagnet core as it approaches saturation. It also shows
clearly the effect of increased leakage flux at larger air gaps, which causes reduced
maximum lift force due to the electromagnet core saturation.
2.3.7 Accuracy of the steady-state model
The steady-state characteristics of the electromagnet have been analysed and model
equations have been developed. The lateral force of the electromagnet will not be
controlled by the suspension control system and therefore the quality of the lateral force
model is not investigated in this work. However, to verify the accuracy of the lift force
model equations, a simple electromagnet controller was implemented to permit accurate
experimental measurements. Experimental data has been obtained by measuring the
Electromagnet analysis 30
electromagnet current whilst suspending a range of different loads at various suspension
Figure 2.6 Electromagnet lift force model with variable µiron (Equation 2.20)
air gaps. Figure 2.7 shows a graph of the predicted lift force (from Equation 2.20)
along with the measured data from the experimental electromagnet which are plotted
as diamonds. The uncertainty of each experimental data point due to measurement error
is represented approximately by the area of the diamond markers.
Magnetic hysteresis within the electromagnet and track cores causes the experimental
current measurements to vary by approximately ±5%, ±1½% and ±1% at air gaps of
1 mm, 3 mm and 5 mm respectively. To isolate this effect from the underlying
steady-state force characteristic, the experimental values plotted in Figure 2.7 are the
mean of measurements taken with both a rising and a falling flux level.
The performance of the air gap fringe flux model is evident at low forces, before the
onset of core flux saturation. The model accuracy is very good for air gaps of 1-4 mm,
and then steadily deteriorates as the air gap to pole width ratio exceeds about ½. At
higher flux densities, when the core reluctance becomes significant even at large air
gaps, the overall model accuracy remains good. The force model error ranges from -
2½% to +7% for air gaps up to 5 mm, and then increases to a maximum of about +9%
and +13% at air gaps of 6 mm and 7 mm respectively.
Electromagnet analysis 31
Figure 2.7 Steady state lift force model (Equation 2.20) and measured values.
2.4 Dynamic analysis
The models developed so far have dealt with the steady-state characteristics of the
electromagnet. Analysis of the dynamic characteristics is also required because the
open-loop instability of the electromagnet necessitates the use of closed-loop feedback
control. The dynamic relationship between force and flux is considered first. This is
followed by models of the flux and coil current dynamics. Eddy current effects are
then modelled and finally, the effect of magnetic hysteresis is discussed.
2.4.1 Magnetic force
The instantaneous force exerted between the electromagnet and its track can be
approximated as a function of air gap flux Φairgap, by using Equations 2.4, 2.5 and 2.7
to give:
2.25Fmagnet
Φ2
airgap
µoA
pole
newtons
Electromagnet analysis 32
The predicted leakage flux between the electromagnet poles varies from being less than
the air gap flux at small suspension air gaps to being larger than the air gap flux at the
maximum air gap. This effect is significant because even for a constant force, the
electromagnet core flux must double as the air gap varies from minimum to maximum.
Since a change of core flux is required both for changes in force and changes in
suspension air gap length, the dynamics of the flux circuit must now be considered.
2.4.2 Magnetic flux
The relationship between electromagnet core flux Φmagnet, applied coil terminal voltage
V and coil current I, is given by:
where N is the number of coil turns and Rcoils is the total coil resistance.
2.26V NdΦ
magnet
dtI R
coilsvolts
Equation 2.26 is useful for dimensioning the voltage requirement of the power
controller for the electromagnet. The maximum supply voltage needed is the sum of
the parasitic voltage needed for the maximum steady state current requirement plus the
voltage needed to provide the maximum required flux slew rate.
The dynamic behaviour of the electromagnet current must now be considered to
determine the time constant associated with changes in applied coil terminal voltage,
current, flux and air gap.
2.4.3 Coil current
A first order approximation to the electromagnet flux, Φ magnet, can be obtained by
substituting Equations 2.6 and 2.7 into Equation 2.4, giving:
where N is the total number of coil turns, Apole is the pole face area and g is the
2.27Φmagnet
µo
Apole
NI
2 gwebers
suspension air gap length. Substituting Equation 2.27 into Equation 2.26 gives:
Electromagnet analysis 33
where L coils is an approximation to the self inductance of the coils which is given by:
2.28Vd
dtILcoils IR
coilsvolts
The response of Equation 2.28 to a step change in applied coil terminal voltage
2.29Lcoils
µo
N 2 Apole
2 ghenrys
(assuming constant air gap and hence coil inductance), is given by:
where the electrical time constant Tcoils is given by:
2.30IV
Rcoils
1 et /T
coils amperes
2.31Tcoils
Lcoils
Rcoils
seconds
Equations 2.30 and 2.31 show that the electromagnet coil current experiences a first
order lag characteristic relative to the applied terminal voltage. The current lag time
constant given by Equation 2.31 also approximates the flux lag time constant due to the
coil for changes in applied coil terminal voltage or air gap (see Equation 2.27).
A more accurate model of the coil inductance can be obtained by using Equation 2.24
for Φmagnet instead of Equation 2.27 which gives:
This expression varies with air gap and core flux density. Table 2.3 lists the predicted
Lcoils
N2R
track2R
airgapR
leakage
(Rtrack
2Rairgap
)(Rleakage
Rmagnet
) Rmagnet
Rleakage
henrys 2.32
inductance (Equation 2.32) and the consequent lag time constant (Equation 2.31) for a
range of air gaps, but for a constant iron path permeability.
Table 2.3 illustrates how the leakage flux increases the electromagnet flux
(Equation 2.27) at larger air gaps, preventing the inductance from falling as the inverse
of air gap (Equation 2.29). The finite inductance at zero air gap is due to the finite
permeability of the iron paths. This inductance is an overestimate because the
mechanical joints in the iron circuit have been neglected. The table clearly shows that
Electromagnet analysis 34
as the air gap is reduced, the time constant associated with coil current and
electromagnet flux is increased.
Table 2.3 Predicted coil inductance and time constant
Air gap Model Lcoils Model Tcoils
0 mm 1440 mH 1800 ms
1 mm 108 mH 135 ms
2 mm 69 mH 86 ms
3 mm 56 mH 70 ms
4 mm 50 mH 62 ms
5 mm 47 mH 59 ms
6 mm 45 mH 56 ms
7 mm 44 mH 55 ms
2.4.4 Eddy currents
The inductance of the electromagnet coils has been modelled and it has been shown that
there is a phase lag characteristic between a change of terminal voltage or air gap and
the consequent change of flux. Changes in flux level also generate electro-motive
forces in the electromagnet and track cores which cause eddy currents to circulate
within the cores. The eddy currents generate a magneto-motive force which opposes
the flux change. The resistivity of the core material ensures that the eddy currents
decay, so the flux is subject to a lag characteristic. Analysis of the eddy currents is
complicated by the distributed nature of the eddy current circuits, each of which
encloses only a limited portion of the total core flux.
Yamamura and Ito42 performed a sophisticated frequency domain analysis of the
temporal and spatial effects of eddy currents on core flux. Their analysis produced a
model which predicted a first order flux lag with a time constant which increases from
the periphery of the core towards its centre. This spatial flux distribution is due to the
presence of more effective ‘enclosing turns’ around the centre of the core compared to
Electromagnet analysis 35
the periphery of the core. However, the model required to assist in the synthesis of a
control algorithm for the electromagnet does not require a spatial flux distribution
model. A novel time domain analysis is therefore proposed which produces a temporal
model without a spatial model.
The time domain analysis is performed by considering elemental core circuits, which
are then translated to equivalent external coil circuits, ie. coil circuits located outside
the core. The external circuits are equivalent to the internal circuits in terms of their
total core flux coupling. The contributions of each equivalent external circuit are then
combined to give the total flux lag characteristic for a given core geometry. This
analysis, for long, thin pole cross sections, is described in Appendix A and results in
a first order flux lag model with a time constant given by:
where p is the electromagnet pole width, lc is the length of the iron flux circuit, g is the
Teddy
µo
lcp2
24ρgseconds 2.33
suspension air gap and ρ is the resistivity of the core material. To simplify the
analysis, this model neglects the core reluctance, air gap flux fringing and leakage flux.
The temporal component of the model proposed by Yamamura and Ito makes the same
simplifying assumptions and gives a very similar time constant. The only difference
lies in the denominator coefficient which for the frequency domain analysis is 2π2
versus 24 for the time domain analysis. The frequency domain approach therefore
predicts a time constant which is 22% larger than that predicted by the time domain
analysis.
Flux lag due to eddy currents appears in two distinct guises. The first is a reaction to
variation of the flux level in the electromagnet and track due to changing airgap or
changing coil magneto-motive force. For this case, the iron circuit includes both the
electromagnet and the track. To improve the accuracy of the model, a correction factor
is applied to model the effects of air gap flux fringing (see Equation 2.11), and leakage
flux (see Equation 2.15). The effect of leakage flux is modelled by adding it to the
suspension flux passing through the track. The time constants predicted by the model
are shown in Table 2.4 for a range of air gaps.
The second cause of eddy currents is due to the motion of an electromagnet along its
track. Even though the magneto-motive force and air gap are constant, the motion of
the electromagnet is continuously magnetising fresh track at the front of the
electromagnet. A similar effect occurs at the trailing end of the electromagnet as the
Electromagnet analysis 36
track is demagnetised. The flux lag due to the eddy currents reduces the suspension
Table 2.4 Predicted eddy current lag time constant for the electromagnet and track
Air gap Eqn 2.33 Teddy Leakage & fringe
flux correction
Model Teddy
1 mm 11.7 ms 1.36 15.8 ms
2 mm 5.9 ms 1.73 10.2 ms
3 mm 3.9 ms 2.09 8.1 ms
4 mm 2.9 ms 2.46 7.1 ms
5 mm 2.4 ms 2.82 6.8 ms
6 mm 2.0 ms 3.18 6.4 ms
7 mm 1.7 ms 3.54 6.0 ms
force and generates a drag force which retards the motion of the electromagnet.43,44
The iron circuit length for the model (Equation 2.33) is in this case just the track flux
path length. To improve the accuracy of the model, a correction factor is again applied
to model the effects of air gap flux fringing (see Equation 2.11). The time constants
predicted by the model are given in Table 2.5.
Due to the motion of an electromagnetically suspended vehicle along its guideway, the
track flux lag characteristic appears spatially along the length of the electromagnet. The
speed at which the electromagnet length corresponds to 1 flux lag time constant
represents a breakpoint above which a significant loss of suspension force occurs. The
predicted eddy current lag time constant for the track is 3.5 ms at an air gap of 1 mm
(see Table 2.5). This gives a speed break point of 57 m/s for the experimental
electromagnet which is 0.2 m long. Since the experimental track for the research
vehicle is only 5 metres long, the consequent low maximum vehicle speed prevents this
effect from being observed.
2.4.5 Magnetic hysteresis
The magnetisation characteristic for the electromagnet and track steel cores includes
significant magnetic hysteresis.45 Therefore, in addition to the iron permeability being
a nonlinear function of flux density, it is also a nonlinear function of flux history.
Electromagnet analysis 37
Since hysteresis affects the core permeability and hence the core reluctance, it is an
Table 2.5 Predicted flux lag time constant due to eddy currents in the track core
Air gap Eqn 2.33 track
Teddy
Fringe flux
correction
Model track
Teddy
1 mm 3.31 ms 1.07 3.53 ms
2 mm 1.65 ms 1.13 1.87 ms
3 mm 1.10 ms 1.20 1.32 ms
4 mm 0.83 ms 1.28 1.06 ms
5 mm 0.66 ms 1.34 0.88 ms
6 mm 0.55 ms 1.40 0.77 ms
7 mm 0.47 ms 1.47 0.69 ms
effect which increases in significance at smaller air gaps. There are very few references
to hysteresis in the published literature on electromagnetic suspension control, but
Limbert et al46 referred to doubling the damping feedback gain of their suspension
controller to overcome problems which they attributed to hysteresis. Magnetic
hysteresis is very difficult to model analytically with even moderate accuracy because
of its nonlinear dependence on the history of the core flux.
The steady-state model error caused by the iron core hysteresis increases for smaller air
gaps as described in Section 2.3.7. However, due to the small size of the model error
for air gaps greater than 2 mm, and the difficulty of modelling hysteresis, the design
of the suspension control system is to proceed without the aid of an analytical model
for magnetic hysteresis.
Finally, annealing47 the electromagnet core to relieve the stresses built up during
manufacturing increases the iron core permeability and significantly reduces the
hysteresis envelope for minimal cost. This reduces the detrimental effects of both
hysteresis and iron core reluctance prior to saturation for the experimental
electromagnets.
Electromagnet analysis 38
2.4.6 Accuracy of the dynamic model
The flux lags associated with changes in both air gap and applied coil terminal voltage
have been modelled. The flux lag time constant is a function of both the coil
resistance, which is temperature dependent, and most significantly the inductance, which
is a function of air gap. Eddy currents generated within the cores also cause a flux lag
characteristic. Both the coil and the eddy current circuits are magnetically coupled to
the core with little leakage inductance. The combined effect is therefore modelled by
a first-order flux lag (see Appendix A) with a time constant equal to the sum of the coil
and eddy current time constants.
The quality of the flux lag model is now tested by comparing the model predictions
with experimental measurements from the electromagnet and track. All of the measured
flux responses exhibited a dominantly first order lag characteristic.
The electromagnet coil time constant was determined by applying voltage steps to the
coil and measuring the flux rise time constant using a Hall plate flux probe and a
storage oscilloscope. The measured flux time constant includes a lag contribution from
both the coil and the eddy current circuits. The eddy current time constant is therefore
subtracted from the measured value to give the coil time constant. The accuracy of the
coil time constant is insensitive to errors in the measured eddy current time constant
because of the large relative size difference. Table 2.6 lists the model predictions and
experimentally measured values for the electromagnet time constant for a range of air
gaps.
The error between the predicted and measured electromagnet coil time constants ranges
Table 2.6 Predicted and experimental electromagnet coil time constants
Air gap Model Tcoil Experimental Tcoil Error factor
1 mm 135 ms 111 ms 1.21
4 mm 62 ms 76 ms 0.82
7 mm 55 ms 68 ms 0.81
from around +20% to -20%. With a measurement accuracy of about ±5% the model
Electromagnet analysis 39
accuracy is seen to be quite satisfactory. This model relies heavily on the static model
equations thus giving additional confidence in their accuracy. The error at 1 mm is in
part due to an overestimate of leakage flux which is inherent in the static force model
equations (see Section 2.3.5). The discrepancies that exist suggest that the iron
permeability may be lower than assumed and that the leakage flux may be slightly
larger than modelled for large air gaps.
In order to measure the eddy current time constant, the effective coil circuit time
constant was significantly reduced. This was achieved through the use of a closed-loop
controller incorporating high gain (x 100) current feedback. Analysis of this
closed-loop configuration (see Appendix B) shows that current feedback reduces the
effective coil time constant, whilst leaving the eddy current time constant unaltered.
Step changes in coil current reference were input to the controller and the flux change
time constant was again measured using a Hall plate flux probe and a storage
oscilloscope. The measured flux time constant now includes the eddy current lag time
constant plus 1% of the coil circuit time constant. The contribution of the coil circuit
is therefore subtracted from the measured time constant to give the eddy current time
constant. The relative size of the two components gives an eddy current time constant
error approximately equal to the measurement error plus one third of the coil time
constant error. Table 2.7 lists the predicted and measured values for the eddy current
time constant for the electromagnet and track cores over a range of air gaps.
The predicted values are all about 3 to 3.5 times the experimental values. The rather
Table 2.7 Predicted and experimental eddy current lag time constants for the
electromagnet and track
Air gap Model Teddy Experimental Teddy Error factor
1 mm 15.8 ms 4.6 ms 3.4 x
4 mm 7.1 ms 2.4 ms 3.0 x
7 mm 6.0 ms 1.8 ms 3.3 x
large discrepancy between the model and the experimental system is attributed mostly
to the model assumption of uniform core flux distribution and zero core reluctance. In
Electromagnet analysis 40
reality, the flux lag is composed of spatially distributed components, with time constants
which are small near the perimeter of the core, but which increase towards the centre
of the core. The interaction between distributed flux components is further complicated
by the effects of the core permeability.
In addition to the above factors, the model parameters represent an additional error
source. All of the core dimensions were measured with high accuracy, however, the
core resistivity could not be measured with available equipment. Manufacturing records
for the cores do not exist, but they appear to be made of mild steel. If, however, the
cores are made of magnetic steel with even as little as 0.5% Silicon content, then the
resistivity would be approximately double that of mild steel.48 In view of the
uncertainty associated with the core resistivity, and the otherwise good fit of the first
order lag model, further experimental work with a known core resistivity would be
required to determine the exact accuracy of the eddy current model time constant.
For the purposes of developing an electromagnet control algorithm, the quality of the
first order lag model is considered to be sufficiently accurate when combined with the
measured time constants.
It is assumed that if the track cores are made of the same material as the electromagnet
cores, then the predicted eddy current time constants for the track alone (see Table 2.5)
are also approximately three times too large.
2.5 Concluding remarks
Models have been developed for the significant steady-state and dynamic characteristics
of the experimental suspension electromagnet. The force between the electromagnet
and its reaction rail is dominantly a function of the core dimensions, the square power
of the coil current-turns and the inverse square power of the length of the suspension
air gap. The resulting negative stiffness makes the electromagnet open-loop unstable.
The operational envelope for a given electromagnet geometry depends on the maximum
achievable air gap flux. This is determined by the flux saturation level of the cores,
and the air gap which determines the reduction of air gap flux due to the significant
flux leakage between the electromagnet pole pieces. Finite core permeability at low
flux levels is also significant for small air gaps. The electromagnet steady-state model
equations therefore incorporate variable core permeability and leakage flux.
Electromagnet analysis 41
The dynamic characteristics of the electromagnet force are dominated by the dynamic
behaviour of the electromagnet coil current and the eddy currents within the cores.
These are a function of the coil inductance and resistance, the core dimensions and
resistivity, and the suspension air gap and core permeability. In addition to the
electromagnet core flux varying with force, the flux also varies with air gap (at constant
force) due to leakage flux varying with air gap. The dynamic behaviour of the
electromagnet flux is modelled by a first order lag with a time constant given by the
sum of the coil and eddy current time constants.
The accuracy of the steady-state model equations is very good. The characteristic
behaviour of the dynamic model equations is also good, but the predicted time constant
for the eddy current lag is about 3 times that of the measured time constant. This
discrepancy may be largely due to an erroneous estimate for the core resistivity.
However, further experimentation with a material of known resistivity is required to
validate the model time constant more fully.
Electromagnet force control 42
3
Electromagnet force control
3.1 Introduction
The multivariable vehicle suspension control system proposed in Chapter 1 requires
electromagnetic force actuators which are independent, and suitably linear and stable,
in order to decouple and control the vehicle mode motions successfully. This chapter
describes the steps involved in the synthesis and validation of a novel electromagnet
force control scheme which meets the above requirements.
The open-loop force/voltage transfer function of the electromagnet is first examined to
identify the dynamic structure and the parameters which characterise the experimental
suspension electromagnet. Existing force control strategies, which use linear algorithms,
are then examined and found to be unsatisfactory for providing independent and linear
force actuation. A novel control scheme, employing a detailed nonlinear model of the
electromagnet, is proposed to meet the force actuation requirements. After describing
the design of the proposed force controller, the results of some experimental
performance tests are presented and discussed. Finally, some conclusions are drawn on
the work described in this chapter.
Before examining the electromagnet transfer function, the operational envelope for the
experimental suspension electromagnet is defined, and some operational parameters are
identified and discussed.
3.2 Operational envelope
The experimental electromagnet has a mass of 7.3 kg and was designed to support a
maximum load of about 50 kg, giving a lift/weight ratio of about 7. A 50 kg load calls
Electromagnet force control 43
for a maximum lift force capability of about 550 N to allow for some acceleration of
the suspended load. The design of the experimental electromagnet results in a large
leakage flux which limits the maximum air gap to about 5 mm for a lift force of 550 N
(see Figure 2.7). This allows a nominal operating air gap of 3 mm, with a deflection
of ±2 mm, to be realised. Mechanical limit stops are positioned to restrict the air gap
range to 0.5-5.5 mm. The current range required to suspend 50 kg is 3.5-20 A for air
gaps of 1-5.5 mm respectively. This gives a power/lift ratio of about 1.6 W/kg at the
nominal operating air gap.
The minimum suspended mass for the experimental suspensions is 15 kg per
electromagnet. Using Equation 2.22, the predicted air gap flux density varies from
about 0.35-0.65 Tesla over the full operational envelope. The corresponding predicted
electromagnet core flux density using Equation 2.24 is 0.45-1.50 Tesla. The operational
envelope for the electromagnet force actuator is summarised in Table 3.1. The
power/lift and lift/weight ratios for the experimental electromagnet are very similar to
those achieved by electromagnets with a nominal suspension load of 250-1000 kg.49
Table 3.1 Operational envelope for the electromagnet
Parameter Operating range
Lift force 0-550 N
Suspended mass 15-50 kg
Air gap 3 mm ±2 mm
Current 0-20 A
3.3 Electromagnet transfer function
A transfer function for the electromagnet is now derived to characterise its dynamic
behaviour and to analyse the nature of the force instability. The approximate model of
the electromagnet developed in Section 2.3.1 is used as the starting point. This model
is linearised around a nominal operating point to permit modelling of the electromagnet
in the frequency domain using the Laplace operator50 s.
Electromagnet force control 44
Since the electromagnet is in practice excited by a variable voltage source, the
electromagnet force/voltage transfer function is derived and the location of the
open-loop poles and zeros is identified. Figure 3.1 illustrates the configuration of the
electromagnet and track and the variable nomenclature used in this chapter. To simplify
the analysis, the natural frequency of the electromagnet reaction rail is assumed to be
sufficiently high to be neglected.
Four equations which characterise the electromagnet motion for small perturbations
Figure 3.1 Electromagnet and track configuration
around a nominal operating point (io, co) are identified. These are then combined to
form the force/voltage transfer function. All equations are expressed in the frequency
domain using the Laplace operator s.
The air gap flux is the key element in the operation of the electromagnet, and it is
approximately proportional to the coil current divided by the air gap (see
Equation 2.27). This is linearised for small perturbations by:
and where ∆C, ∆Φ, and ∆I are the perturbations of air gap, its flux, and current
3.1∆Φ(s) kφi∆I(s) kφc
∆C(s) where kφi
∂φ(io,c
o)
∂i, kφc
∂φ(io,c
o)
∂c
respectively, and the partial derivative coefficients, kφi and kφc are both positive.
Electromagnet force control 45
The electromagnet force, which is approximately proportional to the square power of
the air gap flux (see Equation 2.25), is linearised by:
and where ∆F is the force perturbation, and the partial derivative coefficient kfφ is
3.2∆F(s) kfφ∆Φ(s) where k
fφ
∂f(io,c
o)
∂φ
positive.
The acceleration of the suspended load due to the electromagnet force is now given by:
where m is the suspended mass.
3.3s 2 ∆C(s)∆F(s)
m
The final equation determines the relationship between electromagnet coil voltage, air
gap flux, and coil current (see Equation 2.26). This relationship is expressed by:
where ∆V is the coil voltage perturbation, N is the number of coil turns, and Rcoils is the
3.4∆V(s) sN ∆Φ(s) Rcoils
∆I(s)
coil resistance.
The force/voltage transfer function for the electromagnet is now obtained by substituting
current from Equation 3.1, flux from Equation 3.2, and force from Equation 3.3, into
Equation 3.4. The substitutions are detailed in Appendix B and result in:
This is more conveniently expressed as:
3.5∆F(s)
∆V(s)
kfφkφi
Rcoils
s 2
s 3 Nkφi/R
coilss 2 k
fφkφc/m
where:
3.6∆F(s)
∆V(s)
ki
Rcoils
s 2
s 3 Tflux
s 2 kc/m
Electromagnet force control 46
The partial derivative coefficient ki represents the force/current gain factor, whilst kc
3.7k
ik
fφkφi
∂f
∂φ∂φ∂i
∂f
∂i, k
ck
fφkφc
∂f
∂φ∂φ∂c
∂f
∂c,
Tflux
Teddy
Tcoil
, Tcoil
N kφi/R
coils
represents the magnitude of the electromagnet’s negative stiffness. Tflux is the total flux
lag time constant (see Section 2.4.6), which is comprised of Teddy, the eddy current time
constant, and Tcoil, the coil time constant (see Appendix B).
Equation 3.6 shows that the force/voltage transfer function has two zeros at the origin
of the s-plane and three poles. The negative electromagnet stiffness generates a positive
denominator root, which places one of the poles in the right-hand side of the s-plane,51
thus producing an open-loop unstable force. Figure 3.2 illustrates the open-loop
electromagnet force/voltage transfer function, with the scope of the nonlinear elements
denoted by the two dashed boxes. The configuration of the eddy current flux lag loops
is identical to that of the coil current flux lag loop, but with a larger resistance to reflect
the smaller eddy current time constant. The eddy current loops are omitted from
Figure 3.2 for the sake of clarity. The open-loop transfer function is augmented in
Figure 3.2 with a second-order model of a flexible track which shows how the track
motion couples with that of the electromagnet. The track model has a natural
undamped frequency, ω, and damping ratio, ζ.
Figure 3.2 Block diagram of the electromagnet system
Electromagnet force control 47
Table 3.2 shows how the measured parameters of the experimental electromagnet vary
as a function of air gap and suspension force. The partial derivative coefficients
defined in Equation 3.7, namely ki and kc, are calculated from the measured data used
to produce the graph shown in Figure 2.7. The coil and eddy current time constants are
taken from the measured data summarised in Tables 2.6 and 2.7 respectively.
Due to the nonlinear nature of the electromagnet, the location of the poles of the
Table 3.2 Linearised electromagnet model parameters
Air gap1 mm 3 mm 5 mm
Max
/minParameter
ki 140-400 N/A 50-90 N/A 25-50 N/A 16
kc 260-950 N/mm 95-350 N/mm 60-150 N/mm 16
Tcoil 111 ms 85 ms 72 ms 1.5
Teddy 4.6 ms 3.0 ms 2.1 ms 2.2
Notes: The coefficients ki and kc were measured over the force range of 150-550 N
transfer function varies with the operating point. The unstable pole reaches a maximum
value of s = 77, when kc=950 N/mm, m=15 kg, and Tflux=0.1 s, whilst the two remaining
poles form a complex conjugate pair at s = -44 ±70j. The worst case instability time
constant is therefore 13 ms.
In order to be able to neglect the effects of track flexibility and hence vibration, the
track poles must be sufficiently separated from the electromagnet force poles. For the
open-loop voltage controlled electromagnet, the track poles should ideally have
frequencies of at least 16 Hz. The acceptable natural frequency for any closed-loop
electromagnet configuration requires adequate separation of the track poles from the
dominant poles of closed-loop response.
Electromagnet force control 48
3.4 Force control strategies
The force/voltage transfer function of the electromagnet has been shown to be
open-loop unstable due to the electromagnet’s negative stiffness. A control strategy is
now needed to provide electromagnetic force actuation with sufficient stability,
accuracy, linearity, and bandwidth.
The force actuation stability is most conveniently expressed in terms of the residual
electromagnet stiffness that is superimposed upon the controlled electromagnet force.
The acceptable level for residual stiffness clearly depends on the requirements of the
system using the electromagnet force actuator. For this application, the suspension
control system uses a suspension stiffness gain of 250 N/mm and is designed to
accommodate total transducer errors of up to ±25% (see Section 4.4.2). Therefore, a
force accuracy of ±20% with a corresponding residual stiffness of up to ±50 N/mm will
not significantly impair the suspension controller response. The linearity of the force
actuators will affect the success of the vehicle mode decoupling (see Chapter 5) and so
a force nonlinearity of no more than 10% is acceptable.
The suspension controller requires a force actuation bandwidth of about 50 Hz or more
(see Section 4.4.3). The ideal force actuation transfer function therefore consists of a
single pole low-pass filter with a time constant ≤ 3.2 ms. For small force perturbations,
a linear approximation between flux and force can be assumed. The dynamic behaviour
of the electromagnet force and flux is therefore considered to be equivalent for the
purpose of developing a force control strategy.
The most direct method to reduce the electromagnet force instability, to improve its
linearity, and to increase its force actuation bandwidth, is to apply force feedback, or
some functionally equivalent feedback such as acceleration or air gap flux. An
alternative approach is to target the cause of the instability, and use air gap feedback
to reduce the instability. Since the largest flux time constant for the force/voltage
transfer function is about 100 ms, an air gap feedback strategy must be augmented by
an additional technique to increase the force actuation bandwidth. Before discussing
these force control strategies, it is useful to consider the constraints imposed by the
electromagnet power control hardware.
In order to achieve an acceptable weight and cost for the electromagnet power control
hardware, the power conversion efficiency must be high, and the use of a switch-mode
power controller rather than a linear one is therefore required.52 This imposes an
Electromagnet force control 49
upper limit on the voltage actuation bandwidth which is determined by the required
voltage and current ratings, and the type of semiconductor device used to switch the
power.53 An additional constraint on the power controller performance is due to the
provision of a finite voltage source, with the attendant potential problems of voltage and
hence current slew-rate saturation.
The steady-state electromagnet current ripple which arises due to the use of a
switch-mode power controller,54 causes the magneto-motive force of the electromagnet
to experience a similar superimposed oscillation. This continuous cycling of the
magneto-motive force increases the effective differential permeability of the core
material for small current perturbations. The use of a switch-mode controller therefore
ameliorates the control problems caused by the magnetic hysteresis of the electromagnet
and track cores.
3.4.1 Force feedback
The open-loop force/voltage transfer function (see Equation 3.6) has two zeros at the
origin and three poles, one of which is located in the right-hand side of the s-plane, thus
making the open-loop system unstable. However, the two zeros at the origin suggest
that if sufficient force feedback is applied, the unstable pole can be drawn in close to
the origin, thus reducing the force instability. Figure 3.3 illustrates the closed-loop root
locus of the electromagnet for a force feedback gain varying from 0 to 120. The
maximum gain is determined by limiting the time constant of the fast ‘flux lag’ pole
to 1 ms to accommodate an acceptable minimum response time for the power controller.
Attempts to use excessive gain would result in the additional pole due to the power
controller causing the left branch of the root locus to split away from the real axis. The
resultant pair of high frequency complex conjugate poles would produce an undesirable
oscillatory force response.
Figure 3.3 shows that the use of force feedback can improve the worst case time
constant of the unstable pole from its open-loop value of 13 ms to 44 ms. Since the
closed-loop stiffness is inversely proportional to the square power of the unstable pole
time constant,55 the effective negative stiffness is reduced by a factor of 11.
The principle problem with applying force feedback is that of measuring the force. In
the case of a single electromagnet suspending a load many times its own weight, a force
transducer could be located between the electromagnet and its load. However, for the
Electromagnet force control 50
experimental vehicle, the measured force would suffer disturbances from the other
Figure 3.3 Closed-loop root locus for force feedback
(kc = 950 N/mm, m = 15 kg, Tflux = 116 ms, Gain=0-120)
electromagnets which are directly coupled to the vehicle chassis. Since independent
electromagnet stabilisation is required, it is unacceptable to transform the electromagnet
instability into a multi-variable instability problem. Therefore, direct force measurement
and feedback is not a feasible option. Flux and acceleration are two possible substitutes
for force feedback, so these are considered next.
3.4.2 Flux feedback
Since the electromagnet lift force is proportional to the square power of the air gap flux
(see Equation 2.25), the r.m.s. level of the flux over the entire pole face area represents
the lift force. This factor complicates matters because the pole face flux distribution
is not always uniform. This is due to three dominant factors. Firstly, when changes
in flux are demanded, eddy currents in the electromagnet and track cores cause flux to
be concentrated near the pole edges during the flux transient. Secondly, the motion of
an electromagnet along its track produces a similar effect, with track-borne eddy
currents causing a flux lag which is distributed spatially along the length of the
electromagnet pole face and across its width. Finally, gaps between track joints can
cause local flux disturbances.
Electromagnet force control 51
Two different transducers have been used to measure flux in electromagnetic suspension
systems, namely Hall plate devices and search coils. Hall plate sensors56 must be
mounted near the electromagnet pole face to minimise the measurement of leakage flux.
This exposed location is environmentally harsh and the electromagnet core experiences
both low and very high temperatures which can cause problems with the delicate Hall
plate devices. Positioning the sensor on the pole face is an additional problem due to
the non-uniform distribution of flux across the pole face area. Use of laminated
electromagnet and track cores can significantly reduce the flux distribution problem, but
it is an expensive option. Alternatively, a number of sensors can be used, distributed
over the pole face to provide sufficient measurement accuracy and also redundancy to
cope with device failures.
Hall plate sensors can enable excellent flux control, both dynamically and in the
steady-state, although a nonlinear controller is needed for linear force actuation.
However, due to their questionable robustness, they have not advanced from laboratory
prototypes.
The prime benefit of flux control can be achieved more readily and robustly through
the use of a search coil wound around the electromagnet poles, close to the pole faces.
The coil voltage gives a measure of the rate of change of flux which must be integrated
to provide a measure of the flux. To avoid problems due to integrator drift, a low
frequency roll-off is required. The resultant lack of a d.c. response does not impair the
stabilisation function since the flux dynamics are the problem, not the steady-state flux.
However, linear control of the lift force is impossible since there is no absolute flux
level measurement.
The merits of the search coil sensor are that it senses the average flux level over the
whole pole face area, and it provides a cheap and robust method of reducing the force
instability. However its major disadvantage for this application is that it cannot produce
a linear force actuator.
3.4.3 Acceleration feedback
The final alternative to direct force measurement is to measure electromagnet
acceleration to derive the lift force. An acceleration signal is required by the
suspension controller, and is obtained by attaching an accelerometer to the
electromagnet. If a conventional secondary suspension is used, this technique is
Electromagnet force control 52
available directly. However, the rigid coupling of the electromagnets and
accelerometers to the chassis of the experimental vehicle leads to a tightly coupled
multi-variable instability problem. Acceleration feedback, like force feedback, is thus
incapable of providing independent electromagnet stabilisation.
3.4.4 Air gap and current feedback
The electromagnet is open-loop unstable due to its negative stiffness coefficient. The
instability of the electromagnet force can therefore be reduced by using air gap feedback
with the controller stiffness set as close as possible to the magnitude of the open-loop
electromagnet stiffness. The force actuation time constant can then be significantly
reduced by using high gain current feedback. A current loop gain of 100 is sufficient
to reduce the coil current lag time constant from a worst case value of 110 ms down
to approximately 1 ms. The transfer function of the open-loop current controlled
electromagnet is given by Equation 3.6, but with Tflux now given by:
where kamp is the loop gain of the current controller (see Appendix B). The worst case
3.8Tflux
Teddy
Tcoil
kamp
time constant of the unstable pole for the current controlled electromagnet is 5.6 ms (for
kamp = 100). The higher force actuation bandwidth incurs the penalty of a faster
instability time constant relative to a voltage controlled electromagnet.
Air gap and current feedback can theoretically provide a stable and linear force actuator,
but only for a single operating point in terms of air gap and force. The problem with
this technique is that the negative stiffness coefficient (see Table 3.2) varies by a factor
of about 16 over the full operational envelope of the electromagnet. The closed-loop
root locus for the current controlled electromagnet with air gap feedback (see
Figure 3.4) shows that the closed-loop force becomes more unstable as the air gap
feedback gain error increases. If the feedback gain is too high, a pair of unstable
complex conjugate poles replaces the single unstable pole which is present if the gain
is too low. If the feedback gain is approximately equal to the electromagnet stiffness,
the two poles near the origin can be considered to cancel with the two zeros at the
origin. The resultant force transfer function is then dominated by the flux lag pole.
Electromagnet force control 53
In order to match the variation in the open-loop stiffness over the full operational
Figure 3.4 Closed-loop root locus for air gap and current feedback
(kc=950 N/mm, m=15 kg, Tflux=5.7ms, Gain=0-2kc)
envelope, a nonlinear control algorithm is required. A disadvantage of this technique
is that in order to achieve accurate gain scheduling, the required control algorithm must
embody a complex model of the electromagnet (see Chapter 2). If sufficient accuracy
can be achieved in the model and signal measurements, a reasonably stable and linear
force actuator is theoretically possible. A further disadvantage associated with air gap
feedback stabilisation, is that unlike force or flux feedback, air gap feedback does not
encompass the magnetic hysteresis of the electromagnet and track cores.
Accurate measurement of the r.m.s. air gap over the full length of the electromagnet is
ideally required. In practice, a large area sensor and/or a number of smaller sensors
could be used to measure the average air gap. For example, using triple mode
redundancy, three sensors could provide average air gap measurement, with a graceful
degradation of accuracy if one of the sensors failed. Various industrial sensors using
inductive, capacitive and eddy-current techniques57 are available for non-contacting
displacement measurement.
Electromagnet force control 54
3.4.5 Proposed force control strategy
The major problem associated with producing a well controlled suspension force from
an electromagnet stems from the variation of the parameters of the open-loop transfer
function rather than the fact that it is unstable. However, in reality it is only the
parameters of the linearised model which vary. The parameters of the nonlinear models
developed in Chapter 2 are mostly constant. The coefficient parameters of the static
force characteristic are determined by the physical dimensions of the electromagnet and
track, and the permeability of air, both of which are essentially constant. The dynamic
behaviour of the electromagnet is only slightly less ideal since temperature affects the
coil resistance and core resistivity which in turn affects the coil current and
eddy-current time constants respectively. With such detailed models of the nonlinear
behaviour of the electromagnet available, it would be unwise to discard most of the
information in order to use only classical linear control algorithms.
The only force feedback scheme which is suitably robust is flux derivative feedback
using a search coil. This can stabilise the electromagnet force but it cannot provide
linear force actuation. For electromagnetically suspended vehicles which use
conventional secondary suspensions, the electromagnets are effectively decoupled by the
secondary suspensions and they can therefore be considered to be independent. For
such systems, flux derivative feedback and/or acceleration feedback is employed58,59
to stabilise and linearise the force actuation.
Alternatively, since air gap feedback is required by the suspension controller, air gap
feedback stabilisation can be used with no additional transducer requirements.
However, to achieve satisfactory force stability and linearity, the air gap feedback gain
must be changed dynamically as a function of the operating point. The proposed
central element of the electromagnet force controller is therefore, air gap feedback into
a nonlinear model of the electromagnet static force characteristic.
The force controller must increase the open-loop force actuation bandwidth to the 50 Hz
required by the suspension controller. This corresponds to a force actuation time
constant of about 3.2 ms. The force time constant for the electromagnet is
approximated by the flux time constant which is composed of the coil and eddy current
lag time constants (see Equation 3.7). Since the worst-case open-loop coil and eddy
current time constants (see Table 3.2) are 111 ms and 4.6 ms respectively, both of these
time constants must be reduced. The massive reduction required for the coil time
constant can be most robustly achieved through the use of high gain current feedback,
Electromagnet force control 55
so that is the proposed method. The current controller also enables coil resistance
changes due to varying temperature to be neglected. The additional hardware
complexity of an electromagnet current controller over a voltage controller is not great
and it can readily provide safe current limiting under fault conditions.
Since the reduction ratio required for the eddy current time constant is not large,
pole-zero cancellation using a phase-lead compensator60 is proposed to achieve the
necessary reduction. The increase in core resistivity due to elevated temperatures
reduces the eddy current time constant and is therefore neglected. The phase-lead
compensation should be applied to the electromagnet core flux demand, which
incorporates the leakage flux as well as the air gap flux. In a practical system, it would
be better to design the electromagnets to have an acceptable eddy current time constant
by using either narrower pole pieces, a higher resistivity core material, or laminated
cores (see Appendix A). This would reduce the power controller voltage requirement
and applies irrespective of the stabilisation strategy used.
Figure 3.5 illustrates the configuration of the proposed force control strategy. The eddy
Figure 3.5 Proposed force control configuration
current flux lag loop has been omitted for the sake of clarity. The detailed design of
Electromagnet force control 56
the force control algorithm is described next. The implementation of all control system
components is described in Chapter 6.
3.5 Force controller design
The stabilisation and linearisation element of the proposed force control strategy
consists of feeding the measured air gap and the demanded force into a model of the
electromagnet, which then specifies the requisite current demand. The full nonlinear
model of the electromagnet is given by Equation 2.20 and is illustrated by Figure 2.7.
This equation cannot be used directly because the magnetic reluctance of the
electromagnet core depends on the core flux density. Therefore the core flux density
must be evaluated en route to the current demand. The full force control algorithm is
listed in Figure 3.6, where c is the air gap between the electromagnet and the track, y
is the lateral offset between the electromagnet and track poles (which is assumed
constant), p is the pole face width, and the air gap, track, leakage and electromagnet
reluctances are defined in Chapter 2.
The lateral offset in the control algorithm refers to the fixed pole offset due to the
difference in separation between the electromagnet poles and the reaction rail poles.
The electromagnet lateral offset is not measured in order to reduce the cost and
complexity of the experimental vehicle. Therefore, lateral displacement of the
electromagnet relative to the rail, results in a reduction of the suspension lift force. The
force reduction does not affect the linearity of the force actuation and only slightly
impairs the electromagnet stabilisation function. The same argument applies to the loss
of suspension force due to reaction rail eddy currents induced by the electromagnet
motion along its guideway.
To increase the open-loop force bandwidth to that required by the suspension controller,
the proposed force control strategy employs current feedback and phase-lead
compensation of the electromagnet core flux. The required force actuation time
constant is about 3.2 ms, whilst the worst-case open-loop coil and eddy current time
constants (see Table 3.2) are 111 ms and 4.6 ms respectively. To meet this
requirement, a current controller feedback gain of 100 is appropriate to reduce the
maximum coil current time constant to a maximum value of about 1.1 ms, giving a total
flux lag time constant of 5.7 ms. The total flux lag time constant can be further
reduced to a maximum value of 2.5 ms by employing pole-zero cancellation. This is
implemented in the controller by applying phase-lead compensation to the electromagnet
Electromagnet force control 57
core flux demand using the compensator whose transfer function is given by:
Figure 3.6 Electromagnet force control algorithm (sequential program)
Fper_pole
Fdemand
2
1y/c tan 1(y/c)
1 πp/2c
gross force per pole
Φairgap
Fper_pole
2c
Rairgap
(c)air gap flux
Mairgaps track
Φairgap
2Rairgap
(c) Rtrack
air gap mmfs track mmf
Φleakage
Mairgaps track
/Rleakage
leakage flux
Φmagnet
Φairgap
Φleakage
magnet core flux
Φcomp
Phase_Leadeddy
(Φmagnet
) eddy current compensation
Mmagnet
Φcomp
Rmagnet
(Φmagnet
) magnet core mmf
Icoils
Mairgaps track
Mmagnet
/Ncoil_turns
current demand
The maximum value for the compensated flux time constant of 2.5 ms gives a minimum
3.9Phase_Lead
eddy(s)
1 s Tflux
1 s Tflux
/nwhere T
flux5.7 ms, and n 2.3
predicted force actuation bandwidth of about 64 Hz. The actual values used in the
experimental system are Tflux = 4.6 ms and n = 3, thus giving a slightly higher minimum
force bandwidth.
3.6 Force controller performance
The performance of the electromagnet force control algorithm cannot be conveniently
tested in isolation due to the residual closed-loop stiffness that always exists due to
imperfections in the control algorithm and its implementation. Therefore, a closed-loop
air gap controller was used to assist in testing the accuracy, linearity, and stability of
a practical implementation of the force control algorithm. The suspension control
Electromagnet force control 58
system test results presented in Chapter 4 are used to validate the force actuation
bandwidth of the experimental force controller.
To impose a realistic test environment on the force controller, the suspension controller
developed in Chapter 4 is used to obtain test results for the experimental force
controller. The physical arrangement of the experimental single electromagnet
suspension rig used for the performance tests is illustrated in Figure 3.7. All measured
signals are provided by the control system from the sampled input data.
Two tests are used to determine the effectiveness of the force control strategy and
Figure 3.7 Single electromagnet suspension experimental rig
implementation. Both tests are performed over the operational range of air gaps and
suspension forces. The first test measures the static force accuracy and linearity by
plotting measured suspension forces versus reference force demands. The second test
determines the residual force actuation stiffness, and hence the residual force instability.
Figure 3.8 illustrates experimental static force measurements for suspended loads of
140, 220, 360 and 510 N. The graph plots the suspension load forces versus the
reference force demands required to generate those suspension forces. The
measurements were taken over the operational air gap range of 1-5 mm. All of the
force measurements are within the anticipated force error tolerance of ±15% except for
the 1 mm and 5 mm points with the 510 N load. These are out by just under 19%.
The larger error at the 1 mm air gap is attributed to slight bending of the electromagnet
support beam which causes air gap measurement errors. The higher error for the
maximum air gap is due to the larger inaccuracies involved in modelling the core
permeability at the very high core flux level at that operating point.
Electromagnet force control 59
Figure 3.8 clearly shows that the proposed force controller is successful in producing
Figure 3.8 Experimental controller reference force demands and actual suspension
forces
a dominantly linear static force actuation over the full operational envelope. The force
nonlinearity, determined by measuring the maximum deviation from a straight line fitted
through all of the experimental data, is typically about ±5%, with a worst case value
of ±7% at an air gap of 1 mm.
The residual stiffness exhibited by the force controller is investigated by experimenting
with the closed-loop stiffness of the suspension controller by means of the air gap
feedback gain. The error integral action is first removed and then the controller air gap
feedback gain is reduced to determine the minimum value for which the system remains
stable. The worst case electromagnet negative stiffness predicted by Table 3.2 is
950 N/mm, which occurs at maximum load and minimum air gap. For the experimental
system, the minimum stable air gap feedback gain for the operating point (1 mm,
510 N) was about 50 N/mm. This represents a residual stiffness of 5% of the predicted
electromagnet stiffness. For smaller loads and larger air gaps, the minimum stable air
gap feedback gain was approximately 25 N/mm. This result shows that the proposed
force control strategy achieves a residual stiffness of 10-20% of the suspension
controller stiffness gain of 250 N/mm (see Section 4.4.2).
Electromagnet force control 60
Table 3.3 summarises the measured performance of the experimental electromagnet
force controller in terms of its general accuracy, linearity and residual stiffness. As
expected, the performance of the experimental system was found to be highly dependent
on the accuracy of the air gap measurement. However, calibration of the air gap sensor
offset at the most critical operating point (at a gap of 1 mm), ensured a satisfactory
performance from an inexpensive industrial air gap sensor. The impact of magnetic
hysteresis is just noticeable at the minimum operational air gap, but it did not
significantly impair the behaviour of the force controller.
Table 3.3 Performance of the electromagnet force controller
Parameter Typical Worst case
Accuracy ± 15 % - 19 %
Linearity ± 5 % ± 7 %
Residual stiffness ≤ 25 N/mm ≤ 50 N/mm
3.7 Conclusions
The force/voltage transfer function of the experimental suspension electromagnet has
been analysed. Force control schemes employing linear feedback techniques have been
shown to be unsuitable for providing independent linear force actuation in an
environment where a number of electromagnets are rigidly coupled. Therefore, a new
force control algorithm has been proposed, which employs a detailed nonlinear
electromagnet model, in conjunction with air gap feedback, to provide an independent
force actuator for rigidly coupled electromagnets. The bandwidth of the electromagnet
force is increased through the use of closed-loop current feedback and series
compensation of the electromagnet core flux.
The proposed electromagnet force control scheme has been shown to possess significant
advantages compared with existing stabilisation techniques using flux derivative
feedback due to its dominantly linear force actuation. It does however, suffer from a
slight disadvantage due to its increased reliance on an accurate air gap measurement at
small air gaps. This drawback could be overcome if desired, by developing a hybrid
Electromagnet force control 61
control approach combining flux derivative feedback, for stability, with the proposed
scheme, for force linearity. However, the test results from the experimental
implementation of the proposed force control strategy show that an acceptable
performance has been achieved. Therefore, this force control strategy will now be used
by the suspension control systems described in Chapters 4 and 5.
Suspension mode control 62
4
Suspension mode control
4.1 Introduction
The previous chapter describes an electromagnet control scheme which is capable of
producing an electromagnetic force actuation that is dominantly linear and stable. This
chapter now develops the sophisticated suspension mode control strategy proposed in
Chapter 1 and employs the aforementioned force actuator to provide a non-contacting
suspension system.
The suspension mode controller is a component of the multi-electromagnet vehicle
control system described in Chapter 5. However, it is first developed and tested using
a single electromagnet suspension configuration in order to simplify the design and
verification stages.
The description of the development of the suspension mode control system is partitioned
into six sections. First, the functional requirements of the suspension are identified.
A sophisticated suspension control strategy is then proposed, and the strategy is
developed by analysing the proposed closed-loop suspension system and synthesising
each system component. The characteristic behaviour of the passive lateral
electromagnet guidance motion is then briefly analysed. Finally, some simulated and
experimental responses are presented and discussed, and conclusions are then drawn on
the performance of the proposed system.
4.2 Functional requirements
The primary functional requirement for an electromagnetic suspension system is to
follow the general guideway profile whilst providing a quality of ride consistent with
Suspension mode control 63
passenger comfort.61 This requirement calls for a control system with two conflicting
aims. Guideway following requires the suspension to be effectively coupled to the
track, whilst ride comfort considerations require the suspension to be isolated from track
irregularities.
Passenger ride comfort is a subjective measure which depends on the individual in
Figure 4.1 ISO reduced comfort boundary (for a 1 hour exposure)
question. However, it has been recorded that irrespective of the individual in question,
the human body is particularly sensitive to vibration over a frequency range of about
1-20 Hz. In 1974 the International Standards Organisation (ISO) produced a guide62
which incorporated a specification for vertical acceleration versus frequency
corresponding to the threshold between human comfort and discomfort. For urban
transport applications, the ISO’s one hour exposure characteristic is appropriate (see
Figure 4.1) for which the peak acceleration limit is 0.4 m/s2 over the frequency range
4-8 Hz.
For an electromagnetic suspension with no conventional secondary suspension (ie. no
springs and dampers), suspension travel is limited to the operating air gap range of the
electromagnet. This suspension configuration therefore requires a stiff track, with a
deflection restricted to about 50% of the normal operating air gap deviation.
Suspension mode control 64
The ride comfort analysis for vehicle suspensions is generally performed by considering
the track roughness.63 However, since the guideway for the proposed low-speed
system consists of stiff, smooth track sections, which are joined together, a ride quality
specification in terms of a worst case track joint misalignment is more convenient. A
suspension designed to accommodate significant track steps is desirable since it permits
cost savings in track production and maintenance.64 Therefore, after allowing for
operational air gap deviations, it is desirable for the suspension to accommodate a
worst-case track step size of 50% of the nominal operating air gap deviation. The
suspension acceleration when negotiating such a track step must be within the ISO
comfort specification (see Figure 4.1). To improve upon existing light rail systems,
which have a typical peak vertical acceleration65 level of 0.4 m/s2, a design target of
0.2 m/s2 is considered desirable for the proposed electromagnetic suspension system.
In addition to the track oriented functional requirements discussed so far, the suspension
system must reject disturbance forces due to variations in the weight of the payload,
wind gusts, and vertical forces generated by the linear inductance motor66 which
propels the vehicle along its guideway. Accommodation of disturbance forces of
30-50% of the maximum vehicle weight is typically required.67 Therefore, the design
target for the proposed system is for a transient air gap deflection of 30-50% of the
operating gap deviation, for a disturbance force equal to 30-50% of the maximum
suspended weight. In addition, there must be no steady-state air gap deflection due to
disturbance forces in order to maximise the available air gap deviation.
Having identified the functional requirements of the electromagnetic suspension, the
suspension system structure proposed in Chapter 1 is now examined and developed.
4.3 Suspension control strategy
The vehicle suspension control system structure proposed in Chapter 1 is applied to a
single electromagnet suspension as illustrated in Figure 4.2. The suspension strategy
involves the calculation of the absolute track position, which is then suitably processed
to provide an absolute position reference demand for the high stiffness electromagnet
position controller. The force actuation demand from the position controller is then fed
to the electromagnet force controller which is described in Chapter 3. Strategies for
designing the electromagnet position controller and the guideway following algorithm
are outlined next.
Suspension mode control 65
The electromagnet position controller is the heart of the suspension system. It must
Figure 4.2 Suspension control system hierarchy
control the absolute position of the suspension with zero steady-state error, and it must
also resist disturbance forces and accommodate load variations. The minimum
properties required of the position controller therefore include stiffness, damping, and
position error integral action.
Implementing this control system structure is complicated by the lack of a physical link
between the electromagnet and an absolute datum. Measurement of the absolute
position of the electromagnet is therefore not readily achievable, but absolute
acceleration can be cost-effectively measured using a standard industrial accelerometer.
The acceleration measurement must be integrated to give the absolute velocity, which
must in turn be integrated to give the absolute position. Pure integrators would suffer
from drift problems due to erroneous offsets, so a high-pass filter must augment each
integrator. Since acceleration is measured, the position controller can incorporate
acceleration feedback to assist in the control task. The proposed state-vector used for
feedback thus consists of position, and integral of position error, together with output
velocity and acceleration.
The absolute position of the track can be calculated using the absolute electromagnet
position and the electromagnet air gap measured using an industrial non-contacting
Suspension mode control 66
displacement sensor. The function of the guideway following algorithm is to receive
the track position and to transform it to a suspension position reference signal which
avoids contact with the track and which provides a comfortable ride quality. The
design of the guideway following algorithm depends on several factors. The most
important ones are the operational air gap deviation, the vehicle speed, the size of track
discontinuities, and finally, the parameters of track gradient entries and exits.
Since acceleration is the second derivative of displacement, a guideway following
algorithm consisting of a second-order low-pass filter can be used to limit acceleration
for a given size of track discontinuity on an otherwise flat guideway. However, Pollard
and Williams68 showed that the use of solely linear control algorithms presents
considerable difficulties and operational limitations when designing for gradient entries
and exits. It is therefore envisaged that the use of techniques such as
matched-filtering69 could provide better performance than traditional linear filtering.
For example, a matched-filter could be programmed with the functions used to define
the gradient entries and exits. These may consist of straight track segments, smoothly
curved sections, or a combination of both. Having identified the underlying guideway
profile curvature, linear filtering may then be suitable for rejecting discontinuities
superimposed on the curved profile.
If a guideway profile is encountered which cannot be comfortably negotiated, the
guideway following algorithm must attempt to limit the air gap deviation in order to
prevent the mechanical air gap limit stops from being reached. Similar techniques may
also be required for lateral guidance of the suspension.
The design of suitable guideway following algorithms, for both vertical and lateral
motion, represents a large research project in itself and is peripheral to the main
objective of this research work. Therefore, for the experimental research system, the
use of a second-order linear filter is proposed for the vertical motion guideway
following algorithm. In order to reduce the cost and complexity of experimental system
hardware, the lateral motion is not actively controlled.
The logical arrangement of the suspension system is outlined in Figure 4.3, and the
configuration of the proposed suspension control system is illustrated in Figure 4.4.
Table 4.1 lists the control system variables and parameters.
Suspension mode control 67
Inspection of Figure 4.4 shows that the position error signal consists of a component
Figure 4.3 Logical arrangement of the suspension system
Figure 4.4 Suspension control system configuration
of both the absolute position signal and the position signal relative to the track. The
proportion and frequency spectrum of each component is determined by the guideway
following algorithm. The coupling of the electromagnet suspension to the track is
therefore dominated by the guideway following algorithm. In order to be able to isolate
the suspension design procedure from the guideway dynamics, a sufficient margin must
be provided between the natural frequency of the track and the suspension to track
coupling frequency.
The analysis and synthesis of the position controller, the guideway following filter, and
the state integration filters are described next.
Suspension mode control 68
Table 4.1 Suspension control system nomenclature
Identifier Variable / Parameter
Z(s) Electromagnet absolute position (ˆ - calculated position)
P(s) Track absolute position (ˆ - calculated position)
C(s) Electromagnet to track air gap
Zref(s) Position controller reference signal
Fdemand(s) Position controller force demand
Fdisturb(s) Disturbance force input
kacc Acceleration feedback gain (virtual mass)
kvel Velocity feedback gain (damping)
kpos Position feedback gain (stiffness)
Terr Position error integral action time constant
ωint State integration filter corner frequency
Tforce Electromagnet force actuation time constant
m Suspended mass
4.4 Synthesis of the suspension control system
The proposed suspension control system is developed by analysing the closed-loop
system and then synthesising each system component. The position, velocity and
acceleration feedback gains are designed first, then an acceptable force actuation
bandwidth is determined. Next, a suitable error integral action time constant is chosen.
Finally, the guideway following algorithm is designed, and an acceptable cut-off
frequency for the state integration filters is determined. The development is concluded
by identifying a simplified transfer function which characterises the suspension system
response.
Suspension mode control 69
4.4.1 Position control transfer function
In order to be able to design the position controller and the guideway following
algorithm on a largely independent basis, two restrictions must be imposed on the
guideway algorithm. Firstly, the guideway algorithm must pass the track position signal
for frequencies below ωint so that the bandwidth of the position error signal extends
down to dc. Also, the guideway following algorithm must have a gain not exceeding
unity at all frequencies so that the position error feedback gain is determined solely by
kpos. In practice, these are fundamental requirements of all guideway following
algorithms and therefore they do not impose any operational restrictions.
The Laplace transform of the full position control law illustrated in Figure 4.4 is given
by:
The acceleration of the suspended load due to the control force demand and the
4.1
Fdemand
(s)
kpos
kpos
sTerr
s 2 Z(s)
s ωint
2Z
ref(s) k
vel
s 2 Z(s)
s ωint
kacc
s 2 Z(s)
disturbance force (see Figure 4.4) is given by:
The Laplace transfer function for the full closed-loop position controller is obtained by
4.2s 2 Z(s)F
demand(s)
m 1 sTforce
Fdisturb
(s)
m
substituting Equation 4.1 into Equation 4.2. After rearranging and collecting terms (see
Appendix C), the reference position transfer function is given by:
and the disturbance force transfer function is given by:
4.3
Z(s)
Zref
(s)
kpos
s 1/Terr
s ωint
2
s 6 mTforce
s 5 kacc
m 2ωint
mTforce
s 4 kvel
2ωint
(kacc
m) ω2
int mTforce
s 3 kpos
ω2
int(kaccm) ω
intk
vels 2 k
pos/T
err
The reference position transfer function has 6 poles and 3 zeros so the closed-loop
4.4
Z(s)
Fdisturb
(s)
1 sTforce
s ωint
2
s 5 mTforce
s 4 kacc
m 2ωint
mTforce
s 3 kvel
2ωint
(kacc
m) ω2
int mTforce
s 2 kpos
ω2
int (kaccm) ω
intk
vels k
pos/T
err
position control system is a sixth-order system. However, since the guideway following
algorithm passes frequencies below ωint, the position feedback signal can be assumed
Suspension mode control 70
to be dc coupled. This results in the transfer function zeros due to the state integration
filters moving to the origin of the s-plane, where they cancel with the poles at the
origin. The closed-loop response is therefore effectively that of a fourth-order system,
with two poles due to the sprung mass, one pole due to the position error integral
action, and another pole is contributed by the force actuator. Furthermore, the high
stiffness required of the position controller permits the use of a slow error integration
time constant. The position controller can therefore be designed as a dominantly
third-order system with a fourth pole located close to the origin due to the error integral
action. If the force actuation time constant is sufficiently small, then the system
reduces to a dominantly second-order system. However, since the electromagnet force
actuation bandwidth is limited by cost and complexity considerations, the force
actuation pole cannot be neglected.
Synthesis of the position control algorithm can be considerably simplified if the
dominant response of the controller is that of a second-order system.70 To permit this
design approach, the position control system is reduced to a second-order system by
assuming infinite force bandwidth, by neglecting error integral action, and by assuming
that the suspension velocity and position are measured directly. After designing the
state feedback gains for the reduced-order system, the constraints for the neglected
terms are calculated to ensure an acceptable full closed-loop transfer function.
Since the guideway following algorithm determines the passenger ride characteristic,
the detailed characteristic behaviour of the position controller transfer function is not
critical. The primary requirements are that the position controller bandwidth is
sufficiently high, and that the response is adequately damped. In practice, the high
stiffness required for the force disturbance rejection ensures sufficient bandwidth, and
the damping is an independent design factor.
The reduced-order position transfer function is obtained by substituting values of
Tforce = 0, Terr = ∞ and ωint = 0 into Equations 4.3 and 4.4. This substitution gives:
The characteristic behaviour of this second-order transfer function can be conveniently
4.5Z(s)
kpos
s 2 (kacc
m) skvel
kpos
Zref
(s)1
s 2 (kacc
m) skvel
kpos
Fdisturb
(s)
expressed in terms of its undamped natural frequency, ωn, and damping ratio, ζ. The
standard form for a second-order transfer function and the location of its poles is given
by:
Suspension mode control 71
The undamped natural frequency and damping ratio of the closed-loop position control
4.61
s 2/ω2
n 2ζ s /ωn
1with poles at: s ω
nζ ±ω
nζ2 1
transfer function (Equation 4.5) are thus given by:
4.7ωn
kpos
kacc
mζ
kvel
2 (kacc
m)kpos
4.4.2 Position, velocity and acceleration feedback gains
The most exacting requirement for the position controller is the rejection of disturbance
forces. The controller must limit the transient deflection due to a force disturbance of
50% of the full load force, to half of the maximum suspension deviation (see
Section 4.2).
The maximum total suspended load for the experimental electromagnet is 50 kg and its
nominal operating air gap is 3 mm with an operational deflection of up to ±2 mm (see
Table 3.1). Therefore, accommodation of a 50% load change, within half the
operational air gap deflection, requires a suspension stiffness, kpos, given by:
kpos = 50% × 50 kg × 9.81 Nkg-1 ÷ 1 mm ≈ 250 N/mm
If kacc is initially assumed to be zero, kpos = 250 N/mm and m = 15 kg, then the
undamped natural frequency of the position controller (see Equation 4.7) is 20 Hz,
while for m = 50 kg, the undamped natural frequency falls to 11 Hz. The use of
acceleration feedback can reduce this variation in natural frequency due to a load
change by providing ‘virtual mass’. Acceleration feedback is functionally equivalent
to force feedback, so it effectively increases the force bandwidth and enhances linearity
and stability. However, due to the constrained achievable force bandwidth (see
Chapter 3), kacc must be kept to a sensible minimum. The breakpoint for the
effectiveness of acceleration feedback in ameliorating the impact of load changes occurs
when kacc = minimum mass, so this is proposed as an acceptable compromise. With
kacc = 15 kg, the undamped natural frequency is reduced to 15 Hz and 10 Hz at no load
Suspension mode control 72
and full load respectively. The acceleration feedback gain is effectively equivalent to
a force feedback gain of 1 at the minimum load.
Since the underlying electromagnetic force actuator is open-loop unstable, a well
damped response is considered prudent in order to produce a robust suspension which
is insensitive to force actuation errors by the electromagnet force controller. In
addition, since the operational air gap range is limited, overshoot of the position
response is undesirable. Therefore, a worst case damping ratio given by the critical
damping ratio71 of 0.707 is required. To allow a margin for feedback gain deviations
due to transducer errors of up to ±25%, a minimum nominal value for ζ of 0.8 is
required. This requires a velocity feedback gain (see Equation 4.7) of
kvel = 6500 N/(m/s), which gives a nominal damping ratio of 0.8 and 1.2 at full and
minimum load respectively. In practice, the minimum damping ratio will be higher if
the suspended load is a passenger. This is because the passenger’s mass will not be
rigidly coupled to the vehicle at the natural frequency of the position controller.
4.4.3 Force actuation bandwidth
Having designed the response of the reduced-order position controller, the required
force actuation bandwidth is now determined. The transfer function for the position
control system, neglecting only the low frequency components due to the state
integration filters and position error integral action, is obtained by substituting values
of Terr = ∞ and ωint = 0 into Equation 4.3. This substitution yields:
Table 4.2 lists the location of the closed-loop poles for minimum and maximum
4.8Z(s)
Zref
(s)
kpos
s 3 mTforce
s 2 (m kacc
) skvel
kpos
suspended masses over a range of force actuation bandwidths. The location of the poles
is strongly influenced by both the force actuation bandwidth and the suspended mass,
and the damping ratio relates to the complex conjugate pole pair or the dominant pair
of real poles. As expected, the table shows that the damping ratio falls sharply once
the force actuation bandwidth approaches that of the reduced-order position control
system. To ensure an adequately damped response, and to allow a margin for
transducer errors, a force actuation bandwidth of about 50 Hz is required, which is
equivalent to a force time constant, Tforce = 3.2 ms.
Suspension mode control 73
Table 4.2 Variation of closed-loop poles with force bandwidth
Force
bandwidth
Closed-loop poles
(m=15kg)
Closed-loop poles
(m=50kg)
Hz rad/s Location ζ Location ζ
∞ ∞ -∞, -167, -50 1.19 -∞, -50 ±j37 0.80
1000 6283 -12347, -170, -50 1.19 -8067, -50 ±j37 0.80
100 628 -993, -214, -49 1.28 -708, -55 ±j38 0.82
50 314 -290 ±j155, -48 0.88 -284, -62 ±j41 0.83
40 251 -227 ±j188, -48 0.77 -190, -69 ±j44 0.84
30 188 -165 ±j197, -47 0.64 -88, -78 ±j67 0.75
20 125 -102 ±j186, -46 0.48 -63, -50 ±j86 0.50
4.4.4 Position error integral time constant
The final feedback gain factor to be determined is the time constant for the position
error integral action. Since the stiffness of the controller is high, the functional
requirement for the integral action, that of eliminating steady-state position errors, can
be performed quite slowly. This enables the closed-loop pole due to the integral action
to be placed close to the origin of the s-plane where it has negligible impact on the
other system poles. The slowest closed-loop pole is located at s = -48 (see the
highlighted row in Table 4.2), and has a time constant of 21 ms. Therefore, a value for
the integral action time constant, Terr, of 1 second will cause negligible disturbance of
the other closed-loop poles. Position error integral action is thus introduced without
impairing the dynamic response or damping of the position controller.
4.4.5 Guideway following algorithm
The guideway following algorithm is required to limit the acceleration due to a ±1 mm
step change in track height to about ±0.2 m/s2 (see Section 4.2). Since acceleration is
the second derivative of position, the proposed second-order low-pass filter can be
Suspension mode control 74
designed to limit the acceleration to any desired level for a given track step size. The
Laplace transform of such a filter, with a damping ratio of 1, is given by:
where ωfollow is the guideway following filter corner frequency. For a step input of
4.9Guideway_filter(s)ω
follow
s ωfollow
2
amplitude ∆pos, the acceleration of the filter response is given by:
The corresponding acceleration response in the time domain is given by:
4.10Acc(s) s 2 ∆pos
s
ωfollow
s ωfollow
2∆pos
s ωfollow
s ωfollow
2
Inspection of this time domain response reveals that the peak acceleration occurs at time
4.11acc(t) ∆pos ω2
follow eω
followt
1 ωfollow
t
t=0. The guideway filter time constant required to limit the acceleration to accmax is
thus obtained by setting t=0 and acc(t)=accmax. After rearranging, this gives:
Therefore, to accommodate a track step size of ±1 mm, with a peak acceleration of the
4.12ωfollow
accmax
∆pos
filter output of ±0.2 m/s2, a low pass filter, with two poles at s = 10,10 is required.
However, the poles of the closed-loop position controller transfer function limit the
initial acceleration of the suspension to zero. Since an exact numerical solution for the
maximum acceleration of the full suspension control system is unduly complicated, the
two pole guideway following filter is simulated along with a representative pole from
the position controller transfer function. Examination of Table 4.2 shows that the
position controller can be effectively modelled for this purpose by a single pole at
s = 48. The simulation results show a peak acceleration for three poles at s = 10,10,48
of 0.05 m/s2, which is much lower than that required. The guideway following filter
poles are therefore moved so that s = 25,25,48 where they produce a peak acceleration
of 0.21 m/s2. The bandwidth of the guideway following filter, ωfollow is thus set to 25
rad/s, and the overall track/suspension position response is approximated by three poles
at s = 25,25,48.
Suspension mode control 75
To prevent the suspension system from exciting track vibrations, the natural frequency
of the track must be above the track following frequency of 4 Hz.
4.4.6 State integration filters
The position controller synthesis has so far assumed dc coupled feedback signals for
acceleration, velocity and position. However, since the state integration is augmented
with high-pass filtering, it is necessary to determine the constraints that must be
imposed upon the filter corner frequency in order to limit the dislocation of the
closed-loop poles by the filters.
Neglecting the high frequency component due to the force actuation time constant, the
transfer function of the closed-loop position control system is obtained by substituting
the value Tforce = 0 into Equation 4.3. This substitution gives:
The integration filter corner frequency, ωint, is present in the characteristic polynomial
4.13
Z(s)
Zref
(s)
kpos
s 1/Terr
s ωint
2
s 5 kacc
m s 4 kvel
2ωint
(kacc
m) s 3 kpos
ω2
int(kaccm) ω
intk
vels 2 k
pos/T
err
as a factor in the s4 and s3 terms. These terms are therefore used to determine
constraints on ωint such that the characteristic polynomial is not materially altered by
the integration filters. The first term produces the constraint given by:
The second constraint is simplified using the first constraint and is given by:
4.142ωint
(kacc
m) kvel
∴ ωint
kvel
2(kacc
m)
For the controller feedback gains determined earlier (see Table 4.3), the filter corner
4.15ωint
ωint
(kacc
m) kvel
kpos
∴ ωint
kpos
kvel
frequency, ωint, must be much lower than 38 rad/s.
An additional constraining factor is due to the bandwidth of the guideway following
algorithm. Since this provides a relative position signal at frequencies below 25 rad/s
(see previous section), the state integration filters must pass frequencies above 25 rad/s.
Suspension mode control 76
Therefore, ωint must be much less than 25 rad/s in order to be able to assume that the
position feedback is dc coupled, and hence assume that the state integration filter zeros
cancel the poles at the origin of the closed-loop transfer function (see Equation 4.13).
The low frequency limit that can be achieved in practice is determined by the system
implementation (see Chapter 6). The limiting factors include the hysteresis,
non-linearity, and resolution of the accelerometer and its analogue-to-digital converter,
along with the numerical techniques and precision used to implement the filters. Since
achieving a very low frequency response is costly, an acceptable performance and cost
trade-off is required.
Experimentation with a sensor which met the other system requirements demonstrated
that a low frequency cut-off of 0.6 rad/s is achievable. With ωint = 0.6 rad/s, the two
state integration filters reduce the position signal amplitude by a total of 5-7.5%
at 15-25 rad/s. Due to offsets present within the analogue parts of the experimental
system, two additional high-pass filters are applied to the measured acceleration signal
before it can be used. For frequencies of 15-25 rad/s, this gives rise to a total loss of
position signal amplitude of 10-15%. An overshoot of approximately this size is
therefore expected on the closed-loop suspension position response to a track step. The
size of the overshoot is increased further by the velocity feedback signal. If desired,
it may be possible to compensate for most of the position signal gain loss by increasing
the gain of the air gap signal to match the low frequency gain loss of the position
signal.
4.4.7 Suspension controller design specification
By adhering to the constraints identified in the preceding section for the state
integration filter bandwidth, the closed-loop transfer functions for the position controller
(see Equations 4.3 and 4.4) can be expressed more simply by:
and
4.16Z(s)
Zref
(s)
kpos
(s 1/Terr
)
s 4 (mTforce
) s 3 (kacc
m) s 2 kvel
skpos
kpos
/Terr
Suspension mode control 77
The closed-loop suspension/track position transfer function is obtained by augmenting
4.17Z(s)
Fdisturb
(s)
s (1 sTforce
)
s 4 (mTforce
) s 3 (kacc
m) s 2 kvel
skpos
kpos
/Terr
Equation 4.16 with Equation 4.9. This gives:
This is dominated by the guideway following filter and can therefore be approximated
4.18Z(s)
P(s)
ω2
follow kpos
(s 1/Terr
)
s ωfollow
2 s 4 (mTforce
) s 3 (kacc
m) s 2 kvel
skpos
kpos
/Terr
by Equation 4.9 alone, giving:
The suspension control system parameters designed earlier are summarised in Table 4.3,
4.19Z(s)
P(s)≈
ω2
follow
s ωfollow
2
and the closed-loop poles and zeros that these produce (see Equation 4.18) are listed
in Table 4.4.
Table 4.3 Suspension control system parameters
Parameter Name Value
Stiffness kpos 250 N/mm
Damping kvel 6.5 N/(mm/s)
Virtual mass kacc 15 kg
Error integral time constant Terr 1 s
Force actuation time constant Tforce 3.2 ms
State integration filter frequency ωint 0.6 rad/s
Guideway following frequency ωfollow 25 rad/s
Suspension mode control 78
4.5 Lateral guidance
Table 4.4 Closed-loop suspension system zeros and poles
MassPosition Controller Guideway Filter
Zero Poles ζ complex.poles Poles
15 kg -1 -290 ±j157, -48, 0 0.88 -25, -25
50 kg -1 -281, -63 ±j40, 0 0.84 -25, -25
In order to reduce the complexity of the implementation of the experimental vehicle and
single electromagnet rig, provision has not been made for lateral force control. The
problem of controlling the lateral force is minor compared with that of controlling the
lift force, since the lateral behaviour of the electromagnetic suspension is open-loop
stable.
Without active control, the suspension experiences lateral stiffness due to the geometry
of the shear flux between the electromagnet and track poles. However there is
negligible lateral damping so a brief analysis to ascertain the lateral stiffness and natural
frequency of the lateral motion is considered prudent.
For lateral offsets of up to 2/3 of the electromagnet pole width, the lateral force (see
Equation 2.10) is approximated by:
where Fmagnet is the gross electromagnet force, c is the air gap, y is the lateral offset, and
4.20Flateral
Fmagnet
2c
πptan 1
y
c
p is the pole width. The lateral stiffness can therefore be approximated by:
where mt is the total suspended mass and ag is the acceleration due to gravity.
4.21klateral
Flateral
ym
ta
g
2c
πpytan 1
y
c
Although this is a nonlinear function, it can be linearised for small lateral offsets by:
Suspension mode control 79
4.22klateral
≈2m
ta
g
πpfor y<c
The laterally sprung mass is therefore assumed to have a dominantly second-order
response, with the natural undamped angular frequency (see Section 4.4.1) of the lateral
motion approximately given by:
where mc is the effective, rigidly coupled suspended mass.
4.23ωn
≈2m
ta
g
mcπp
For any suspension load which is rigidly coupled to the electromagnet, the model
predicts a natural undamped frequency of 4 Hz. However, with a human passenger
load, the natural frequency of the human body lateral coupling is likely to be below
4 Hz, and so the natural undamped frequency of the lateral motion could be as high as
7 Hz. In practice, it is likely to lie somewhere in the range of 4 Hz to 7 Hz. The
natural lateral suspension frequency is thus similar to that designed for the vertical track
following algorithm. A beneficial side effect of human body to vehicle coupling for
the experimental systems is that it provides a degree of lateral damping.
The presence of the air gap in the lateral force model (see Equation 4.20) presents the
interesting possibility of providing lateral damping by controlling the lateral force
through modulation of the air gap. Clearly, such a technique would impose a
disturbance on the vertical motion, but it would provide lateral damping without the
need for additional electromagnets. However, since no problems have been experienced
with regard to lateral motion oscillations, this technique has not been investigated.
4.6 Performance of the experimental mode suspension
In order to validate the theoretical basis of the design of the proposed suspension
control system, the system was simulated and various step responses were obtained and
examined. An experimental single electromagnet suspension system was then
developed (see Chapter 6) and the experimental system was tested by comparing a
number of simulated and experimental responses.
Suspension mode control 80
The control system was simulated using the Advanced Continuous Simulation Language
(ACSL)72 which provides high level simulation constructs and has powerful signal
monitoring capabilities. The model of the suspension control system is listed in
Appendix C and incorporates a discrete-time suspension controller with analogue-digital
conversion quantisation, and continuous-time models of the electromagnet and control
system transducers.
Figure 4.5 (Figure 3.7 repeated) shows the physical arrangement of the experimental
Figure 4.5 Single electromagnet suspension experimental rig
single electromagnet test rig in which the long pivoted beam allows the electromagnet
and sensors to move with negligible stiffness and damping in the vertical direction. The
control system was designed using the suspension configuration shown in Figure 4.4
with the parameters defined in Table 4.3.
Step changes in track height are simulated by injecting an offset into the track profile
calculation. The full bandwidth response of the position controller is tested by
bypassing the guideway following algorithm, thus effectively setting the guideway
following filter bandwidth to infinity. This results in the position error signal consisting
solely of the air gap signal with no contribution from the absolute position signal. After
testing the position controller response, the system is reconfigured with the correct
guideway following filter to verify that the suspension meets the required ride comfort
specification.
4.6.1 Position controller
Figure 4.6 and Figure 4.7 show the simulated and experimental position controller
responses respectively, for a 1 mm step size, and a suspended mass of 15 kg.
Figure 4.8 and Figure 4.9 show the respective responses for a suspended mass of 45 kg.
Suspension mode control 81
The air gap responses are well damped, with a small, very low frequency overshoot.
Figure 4.6 Simulated suspension response to a 1 mm air gap reference step
(ωfollow = ∞ rad/s, m = 15 kg)
Figure 4.7 Experimental suspension response to a 1 mm air gap reference step
(ωfollow = ∞ rad/s, m = 15 kg)
The 7% overshoot on the simulated responses is due solely to the high-pass filtering of
the velocity feedback signal. This is augmented on the experimental responses by an
additional 5% overshoot due to the force actuation error as the electromagnet operating
point changes with the air gap. The experimental response for the 45 kg suspended
mass has a further 5% overshoot (bringing its total to 17%) which can be attributed to
the damping ratio being slightly below the critical damping ratio.
The sharp peaks for the acceleration responses are caused by current slew rate limiting
in the electromagnet current controller. This occurs because the force slew rate is
Suspension mode control 82
designed for comfortable passenger acceleration levels rather than the high levels
Figure 4.8 Simulated suspension response to a 1 mm air gap reference step
(ωfollow = ∞ rad/s, m = 45 kg)
Figure 4.9 Experimental suspension response to a 1 mm air gap reference step
(ωfollow = ∞ rad/s, m = 45 kg)
experienced during these tests. The time to the acceleration peak is approximately 9 ms
for both the experimental and simulated responses. The peak acceleration amplitudes
of the experimental responses are 10-20% lower than the simulated responses which
suggests that the bandwidth or force slew rate of the experimental system is slightly
lower than the design target.
The low amplitude oscillation superimposed on the transient experimental acceleration
responses (up to time = 150 ms) is partly due to the current slew rate limiting, but is
mostly due to the force pulse causing the experimental rig to vibrate. The natural
Suspension mode control 83
frequency of the rig is about 60 Hz and the oscillation decays rapidly. However, the
natural frequency can be reduced down to about 10 Hz by using flexible rubber
mountings. As expected, at rig natural frequencies close to that of the position
controller (approximately 10-15 Hz), the rig could be excited with a step position
reference to produce continuous steady-state oscillation. The rig oscillation presented
no electromagnet stability problems, but it is obviously unacceptable. Setting the rig
natural frequency to double that of the air gap controller is sufficient to cause any
oscillation to decay rapidly.
The acceleration ripple which is apparent in the steady state on both the experimental
and simulated responses is due to limit cycling73 of the air gap. The period and
magnitude of the limit cycle is primarily a function of the air gap measurement
resolution and hysteresis, and conversion quantisation, but it is also affected by the
controller sampling period and the electromagnet operating point. The amplitude of the
limit cycles for the experimental responses is approximately one third that of the
simulated responses. This is attributed to measurement noise in the experimental
system increasing the effective analogue to digital conversion resolution.74
The results discussed above give confidence in the linearity and bandwidth of the
electromagnet force actuator developed in Chapter 3. However, additional responses
were obtained to test the performance of the suspension over its full operational
envelope. Figure 4.10 presents the results of a test in which three step responses were
obtained, each stepping up and then down by 1 mm, from initial air gaps of 1.5, 2.5 and
3.5 mm. The consistent responses demonstrate the fact that the electromagnet force
actuator provides an acceptably linear response over the full operational air gap range.
The disturbance force rejection requirement of the suspension is tested by rapidly
applying and removing a 15 kg load. This produces a peak air gap deflection of
±0.73 mm, whilst the expected theoretical deflection is ±0.6 mm. The 20% discrepancy
is attributed to sensor and force actuation errors. The air gap deflection recovers by
90% within 2 seconds of the application of the force disturbance, and is subsequently
eliminated.
The overall correspondence between the steady-state and dynamic characteristics of the
simulated and experimental responses is good, and the air gap response is well damped.
Further position controller test results, including disturbance force rejection and the
frequency domain response are presented in Chapter 5 for the multi-electromagnet
vehicle.
Suspension mode control 84
Figure 4.10 Experimental suspension responses to three 1 mm air gap reference steps
(ωfollow = ∞ rad/s, m = 30 kg)
4.6.2 Full suspension system
The ability of the suspension control system to meet the required ride comfort level is
now examined. Since the reaction rail of the experimental electromagnet rig cannot be
moved, track steps are simulated by injecting a step disturbance into the track position
calculation before the guideway following algorithm. Figure 4.11 and Figure 4.12
illustrate the simulated and experimental responses respectively, for a (simulated) 1 mm
step in track height, with a suspended mass of 15 kg. Figure 4.13 and Figure 4.14
show the respective responses for a suspended mass of 45 kg.
The peak acceleration of the simulated responses is 0.21 m/s2 which is very close to the
design value of 0.21 m/s2. The peak accelerations of the experimental responses are
0.28 and 0.30 m/s2 for the 15 and 45 kg suspended masses respectively, which are 27%
and 36% above the design target. Since the peak acceleration is quite sensitive to the
location of the position controller poles, the experimental acceleration levels are
attributed to errors in the assumed position controller poles. These errors arise due to
inaccuracies in the sensor measurements in general, and the electromagnet force actuator
in particular. The response for the 45 kg mass shows evidence of a low amplitude
oscillation of the experimental rig at a frequency of about 5 Hz. Both experimental
acceleration responses are however, well within the ISO ride comfort specification of
0.4 m/s2, and a consistent response is achieved even when the suspended mass is tripled.
The sensitivity of the acceleration response to the location of the position controller
poles can be ameliorated through the use of a more complex guideway following
Suspension mode control 85
algorithm, for example, by using a third-order filter.
Figure 4.11 Simulated suspension response to a 1 mm simulated track step
(m = 15 kg)
Figure 4.12 Experimental suspension response to a 1 mm simulated track step
(m = 15 kg)
The low frequency overshoot on the air gap step responses for both the simulated and
the experimental results is around 23%. The overshoot is dominantly due to the state
integration filters which reduce the amplitude of the position and velocity feedback
signals at low frequencies. The 23% overshoot is close to that anticipated with a
10-15% contribution from the position signal (see Section 4.4.6) plus a 7% contribution
from the velocity signal which was observed in the position controller test. For the
experimental system, this overshoot is considered acceptable.
Suspension mode control 86
The track step response test described above is in fact rather harsh, since a step change
Figure 4.13 Simulated suspension response to a 1 mm simulated track step
(m = 45 kg)
Figure 4.14 Experimental suspension response to a 1 mm simulated track step
(m = 45 kg)
in track height for a moving electromagnet is transformed to a ramp change as the
electromagnet passes by the step. Since full-scale suspension magnets are quite long75
(typically 1 m, for speeds of 0-25 m/s) this transformation adds a further filter to the
system, which can be approximated by a first order lag with a pole frequency of
0-25 rad/s (speed divided by electromagnet length). A real track step would therefore
produce a lower peak acceleration than the simulated track steps.
Suspension mode control 87
4.7 Conclusions
A novel suspension control scheme has been proposed, developed, and tested, which
permits the conflicting requirements of disturbance force rejection and guideway
following to be designed independently. Simulated and experimental test results show
that the system meets the functional requirements, and that the response is stable and
well damped.
For a full-scale system, it may be desirable to reduce the air gap overshoot associated
with step changes in the track height. Further research investigating higher performance
accelerometers, more complex state integration filters, and air gap gain compensation
should permit a reduction in the air gap overshoot.
The development of a sophisticated guideway following algorithm suitable for use on
guideways with gradients is an area that requires further research. The use of
matched-filter techniques is considered to present potential improvements over low-pass
filtering. This research could be performed very largely through simulation studies.
Having established the validity of the proposed suspension control system, it is now
ready to be applied to the suspension control of the experimental multi-electromagnet
vehicle.
Vehicle suspension control 88
5
Vehicle suspension control
5.1 Introduction
This chapter describes the development and validation of the multi-electromagnet
vehicle suspension control strategy outlined at the end of Chapter 1. The suspension
controller developed in Chapter 4 is used to control independent vehicle modes, and the
force controller developed in Chapter 3 is used to provide independent electromagnet
force actuation.
The development of the vehicle control system is divided into six parts. First, the
characteristics of the vehicle chassis and guideway are identified. The decoupling and
control requirements are then considered and a vehicle suspension control strategy is
developed. Next, the control system is synthesised using the independent force and
suspension controllers developed in Chapters 3 and 4 respectively. The lateral motion
of the vehicle is then briefly analysed. Experimental test results are presented next, to
verify that the proposed vehicle suspension system meets the operational requirements.
Finally, some conclusions are drawn on the merits and limitations of the proposed
system.
5.2 The experimental research vehicle and guideway
The chassis of the experimental research vehicle is designed to be very stiff in order to
provide a rigid coupling between the electromagnets and this exacerbates the problems
of controlling the vehicle.76 Figure 5.1 shows a photograph of the experimental
vehicle and its guideway. The vehicle is suspended by four electromagnets, one at each
corner, and is propelled and braked by a linear induction motor mounted centrally
Vehicle suspension control 89
underneath the chassis. The vehicle is equipped with all of the signal processing and
power control equipment necessary to implement the proposed control schemes.
The chassis is constructed from welded steel tubing, with an aluminium alloy used for
Figure 5.1 Vehicle chassis
the electromagnet hangers to prevent conduction of the electromagnet flux. The
torsional stiffness of the chassis, measured axially about the length of the vehicle is
about 112 kNm/rad. The flexibility of the electromagnet support hangers gives rise to
a coupling stiffness of about 1.5 kN/mm between each electromagnet pole face and the
vehicle chassis. The chassis is assumed to be rigid apart from the torsional and
electromagnet hanger flexibility.
The experimental electromagnets and reaction rail are too narrow to permit direct
measurement of the electromagnet air gap. Therefore, the gap sensors measure the
distance from the chassis down to the top of the track, and the electromagnet air gaps
are calculated using an appropriate formula. This arrangement is not ideal due to the
flexibility of the electromagnet hangers which introduces small errors in the air gap
measurement. In order to limit the magnitude of such errors, the air gap sensors are
calibrated with a preloaded mass of 35 kg per electromagnet. A suspended mass of
20-50 kg per electromagnet thus gives rise to a maximum measurement error of
±0.1 mm due to steady-state hanger deflection. The accelerometers are rigidly mounted
Vehicle suspension control 90
to the underside of the electromagnets and are therefore assumed to be perfectly
coupled.
The guideway is constructed from stiff steel girders which carry ferromagnetic reaction
rails for the electromagnets and a steel-backed, aluminium alloy reaction rail for the
linear motor. When excited with a force impulse, the guideway exhibits lightly damped
oscillation modes at natural frequencies of around 20 Hz and 40 Hz.
The mass of the fully equipped vehicle is 88 kg which permits a maximum passenger
load of about 110 kg. Table 5.1 lists the mass of each of the major vehicle components
and the dimensions between the centres of the electromagnets. The linear motor which
propels the vehicle produces a maximum thrust of about 50 N and an associated
repulsion force of around 200 N.
Table 5.1 Component masses and dimensions of the experimental vehicle
Vehicle parameter Index Size
Chassis Mass mchassis 27 kg
Total Electromagnet Mass mmagnets 29 kg
Linear Induction Motor Mass mLIM 14 kg
Control System Equipment Mass mcontrol 18 kg
Total Vehicle Mass m 88 kg
Chassis length (between electromagnet centres) L 0.8 m
Chassis width (between electromagnet centres) W 0.4 m
5.3 Control strategy for the vehicle suspension
The free-body motion of the vehicle has six degrees of freedom which can be
considered in terms of three cartesian modes of linear motion, namely, heave, sway and
track progress, and the three corresponding cartesian rotation modes, pitch, roll and
yaw. These modes are illustrated in Figure 5.2, along with the electromagnet indices.
Vehicle suspension control 91
The experimental vehicle has no provision for active lateral force actuation, so the sway
Figure 5.2 Free-body motion of the vehicle
and yaw modes are not actively controlled. Lateral guidance is, however, provided by
the inherent lateral stiffness which exists between the suspension electromagnets and
the reaction rails, whilst the progress of the vehicle along the guideway is controlled
by the linear induction motor. The four lift suspension electromagnets on the
experimental vehicle therefore control three free-body motions, namely, heave, pitch and
roll. The redundancy associated with four electromagnets controlling only three
free-body motions, results in the control of a fourth degree of freedom, namely,
torsional distortion of the vehicle chassis.
The suspension requirements for the independent vehicle modes are functionally
equivalent to those identified in Chapter 4, but different suspension parameters are
required for the different vehicle modes. For example, in order to maintain the correct
nominal air gaps, and hence maximise the available air gap deviation range, the heave,
pitch and roll modes require position error integral action. However, an important
operational requirement for an electromagnetically suspended vehicle is for the
suspension forces to be evenly distributed among the electromagnets. This is required
because suspension electromagnets are designed with only a moderate overload
capability due to the weight penalty that it incurs (see Appendix A). Since a cost
effective vehicle and guideway configuration will always have some finite torsional
displacement error, the application of torsional position error integral action would
cause a severe load imbalance between diagonal pairs of electromagnets. This would
clearly be unsatisfactory from an electromagnet utilisation viewpoint. The application
of position error integral action to the vehicle torsion motion is thus precluded. For
full-scale vehicles, it may also be desirable to have different settings for the heave,
Vehicle suspension control 92
pitch and roll mode suspensions,77,78 as is found on conventional trains and road
vehicles.
Since the required suspension characteristics for the vehicle modes differ, independent
vehicle mode suspension controllers are required.79,80 The proposed vehicle control
strategy therefore employs the suspension control algorithm developed in Chapter 4 to
control independently each of the vehicle modes, heave, pitch, roll and torsion.
Figure 5.3 shows the configuration of the proposed vehicle suspension control strategy,
where C, Z and F represent air gaps, accelerations and forces/torques respectively. The
electromagnet feedback signals for air gap and acceleration are first transformed to
vehicle mode coordinates and then fed into independent mode suspension controllers.
The force and torque demands from the suspension controllers are then transformed to
electromagnet force demands. The electromagnet force controller proposed and
developed in Chapter 3 is then used to achieve independent electromagnet force
actuation. The synthesis of the proposed vehicle control strategy is described next.
Figure 5.3 Configuration of the vehicle suspension control system
5.4 Synthesis of the vehicle control system
The synthesis of the vehicle suspension control system is performed in three stages.
First, the transformations required to convert the electromagnet coordinate signals to and
from vehicle mode coordinates are identified. Then, for convenience of design, the
vehicle mode angular measurements are normalised so that they are equivalent to the
linear motions which they generate at the electromagnets. This is achieved by
reformulation of the decoupling transformations, and conversion of the vehicle mode
inertias. Finally, the parameters of the suspension control algorithm developed in the
previous chapter are configured for each vehicle mode.
Vehicle suspension control 93
5.4.1 Decoupling the electromagnet motions
The maximum angular displacement of any vehicle motion is less than 1°. Therefore,
the vehicle mode motions (see Figure 5.2) can be accurately approximated by
Equation 5.1 where h, θ, φ and ψ are the positions of the vehicle modes, heave, pitch,
roll and torsion respectively, zm1-zm4 are the positions of the electromagnets, and L and
W are the vehicle length and width measured between the electromagnet centres (see
Table 5.1).
5.1
h
θφψ
≈ 1
4
1 1 1 1
2/L 2/L 2/L 2/L
2/W 2/W 2/W 2/W
2/W 2/W 2/W 2/W
zm1
zm2
zm3
zm4
The corresponding transformation between vehicle mode forces and torques, and the
electromagnet forces is given by Equation 5.2. This transformation matrix reflects the
additive nature of force and torque translations, compared with the averaging nature of
displacement translations.
5.2
Fm1
Fm2
Fm3
Fm4
≈ 1
4
1 2/L 2/W 2/W
1 2/L 2/W 2/W
1 2/L 2/W 2/W
1 2/L 2/W 2/W
Fh
Tθ
Tφ
Tψ
The masses of the major components from which the vehicle is constructed are listed
in Table 5.1. In order to determine the contribution of each of these masses to the
inertia of the rotational vehicle modes, the distribution of the mass of each component
in a horizontal plane is considered. The heights of the component masses relative to
a horizontal reference plane are considered later.
The centres of gravity of each electromagnet and the induction motor are assumed to
be located at their respective centres of geometry. The electromagnets are located at
the four corners of the rectangular chassis, whilst the motor is located at the geometric
centre of the chassis. The mass of the chassis is concentrated around its periphery, so
Vehicle suspension control 94
it is modelled by four point masses, located at the electromagnet centres, with each
point mass equal to one quarter of the total chassis mass.
The mass of the control system electronics is distributed fairly uniformly over the
rectangular space between the electromagnets. It is therefore modelled by four point
masses, each of one quarter of the total electronics mass, which are located mid-way
between each electromagnet and the geometric centre of the chassis. Equation 5.3
shows the contribution of each component mass to the mass or inertia of each vehicle
mode. The resultant mass and inertias are calculated using the masses and dimensions
given in Table 5.1.
5.3
mh
Iθ
Iφ
Iψ
≈
1 1 1 1
(L/2)2 (L/2)2 (L/4)2 0
(W/2)2 (W/2)2 (W/4)2 0
(W/2)2 (W/2)2 (W/8)2 0
mchassis
mmagnets
mcontrol
mLIM
88 kg
9.7 kgm 2
2.4 kgm 2
2.3 kgm 2
5.4.2 Normalising the vehicle mode motions
For the sake of convenience in considering the design parameters for the vehicle mode
controllers, the rotational vehicle mode motions are normalised so that they are
equivalent to the linear motions which they produce at the electromagnets. The
reformulated transformations are given by Equations 5.4 and 5.5.
In a similar fashion, the inertia of each vehicle mode is normalised so that it is
5.4
zh
zp
zr
zt
≈ 1
4
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
zm1
zm2
zm3
zm4
where zp
Lθ2
, zr
Wφ2
, zt
Wψ2
equivalent to a point mass at the electromagnet centres. Equation 5.6 shows the
normalised mode masses reformulated from Equation 5.3. The angular torsional
stiffness of the vehicle chassis is also normalised using the factors in Equations 5.4
and 5.5 giving an equivalent linear stiffness of 1400 N/mm.
Vehicle suspension control 95
5.5
Fm1
Fm2
Fm3
Fm4
≈ 1
4
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
Fh
Fp
Fr
Ft
where Fp
2Tθ
L, F
r
2Tφ
W, F
t
2Tψ
W
5.6
mh
mθ
mφ
mψ
≈
1 1 1 1
1 1 1/4 0
1 1 1/4 0
1 1 1/16 0
mchassis
mmagnets
mcontrol
mLIM
88 kg
61 kg
61 kg
57 kg
where mθ
4Iθ
L2, mφ
4Iφ
W 2, mψ
4Iψ
W 2
5.4.3 Control of the vehicle mode motions
The centre of mass of the vehicle lies centrally above the horizontal plane linking the
accelerometers which are used to derive the position feedback signals. The pitch and
roll modes are therefore coupled to the heave mode.81 Since the height of the centre
of mass of the vehicle above the accelerometer plane is much smaller than the vehicle
length or width, the degree of mode coupling is low. The vehicle mass distribution
causes negligible cross-coupling between the pitch and roll modes, and from the heave
mode to the pitch and roll modes. Since the amount of vehicle mode cross-coupling
is small, decoupling is considered unnecessary, and the suspension controllers are
applied directly to the vehicle mode motions.
5.4.4 Configuration of the vehicle mode suspension controllers
The functional requirements for the vehicle suspension system are equivalent to those
discussed in Section 4.1. The first requirement is that the suspension should deflect no
more than 30-50% of the maximum allowable air gap deviation, when subjected to a
disturbance or load force equal to 30-50% of the maximum suspended weight. The
second requirement is for a peak acceleration of less than 0.04 g, when negotiating a
Vehicle suspension control 96
1 mm step in track height. In addition, in order to maximise the available air gap
deviation, there must be no steady-state air gap deflection due to disturbance or load
forces for the heave, pitch and roll modes.
The suspension control algorithm developed in Chapter 4 is now configured for each
of the vehicle modes. The electromagnets are rated for a maximum continuous force
of 500 N, so the maximum suspension force for the vehicle is 2000 N. Therefore, in
order to meet the disturbance force rejection requirement at any point on the chassis,
the heave, pitch and roll mode position controllers require a stiffness of 1000 N/mm.
In addition, in order to meet the deflection requirement with the disturbance force
equally divided between two diagonally opposed corners of the vehicle, the torsion
motion also needs a stiffness of 1000 N/mm. Since the chassis has an inherent damped
torsional stiffness of 1400 N/mm, active suspension stiffness for the torsion mode is not
required.
The stiffness of each electromagnet hanger is approximately 1500 N/mm which
translates to a vehicle mode hanger stiffness of 6000 N/mm since the hangers operate
in parallel. This is much larger than the stiffness required for the vehicle mode position
controllers, and it is therefore neglected when designing the suspension controllers.
The design procedure for the position controller described in Section 4.4 is applied for
the vehicle heave, pitch and roll modes. The design procedure uses as its starting point,
the mode stiffnesses determined above, the chassis mode masses calculated in
Equation 5.6, and the maximum passenger load of 110 kg. For the purpose of mode
position controller synthesis, the heave mode is conservatively assumed to have the
same minimum mass as the other modes. The linearity of the suspension control
algorithm and its design procedure, results in vehicle mode feedback gains that are four
times the size of those for the single electromagnet suspension design. Table 5.2 lists
the parameters for the heave, pitch and roll mode position controllers.
The natural frequencies of the various vehicle modes and components are now
considered to check for any potential resonance problems. The predicted closed-loop
suspension poles for the vehicle heave, pitch and roll modes are essentially the same
as those for the single electromagnet suspension, with a resultant position controller
bandwidth of around 10 Hz (see Table 4.4). By assuming a dominantly linear,
second-order behaviour for the vehicle torsion motion, its estimated undamped natural
frequency82 is given by:
Vehicle suspension control 97
Similarly, the estimated undamped natural frequency of each electromagnet and hanger
Table 5.2 Parameters for the heave, pitch and roll mode position controllers.
Parameter Feedback signal Value
Stiffness Position error 1000 N/mm
Damping Velocity 26 N/mm/s
Virtual mass Acceleration 60 kg
Integral time constant Position error integral 1 s
5.7ωtorsion
ktorsion
mtorsion
1400000
57≈ 157 rad/s, f
torsion≈ 25 Hz
is given by:
5.8ωhanger
khanger
mhanger
1500000
7.5≈ 450 rad/s, f
hanger≈ 71 Hz
The predicted natural frequency of the electromagnet hangers is well separated from the
design values for the vehicle heave, pitch, roll and torsion modes, so problems of
resonant coupling are unlikely to occur. However, one of the guideway resonance
modes has a natural frequency of around 20 Hz which is likely to interact with the
torsion mode. In view of this fact, and the open-loop unstable nature of the
electromagnetic force actuators, the use of velocity and acceleration feedback for the
torsion mode is considered prudent.
Therefore, the damping of each electromagnet is made equivalent to that of the
independent electromagnet suspension of Chapter 4 by using the same velocity feedback
gain for the torsion mode as is used for the other mode controllers. The same argument
is applied to the acceleration feedback.
The remaining vehicle suspension system design parameters are the corner frequencies
for the guideway following filters and state integration filters, and the force actuation
time constant. For the vehicle mode suspensions, these are 4 Hz, 0.1 Hz and 3.2 ms
Vehicle suspension control 98
respectively, which are the same as those for the single electromagnet suspension due
to the linearity of the position control algorithm and design procedure. Table 5.3
summarises the parameters for the full vehicle suspension control system.
Table 5.3 Parameters for the vehicle mode suspension controllers
Parameter Name Value
Stiffness* kpos 1000 N/mm
Damping kvel 26 N/(mm/s)
Virtual mass kacc 60 kg
Error integral time constant* Terr 1 s
Force actuation time constant Tforce 3.2 ms
State integration filter frequency ωint 0.6 rad/s
Guideway following frequency ωfollow 25 rad/s
* For the torsion mode suspension controller: kpos = 0 N/mm, Terr = ∞ s.
5.5 Lateral vehicle guidance
The inherent lateral stiffness of the electromagnets provides guidance forces for the
vehicle sway and yaw motions. The undamped natural frequency of the lateral motion
is derived for small lateral perturbations of the electromagnets in Chapter 4, and is
repeated here for convenience. It is approximated by:
where ωn is the undamped natural frequency, mt is the total suspended mass, mc is the
5.9ωn
≈2m
ta
g
mcπp
rigidly coupled suspended mass, ag is gravitational acceleration, and p is the width of
the electromagnet pole pieces. The mass ratio mt /mc is unity for the unloaded vehicle,
and approximately 2 when carrying a passenger. Since the pole width is 9.5 mm, the
undamped natural mode frequencies are around 4 Hz for the unloaded vehicle, and 6 Hz
for the vehicle supporting a passenger. The lateral guidance thus experiences a
second-order guideway following characteristic with a bandwidth very similar to that
Vehicle suspension control 99
of the actively controlled vertical modes, but without the high absolute position
stiffness.
5.6 Performance of the experimental vehicle suspension
In order to verify that the proposed vehicle suspension control system is capable of
meeting the desired suspension performance, an experimental vehicle was developed,
and experimental test results were obtained. The configuration and parameters of the
vehicle control system are described earlier in this chapter, whilst the suspension control
algorithm and electromagnet force control algorithm are described in Chapters 4 and 3
respectively. The implementation of the hardware and software for the experimental
vehicle is described in Chapter 6.
The performance of the experimental vehicle is tested in three phases. The first phase
tests the mode position controllers by examining the reference position step response,
the disturbance force rejection response, and the stability margins. The second phase
tests the degree of cross-coupling between the vehicle mode motions and examines the
load sharing performance. Finally, the third phase tests the suspension ride quality
when subjected to simulated track steps.
Selected tests were performed with the vehicle both stationary and in motion, and no
significant response differences were observed between the two cases.
5.6.1 Performance and stability of the mode position controllers
The performance of the vehicle mode position controllers is tested by analysing the
reference position step response and the force disturbance step response. The stability
margins are then measured by examining the response in the frequency domain using
a Bode plot.83 In addition, a series of position step responses at different air gaps is
presented to gauge the linearity of the suspension control system over the full
operational air gap range.
The position step response of the vehicle heave mode controller for a 1 mm reference
step amplitude is shown in Figure 5.4 for the unladen vehicle, and in Figure 5.5 for the
vehicle with an 80 kg passenger. The transient air gap response should be equivalent
to the simulated and experimental responses for the single electromagnet suspension
Vehicle suspension control 100
presented in Chapter 4 (see Figures 4.6 and 4.7). The 50% rise time for the vehicle is
Figure 5.4 Experimental heave response to a 1 mm position reference step
(no passenger)
Figure 5.5 Experimental heave response to a 1 mm position reference step
(80 kg passenger)
about 10% smaller than that for the single electromagnet suspension, but the responses
are otherwise consistent. The oscillation present on the acceleration signal has a
frequency of just under 70 Hz and is attributed to the electromagnet hangers which have
an estimated undamped natural frequency of 71 Hz (see Equation 5.8). The low
frequency overshoot after the transient part of the response has a peak value of 15%.
This is due mostly to the a.c. coupling of the absolute feedback signals, which causes
reduced position and velocity signal amplitudes at low frequencies (see Section 4.4.6).
Some of the overshoot is also contributed by the position error integral action adjusting
Vehicle suspension control 101
to the new force actuation error that results from the change in electromagnet operating
point.
Since the passenger mass is loosely coupled to the vehicle at the frequencies
encountered in the transient portion of the step response, the loaded and unloaded
responses are very similar. The consistency between the unloaded and loaded responses
also shows that the mode force actuation is dominantly linear over a wide force
actuation range.
A critical performance requirement for the vehicle suspension is disturbance force
rejection. This is tested by measuring the deflection when starting and stopping the
linear induction motor and when adding and removing a 500 N load. The induction
motor produces a lift force of about 200 N at a thrust of 40 N which causes the vehicle
to rise and fall transiently by 0.22 mm. This corresponds well with the theoretical
value of 0.2 mm due to the heave mode stiffness of 1000 N/mm. Operation of the
linear motor has a negligible impact on the heave mode position step response.
Figure 5.6 shows the air gap and acceleration response to a 50 kg vehicle load
reduction. This produces a deflection of 0.61 mm which again corresponds well with
the theoretical value of 0.5 mm. For both test cases, the disturbance response is well
damped with the position error integral feedback restoring the reference air gap in about
2 seconds.
Figure 5.6 Experimental heave response to a 500 N heave disturbance force
Vehicle suspension control 102
The air gap linearity of the mode suspensions is tested by comparing position step
responses at different air gaps. Figure 5.7 shows three position step responses for the
heave mode position controller. The response of each step is essentially the same,
although there is a slight variation in the overshoot recovery characteristic. The latter
effect is most notable at the small and large air gaps because the electromagnet force
controllers are less accurate near the operational air gap limits than they are around the
nominal operational air gap. However, the step responses show that the suspension is
dominantly linear over the full operational air gap range.
For the suspension stability analysis in the frequency domain, the worst case scenario
Figure 5.7 Experimental heave responses to three 1 mm heave position reference steps
is represented by the heave mode controller. The larger mass for the heave mode
relative to the equivalent masses of the pitch and roll modes gives the heave mode a
slightly lower damping ratio. Figure 5.8 and Figure 5.9 show the theoretical and
experimental Bode plots for the position response of the vehicle heave mode controller
for a 1 mm amplitude sinusoidal position reference. The theoretical plots are calculated
using Equation 4.8 which neglects the low frequency effects due to the state integration
filters and the position error integral action. In addition, no allowance is made for the
closed-loop phase delay which is generated by the discrete-time controller. For
example, the average signal processing time delay of 1.6 ms (see Section 6.5.5) causes
an effective feedback loop phase delay of 37° at a frequency 64 Hz.
As with the time domain responses, only a small difference was observed between the
Bode plots for the unloaded and passenger loaded vehicle due to the relatively flexible
coupling of the passenger body to the vehicle. The signal magnification at low
Vehicle suspension control 103
frequencies is due to the low frequency gain roll-off of the state integration filters
which are used to calculate the absolute velocity and position feedback signals.
In practice, the resonance characteristics of the electromagnet hangers and the track are
Figure 5.8 Bode plot (gain) for the position response of the heave mode suspension
Figure 5.9 Bode plot (phase) for the position response of the heave mode suspension
a function of the proximity between the vehicle and the location of the guideway
supports. The worst case vibration conditions were found experimentally and are
marked on the Bode plots by the individual points measured at 42 Hz and 64 Hz.
These are attributed to the track and electromagnet hangers for which the undamped
natural frequencies were measured/estimated to be about 40 Hz and 70 Hz respectively.
The theoretical Bode plots assume that the track is rigidly coupled to the ground.
Vehicle suspension control 104
The gain stability margin84 for the experimental system is 17 dB compared with a
theoretical value of 23 dB. This discrepancy is attributed to the neglection of
discrete-time effects, and also to system model and implementation inaccuracies, in the
theoretical value. The phase stability margin for the experimental system is about 130°
versus a theoretical value which approaches 180°. This difference is mostly due to the
state integration filters and position error integral action which are neglected in the
theoretical calculation. The Bode plots clearly demonstrate a very good correspondence
between the theory and the experimental results from the suspension system, and also
that acceptable stability margins have been achieved.
The time and frequency domain responses for the other position controlled modes,
namely pitch and roll, are dominantly the same as those for the heave mode, and so
they are not presented here. However, the following two test phases present results
which include some time domain responses for all of the vehicle modes.
5.6.2 Decoupling of the vehicle modes and load sharing
Having established a satisfactory performance from the vehicle mode controllers when
tested independently, the cross-coupling between the different modes is now examined.
Figure 5.10, Figure 5.11, Figure 5.12 show the responses of each vehicle mode to a
1 mm reference step input to the heave, pitch and roll position controllers respectively.
The cross-coupling of the heave, pitch and roll modes to the torsion mode is clearly
negligible on all of the experimental test responses. The constant 0.1 mm torsion
position offset reflects the fact that the vehicle chassis and track have a gap
misalignment of +0.1 mm at one diagonal pair of electromagnets, and -0.1 mm at the
other pair.
Low frequency cross-coupling after the transient portion of the step response is apparent
between the heave, pitch and roll modes. This is caused by the electromagnet force
controllers independently adjusting to their new operating points, and it results in a
maximum coupling ratio of 8%. The only significant cross-coupling during the
transient part of the step responses links the pitch to the heave mode, and the roll to the
heave mode. This occurs because the centre of mass of the vehicle is above the
horizontal plane on which the accelerometers are located, and it gives rise to a coupling
ratio of 6%. The low amplitudes of the transient cross-coupling and the low frequency
cross-coupling clearly illustrate the success of the proposed control strategy in terms of
Vehicle suspension control 105
decoupling the vehicle mode motions.
Figure 5.10 Experimental mode responses to a 1 mm heave position reference step
Figure 5.11 Experimental mode responses to a 1 mm pitch position reference step
Figure 5.12 Experimental mode responses to a 1 mm roll position reference step
Vehicle suspension control 106
The lightly damped roll oscillations generated by the roll step are attributed to a small
cross-coupling to the lateral vehicle motion which is undamped. Finally, the very low
amplitude, high frequency oscillation apparent on all of the mode responses is due to
slight vibration of the electromagnet hangers as discussed in the previous test phase.
The load sharing capability of the vehicle suspension is tested by comparing the
operating conditions of the electromagnets at the point on the experimental guideway
where the torsional misalignment between the vehicle and guideway is at its maximum.
Table 5.4 gives a snapshot of the worst case operating conditions at which each
electromagnet deviates from its nominal value by approximately ±0.2 mm. The table
is augmented with another snapshot of the control system, but this time it is
reconfigured with the torsion position error, and error integral feedback gains set equal
to the respective gains for the other vehicle modes. Such a suspension configuration
is equivalent to using four independent electromagnet suspension controllers.
Table 5.4 Effects of worst case experimental vehicle to guideway misalignment
Parameter Electromagnets / Vehicle modes
Electromagnet air gaps (1,2,3,4) /mm 3.21 2.77 3.21 2.81
Vehicle mode gaps (h,p,r,t) /mm 3.00 0.01 -0.01 0.21
Electromagnet force demands /N 279 307 300 275
Electromagnet current demands /A 7.7 7.1 8.0 6.8
Electromagnet power dissipations /W 53 45 58 42
For controller with torsional position error plus integral feedback:
Electromagnet force demands /N 567 20 602 27
Electromagnet current demands /A 10.8 1.8 11.2 2.2
Electromagnet power dissipations /W 105 3 113 4
The air gaps of diagonal pairs of electromagnets (see Figure 5.2) are approximately
2.8 mm and 3.2 mm, thus the larger air gaps are approximately 14% bigger than the
smaller ones. The theoretical current ratio should therefore also be 14%, and the power
Vehicle suspension control 107
dissipation ratio 30%. In fact, due to a slightly uneven load distribution, the highest
electromagnet power dissipation is 38% above the lowest.
The experimental data for the reconfigured suspension with effectively independent
controllers shows that two of the electromagnets carry 96% of the vehicle load. Also,
as the vehicle moves along the guideway, the torsion error changes, and this causes
massive and rapid force fluctuations as the vehicle load is swapped between diagonal
electromagnet pairs. This clearly demonstrates why such a configuration is
unacceptable from both a steady-state and a dynamic viewpoint.
5.6.3 Ride quality of the vehicle suspension
The final test for the vehicle is for suspension ride quality whilst negotiating a step in
track height. The experimental guideway does not have any track steps so they are
simulated by injecting a step into the track position calculation. The experimental
response to a 1 mm step change in track heave position is shown in Figure 5.13 for the
unladen vehicle and in Figure 5.14 for the vehicle loaded with an 80 kg passenger.
Comparison of the experimental responses with the simulated responses for the single
electromagnet suspension (see Figure 4.11) shows that the experimental responses agree
well with the theory. The peak acceleration is 0.025 g which is comfortably below the
ISO target of 0.04 g. The low frequency overshoot (approximately 25%) is again due
to the a.c. coupling of the feedback signals derived from the accelerometers.
Figure 5.15, Figure 5.16, Figure 5.17 show the experimental responses for a 2 mm
simulated track position step for the heave, pitch and roll modes. These show that a
dominantly consistent response is obtained for each of the vehicle modes. The pitch
response overshoot is 16% which is approximately equal to that attributed to the a.c.
coupling of the velocity and position feedback signals (see Section 4.4.6). The heave
response has an additional, slower contribution due to error integral action as the
electromagnet force controllers adjust to the new operating point.
Finally, the roll response overshoot, at 29%, is 13% higher than that of the pitch mode.
Cross-coupling of the roll mode to the undamped sway mode was observed during the
track step response test, and the additional roll mode overshoot is attributed to this
cross-coupling. The cross-coupling occurs because the centre of mass of the vehicle
is located above the plane on which the accelerometers are located. Active control of
Vehicle suspension control 108
the lateral vehicle modes should reduce the roll mode overshoot to a similar level to
that experienced by the pitch mode.
Figure 5.13 Experimental heave response to a 1 mm simulated track step
(no passenger)
Figure 5.14 Experimental heave response to a 1 mm simulated track step
(80 kg passenger)
Vehicle suspension control 109
Figure 5.15 Experimental heave response to a 2 mm simulated track heave step
Figure 5.16 Experimental pitch response to a 2 mm simulated track pitch step
Figure 5.17 Experimental roll response to a 2 mm simulated track roll step
Vehicle suspension control 110
5.7 Conclusions
This chapter describes the development and validation of the multi-electromagnet
vehicle suspension control strategy outlined in Chapter 1. The experimental system
achieved a very good performance in terms of passenger ride quality, disturbance force
rejection, and electromagnet utilisation. The last feature arises from the ability of the
proposed control strategy to support the use of position error integral action which
facilitates accurate control of the nominal air gaps and hence maximises the available
air gap deviations. These benefits accrue from the development of a detailed
electromagnet model for the force control algorithm, and a sophisticated structure for
the vehicle suspension control system.
Finally, it is apparent that existing vehicle control strategies require detailed
consideration of the nonlinear force actuators throughout the design procedure in order
to get predictable results.85 By contrast, the linearity of the proposed electromagnet
force controller permits the assumption of linear force actuation. This fact, coupled
with the modularity of the proposed suspension control strategy, permits a linear vehicle
suspension design procedure, which consists of decoupling the electromagnet motions
and then configuring the mode suspension controllers.
Control system implementation 111
6
Control system implementation
6.1 Introduction
This chapter describes the selection and design of the various components required to
implement the suspension control system for the experimental research vehicle. The
experimental single electromagnet suspension uses a subset of the vehicle suspension
components.
The full set of vehicle control algorithms developed in Chapters 3, 4 and 5 is
computationally complex. Therefore, in order to provide a flexible experimental system
which is capable of being freely modified for current and future research work, a digital
signal processing approach was chosen. Additional benefits of this approach include
the elimination of drift and offsets in the signal processing, and the ability to implement
readily nonlinear functions. The disadvantages of digital processing are primarily those
due to the time and amplitude discretisation of the signals.
The implementation of the experimental control system is described in five sections.
First, the system requirements for the transducers, converters, and signal processors are
identified. Suitable industrial feedback sensors are then selected and an electromagnet
current controller is designed. Next, the signal processing, signal conversion and data
communication subsystems are designed. The design and configuration of the software
which implements the control algorithms is then described, and finally, conclusions are
drawn about the system implementation.
Control system implementation 112
6.2 System requirements
The experimental control system performs three basic tasks. It measures the air gap and
acceleration of each electromagnet, it calculates the electromagnet current demands
according to the control algorithms developed in Chapters 3, 4, and 5, and it adjusts and
maintains the electromagnet currents at the demanded levels.
In order to perform these tasks, the control system is functionally decomposed into five
subsystems, namely feedback sensors, analogue to digital converters, signal processors,
digital to analogue converters, and finally, electromagnet current controllers. The last
subsystem is implemented in the analogue domain to reduce the digital signal
processing load. The critical requirements for each subsystem are specified in terms
of their bandwidth, range, resolution, and accuracy. These specifications are calculated
by first considering the required signal bandwidths and sampling rates, and then the
required signal ranges, resolutions, and accuracies.
6.2.1 Bandwidths and sampling rates
The suspension system has been designed, and the control algorithms developed, using
continuous-time methods. However, since discrete-time, digital signal processing is to
be used, it is necessary to determine an acceptable maximum time interval between the
iterations of the controller. Shannon’s sampling theorem states that a sample rate of
twice the highest frequency component of a signal is theoretically sufficient to describe
that signal completely.86 However, for real-time, closed-loop control applications, the
feedback loop delay introduced by Shannon’s sampling rate would severely disturb the
location of the closed-loop poles, and adversely affect the closed-loop response. In
order to reduce the adverse effects of the feedback loop phase delay and hence obtain
a good response, the sample interval generally needs to be 1/5 to 1/10 of the time
constant of the dominant system pole.87 This results in a sampling rate which is 15
to 30 times Shannon’s theoretical sampling rate.
In addition to determining an acceptable delay for discrete-time sampling, an acceptable
phase delay for the air gap and acceleration sensors must also be determined. The
phase lag due to the electromagnet current controller is already incorporated in the
design of the suspension control algorithms, so it requires no further consideration.
Control system implementation 113
In order to produce a control system design which makes efficient use of the various
subsystem components, a range of acceptable phase delays is calculated, and this is then
apportioned between the feedback sensors and the discrete-time controller.
Table 4.4 lists the location of the dominant poles and zeros of the closed-loop position
control system for the single electromagnet suspension. For the purpose of this
analysis, the poles and zeros of the vehicle mode position controllers can be considered
to be the same as those of the single electromagnet suspension. The suspension
response is dominantly third-order, with a highest pole frequency of around 300 rad/s
for both the unloaded and fully loaded cases, and the calculated minimum phase delay
at this frequency is 3.0 rad (172°). By using the sample rate guideline identified earlier,
an additional phase delay of 10-20% for the position controller processing and sensor
delays is assumed to be acceptable. This permits an additional phase delay
of 0.3-0.6 rad (17-34°) at 300 rad/s, which can also be considered as a time delay
of 1.0-2.0 ms.
Figure 3.4 shows the location of the open-loop poles for the electromagnet at the worst
case operating point. The unstable pole has a worst case value of 167 rad/s, at which
the total open-loop phase delay is 3.4 rad (195°). Using the same assumption as before,
this permits additional force controller and sensor delays of 0.34-0.7 rad (20-39°)
at 167 rad/s, which can be considered as a time delay of 2.0-4.0 ms.
The phase and time delay requirements for the position controller are more exacting
than those for the force controller, so they are used for the baseline requirements. The
approximate equality between the delays occurs because the suspension control
algorithm parameters were designed to make maximum use of the available
electromagnet force actuation bandwidth. It also requires the same overall control
system bandwidth to be used for both the suspension position control algorithm and the
electromagnet force control algorithm.
The phase delays contributed by the air gap and acceleration sensors occur concurrently,
whilst the signal processing time delay follows sequentially. Since attaining high digital
sampling rates is more costly than high sensor bandwidths for this application, the
signal processing should be allocated a larger share of the available time delay than the
feedback sensors. Consequently, the signal processing is allocated roughly two thirds
of the allowable phase delay, with the remaining one third left for the feedback sensors.
Control system implementation 114
The phase delay generated by the first-order phase lag inherent in suitable low-cost
industrial air gap sensors is given by:
where θsensor is the sensor phase delay, ωsensor is the sensor bandwidth, and ω is the
6.1
θsensor
tan 1 ωω
sensor
≈ ωω
sensor
since ω <ωsensor
∴ ωsensor
≈ ωθ
sensor
signal frequency. The air gap sensor bandwidth required to generate a phase delay of
no more than 0.1-0.2 rad at a signal frequency of 300 rad/s is therefore
1500-3000 rad/s. This is equivalent to a sensor pole time constant of 0.3-0.7 ms.
Suitable low-cost industrial accelerometers use a sprung mass as the sensing element
and have a second-order low-pass response. Therefore, in order to generate the same
total phase delay as the air gap sensor, the accelerometer must have a natural resonant
frequency of 3000-6000 rad/s.
The time delay allowed through the signal processors is given by the allowable
processing phase delay (0.2-0.4 rad) divided by the highest closed-loop pole frequency
(300 rad/s). This gives an allowable signal processing time delay of 0.7-1.3 ms.
Assuming that the signal processing takes about 80% of the sampling interval, the
average time delay due to the discrete-time signal conversion and digital processing is
1.3 times the sample interval, and so a sampling interval of 0.5-1.0 ms is required.
The assumptions made about acceptable levels of phase delay were verified using a
detailed simulation of the full closed-loop suspension system (see Appendix C). The
additional phase delay of 10-20% had little impact on the closed-loop response, whilst
phase delays greater than 80% of the original value caused marginal instability
problems at the worst case operating point. Phase delays of less than 5% had a
negligible impact on the suspension response.
Table 6.1 summarises the bandwidth and sampling interval requirements that have been
calculated for the feedback sensors and the digital control algorithms. These figures are
guidelines between which trade-off adjustments can be made as necessary.
Control system implementation 115
6.2.2 Range, resolution and accuracy
Table 6.1 Bandwidth and sampling interval specifications feedback sensors
Component Parameter Value
Accelerometer natural frequency 480-960 Hz
Air gap sensor bandwidth 240-480 Hz
Electromagnet force controller sampling interval 0.5-1.0 ms
Suspension position controller sampling interval 0.5-1.0 ms
The normal operational envelope for the suspension system (see Chapter 3) covers an
electromagnet air gap range of 1-5 mm. However the air gap range required for the air
gap sensors is 0.5-5.5 mm in order to permit operation up to the mechanical limit stops.
Whilst the normal operational acceleration levels are very low, higher levels are
required for testing the suspension position controller. In addition, since the absolute
velocity and position are calculated from the acceleration measurement, it is imperative
that the acceleration signal is not allowed to saturate. Therefore, to accommodate step
response tests, a maximum acceleration measurement range of about ±1 g is required.
Operation of the electromagnets over their full operational envelope requires the current
to be controlled over the range 0-20 A.
The resolution and hysteresis of the transducers and the associated analogue-digital
converters affect the amplitude of the limit cycle oscillations of the control system.88
Since acceleration is the primary quality control factor for the suspension system, an
acceleration limit cycle amplitude of no more than about 20% of the target comfort
threshold of 0.4 m/s2 (see Chapter 4) is considered desirable. This calls for an
acceleration resolution of 80 mm/s2. Assuming an integration interval of 1 ms for the
velocity and position integrators, and sufficient numerical precision, this results in a
velocity resolution of 80 µm/s, and a position resolution of 80 nm.
The required force actuation resolution is calculated next because it is a useful reference
value for calculating the required air gap and current resolutions. The force resolution
is related to the acceleration resolution by:
Control system implementation 116
where ∆force is the force resolution and ∆acceleration is the acceleration resolution.
6.2∆ force mass ∆ acceleration
At the maximum load of 50 kg per electromagnet, an acceleration resolution of
0.08 m/s2 requires a force resolution of 4 N per electromagnet.
The required air gap signal resolution is determined by considering the force resolution
and the air gap feedback gain. The relationship between the air gap resolution and the
force resolution is given by:
The air gap feedback gain for the electromagnet force controller is determined by the
6.3∆airgap∆ force
air_gap_gain
open-loop electromagnet stiffness. This gives rise to a worst case, full load gain of
950 N/mm (see Table 3.2). Since the required force resolution calculated above is 4 N,
this requires an air gap signal resolution of 4 µm. This resolution calls for an
expensive, precision air gap sensor. However, the electromagnet stiffness at the
nominal operating air gap is only one third of the worst case stiffness. The nominal
operating point therefore requires an air gap resolution of 12 µm which is close to that
obtainable by standard industrial sensors which cost significantly less than precision
sensors. Therefore, in order to reduce costs, the lower resolution is preferred and a
slight degradation of the response at small air gaps is anticipated. The use of multiple
air gap sensors to improve the air gap measurement accuracy over the electromagnet’s
full length, would also increase the effective sensor resolution, as well as providing
sensor redundancy. However, in order to reduce the cost and complexity of the
experimental vehicle, only one air gap sensor per electromagnet is employed.
The air gap feedback signal also forms the very low frequency component of the
calculated track position which is filtered and fed into the suspension position
controller. However, since the position feedback gain per electromagnet is 250 N/mm
(see Chapter 4), this presents a less stringent resolution requirement than that identified
above. The position resolution of 80 nm, calculated previously from the acceleration
resolution and integration interval, is clearly satisfactory.
The required current control resolution is determined in a similar manner to the air gap
signal resolution. In this case, the worst case open-loop electromagnet force/current
ratio is 400 N/A (see Table 3.2), which requires a current resolution of 10 mA.
Control system implementation 117
The acceptable accuracy for the air gap and current transducers is determined by
considering a simplified equation for the electromagnet force and the corresponding
error equation. These are given by:
where f is the lift force, i is the coil current, g is the air gap and k is the constant of
6.4f ki 2
g2∴ f
errk
err2 i
err2g
err
proportionality. The error of the nonlinear force controller model is represented by kerr,
which is about ±7% (see Section 2.3.7). By assuming a current controller tolerance of
about ±2% and an air gap sensor tolerance of around ±3%, the resultant force controller
accuracy is expected to be about ±17%.
The accuracy required for the accelerometers is calculated by determining an acceptable
deviation in the characteristic response of the position controller. Under normal
operating conditions, a load change from full load to no load increases the natural
undamped frequency, ω, and damping ratio, ζ, (see Equation 4.7) by a factor of around
50%. Therefore, a further discrepancy of about ±10% to allow for transducer errors is
considered acceptable. Since both the velocity and position signals are calculated from
the acceleration signal, an error in acceleration measurement gain is modelled as an
equivalent error in the position, velocity and acceleration feedback gains. The
suspension parameters and their worst case error equations are approximately given by:
where kpos and kvel are the position and velocity feedback gains, ωerr and ζerr are the
6.5ω ∝ kpos
, ζ ∝k
vel
kpos
∴ ωerr
, ζerr
accerr
2
ferr
2
parameter errors, and accerr and ferr are the acceleration measurement and force actuation
errors respectively. Since the expected force actuation error is ±17% (see above), an
accelerometer accuracy of ±3% is required to achieve a total suspension parameter
deviation of ±10%.
Table 6.2 summarises the range, accuracy and resolution required to measure, convert,
and process the acceleration, air gap, and current signals. The accuracy is an aggregate
figure which should be achieved on an end-to-end basis. In practice, the very high
accuracy of analogue-digital converters and digital signal processing means that the
accuracy figures can be used as transducer requirement specifications.
Control system implementation 118
6.3 Transducers
Table 6.2 Range, accuracy and resolution specifications
Signal Range Accuracy Resolution
Acceleration ±10 m/s2 ± 3% 0.08 m/s2
Air gap 0.5-5.5 mm ± 3% 12 µm
Current 0-20 A ± 2% 10 mA
The suspension position control and the electromagnet force control algorithms require
the measurement of the electromagnet air gaps and accelerations. The control signal
generated by the control algorithms is a current demand, and this is actuated by a
high-gain, closed-loop current controller. In order to complement the benefits of the
electromagnetic suspension, the feedback sensors should not make contact with the
track.
The transducer requirements identified in the preceding section are now used to select
commercial devices for the feedback sensors, and the design of a special purpose
electromagnet current controller is described.
6.3.1 Accelerometer
A large range of industrial accelerometers is available which measure absolute
acceleration. Most accelerometers employ a mass constrained by some stiffness and
damping, and measure deflection of the mass to determine the acceleration. The
sensitivity of such a device is generally a function of the sprung mass divided by the
stiffness. Since the natural frequency is also a function of the sprung mass divided by
the stiffness, the natural frequency and hence the bandwidth of these devices is closely
related to the sensitivity.
The selection of a suitable accelerometer is thus largely determined by the
sensitivity/bandwidth constraint, and this appears to be somewhat more exacting for
electromagnetic suspension control applications than it is for many other industrial
requirements. Other important device parameters include cross-axis sensitivity, thermal
Control system implementation 119
stability, resolution, linearity and hysteresis. The last two factors are particularly
important because the acceleration signal is double integrated within the control system
to calculate the absolute velocity and position.
The basic performance requirements for the accelerometer are the measurement range
and resolution, the natural frequency and the general accuracy, and these are ±10 m/s2,
0.08 m/s2, 480-960 Hz and ±3% respectively (see Table 6.1 and Table 6.2). These
requirements are met most cost effectively by an industrial micro-machined silicon
sensor, which incorporates temperature compensated signal conditioning circuitry on the
silicon substrate. The device has a full scale range of ±20 m/s2, a resolution better than
0.04 m/s2, a natural frequency of 600 Hz, and a maximum error of ±3%. The full
specification is listed in Appendix D.
6.3.2 Air gap sensor
Non-contacting air gap measurement is used extensively in manufacturing processes
where the measurement target is moving. Capacitive, inductive and eddy-current
measurement techniques form the basis of most industrial non-contacting air gap
sensors. Capacitive and inductive devices use the measurement target as part of a
capacitor or inductor and determine the air gap from the measured capacitance or
inductance. Eddy-current devices measure the power loss from a coil due to the power
dissipation associated with eddy-currents circulating in the flux-coupled target.
The maximum performance requirements for the air gap sensor are determined by the
electromagnet force controller which calls for a measurement range of 0.5-5.5 mm, a
resolution of 12 µm, a bandwidth of around 240-480 Hz, and a basic accuracy of ±3%
(see Table 6.1 and Table 6.2). A standard industrial eddy-current loss air gap sensor
has been selected which has a useable air gap measurement range of 3-10 mm, a
resolution of 15-30 µm, a bandwidth of 190 Hz, and a basic accuracy of ±3% when
suitably calibrated. The full specification for the device is listed in Appendix D. The
specification is slightly lower than desired, but precision devices are typically 3-5 times
the price of the standard industrial devices. Since the overall system performance is
affected by both the air gap sensor and the accelerometer, the marginal performance of
the air gap sensor is partly offset by the high resolution and bandwidth of the
accelerometer.
Control system implementation 120
Eddy-current loss devices may not be appropriate for use on a production system
because the output is dependent on the target material resistivity which varies with
target temperature and material. Rapid movement of the sensor along the track may
also cause reduced accuracy due to eddy current losses induced by the motion of the
sensor along the track.
6.3.3 Electromagnet current controller
The electromagnet force control algorithm developed in Chapter 3 incorporates a
closed-loop electromagnet current controller, with a loop gain of 100. The
implementation of the current controller is now described, and it is based on the design
parameters derived in Chapter 3.
The primary dynamic characteristic of the current controller is determined by the
current feedback gain, which has already been calculated. The secondary dynamic
characteristic is the current slew rate, which is determined by the power supply voltage
and the characteristics of the electromagnet. Therefore, a power supply voltage for the
current controller must be calculated which gives an acceptable minimum current slew
rate.
Since the electromagnet is a component of a system designed to provide a comfortable
suspension ride, a force slew rate of 10% of the maximum load force per force
actuation time constant (Tforce ≈ 3.2 ms) is considered to be ample. The worst case
operating point in terms of the force slew rate occurs at the minimum air gap, since it
suffers the highest eddy current lag time constant. At this operating point (1 mm,
500 N), the electromagnet requires an excitation current, Imagnet, of 4 A. Generation of
a ±10% force change requires only a ±5% change in current due to the square law
relationship between the electromagnet current and the force. The required current
change, ∆Imagnet, is therefore ±0.2 A. The required supply voltage, which is the sum of
the voltage needed to slew the current plus the steady-state coil voltage, is given by:
where αphase_lead is time constant ratio of the eddy current phase-lead compensator, and
6.6Vsupply
Lcoils
αphase_lead
∆Imagnet
Tforce
Imagnet
Rcoils
volts
Rcoils is the resistance of the electromagnet coils. The maximum required supply voltage
is therefore:
Control system implementation 121
The supply voltage required at an air gap of 5 mm with a full load is, surprisingly,
6.7Vsupply
0.111 H3×0.2A
0.0032 s4A × 0.8Ω 24V
about the same as that required at 1 mm. At the 5 mm operating point, the reduced
voltage demand due to the smaller eddy current lag time constant, is offset by the
increased voltage demand due to the higher leakage flux and the higher steady-state
current.
The supply voltage required for a given implementation is slightly higher due to the
additional resistance of the power controlling devices, which are connected in series
with the electromagnet. A suitable technique for amplifying the low power control
signal to the high currents required for the electromagnet is now considered.
If a linear mode (class A) power amplifier design were employed, the maximum
theoretical efficiency would be about 33% since the electromagnet would have an
average voltage of 8 V across it, with the remainder dropped across the series current
regulator. This would produce an average amplifier dissipation of 16 V x 10 A =
160 W, and require a power supply rating of 24 V x 20 A = 480 W. Alternatively, the
use of a switch mode (class D) power amplifier typically enables an efficiency of about
85% to be achieved,89 thus reducing the average amplifier dissipation to about 12 W,
and the power supply rating to about 380 W.
A switch mode current controller is therefore desirable and since no suitable
commercial units were available, an electromagnet current controller was designed.90
Figure 6.1 shows the schematic diagram of the controller which incorporates electrical
isolation between the low power signal circuitry, and the high power circuitry.
The current is sensed using an electrically isolated precision current sensor with an
accuracy better than 1%, and a response time of about 1 µs. The measured current is
compared with the reference current demand and amplified to form the current error
signal. This analogue signal is fed into an industry standard pulse width modulator
(PWM) integrated circuit, which drives an electrically isolated H-bridge power
switch.91 Since only a uni-directional current drive is required, two active switches
are used in the bridge configuration,92 with passive devices used in the two remaining
legs.
Control system implementation 122
The time constant of the closed-loop current response is around 1 ms, so a sampling (ie.
Figure 6.1 Schematic design of the electromagnet current controller.
PWM switching) frequency of at least 10 kHz is desirable. In order to introduce a
negligible phase delay due to sampling, and also to make the switching inaudible, a
PWM frequency of 40 kHz is used. This switching frequency favours the use of
MOSFET devices rather than thyristors or bipolar transistors for the bridge switches.93
The selected MOSFET devices have a maximum, temperature derated, drain-source
resistance of 50 mΩ, and are switched in 100-200 ns by a MOSFET driver integrated
circuit. This results in a total power dissipation of about 16 W at the nominal current
rating of 10 A. The sensor accuracy and closed-loop gain combine to give the current
controller an accuracy better than ±2%.
Each current controller module incorporates an on-board PWM oscillator which permits
independent operation of the module. However, for the experimental vehicle which has
four electromagnets, the slight difference in frequency between each of the PWM
oscillators would generate low frequency beat noise. This would be undesirable since
the noise would be within the bandwidth of the feedback control signals. Therefore,
for the vehicle configuration, a single oscillator source is used to ensure synchronous
operation of the four current controller modules, and hence prevent any low beat
frequency noise.
The current controller and electromagnet are also protected by two safety features.
First, the current controller receives an output disable control which is activated if
necessary by a system watchdog timer. In addition, a current limiter set to a level
of 21 A is incorporated in the design.
Control system implementation 123
The high power supply for the current controllers is provided by commercial,
mains-powered switch mode supplies rated at 27 V and 15 A. The circuit diagrams for
the electromagnet current controller are listed in Appendix D.
6.4 Signal processing, conversion and communication
To ensure sufficient flexibility to investigate control algorithms of increased complexity
in the future, a scalable processing resource is desirable. A further requirement for
flexibility and future research is for a distributed processing system. In addition to
these system features, good language support and development system support for
scalable and distributed processor systems are required.
Researchers in the fields of real-time control and simulation94,95,96,97 have found
that parallel processing architectures can offer significant performance and modularity
advantages over single processor architectures.
Three microprocessor development systems were available for use with this work. The
target microprocessors and co-processors were the Intel 80386/80387, the Motorola
68030/68881, and the Inmos T800 transputer. In a study98 of these and other
processors, the transputer came out most favourably in terms of processor performance,
connectivity, parallel language support, and multi-processor development support.
Academic and industrial researchers99,100,101,102 have also found the transputer to
be particularly suitable for implementing parallel processing architectures. Initial work
using one transputer to perform the signal processing for an experimental single
electromagnet suspension103 confirmed that a T800 transputer based system was
capable of meeting the processing requirements at sampling rates around 1 kHz.
6.4.1 Transputers
The Inmos transputer family104 is a range of microprocessors designed for use as
parallel processing components. The family includes 16 bit integer processors (T2xx),
32 bit integer processors (T4xx), and 32 bit processors with a floating point unit (T8xx).
The T800 transputer (Figure 6.2) contains the following system components:
Control system implementation 124
• 32 bit central processing unit (12.5 MIPS at 25 MHz);
• 64 bit floating point processor (1.5 MFLOPS at 25 MHz);
• 4 Kbytes internal memory;
• 2 timers;
• 4 bi-directional 5-20 Mbits/sec inter-processor serial communication links.
The inter-processor communication links facilitate the use of distributed processor
Figure 6.2 T800 transputer architecture.
systems with distributed memory. Unlike conventional bus systems with shared
memory, the inter-process communication bandwidth rises linearly with the number of
processors. Memory and input/output facilities can be readily extended using external
devices.105 Data transfers using the inter-processor links are performed concurrently
with process execution using direct memory access (DMA). This results in a low
performance degradation even when all links are running at full capacity. The
communication links can be connected directly between processors on a single card or
in a chassis unit. For longer distances, twisted pair, coaxial cable or fibre-optic links
can be simply and effectively used. Communication links between transputers and links
to external devices can be configured using a 32-way cross-point switch if necessary.
In order to provide flexibility when configuring software processes onto transputers, a
built-in hardware scheduler is provided which multi-tasks parallel processes on a single
processor.106 This enables algorithms to be coded in parallel without constraining
them to a particular parallel processor configuration. The scheduler performs a task
switch in about 1-4 microseconds which is extremely fast, thus permitting multiple
parallel threads to be efficiently used even with sampling intervals of approximately
1 ms.
Control system implementation 125
Transputer development systems are available for a range of computer platforms. For
this work, an IBM compatible PC/AT is used with a plug-in card which accommodates
the development host processor. The development system software and application
software use the PC as a terminal and file server. Development system support for
many programming languages is available, most of which have parallel programming
extensions. For this work, occam is used because it provides a flexible and robust
environment, with particular regard to the data security issues of parallel processing.
6.4.2 Control system hardware structure
A modular hardware system has been developed to run the vehicle controller and the
single electromagnet rig controller. The system functions are partitioned and
implemented using six cards which can be freely configured. The functional
specification for each card is listed in Table 6.3. The TRAM motherboard takes
standard Inmos TRAM modules which consist of a transputer processor plus random
access memory (RAM).
The cards for the vehicle control system are partitioned between two chassis. One
contains the feedback sensor signal conditioning, the analogue to digital converter and
three TRAM motherboards. The other contains the digital to analogue converter and
the four electromagnet current controller modules. For the single electromagnet control
system, all the cards and modules are located in a single chassis.
In order to prevent earth-loop and other noise problems, the chassis and the
development computer system are all electrically isolated from each other. Digital fibre
optical connections operating at 10 Mbit/s are used for all inter-unit transputer link
communications. Rack input and output interface cards provide the optical input and
output facilities between equipment units. Since the optical link connections are short,
optical signal power loss is not a problem, and so plastic optical fibre cable is used
which is inexpensive.
The issues affecting the design of the analogue signal conditioning and conversion
cards, and the TRAM motherboard and optical interface cards are discussed next. The
chassis configurations and circuit diagrams for all the cards are detailed in
Appendix D.
Control system implementation 126
6.4.3 Analogue signal conditioning and conversion
Table 6.3 Control system hardware function specifications.
CARD FUNCTION
SAP Sensor Analogue Processor:
- Current to voltage conversion for 4 air gap sensors.
- Filtering & amplification for 4 accelerometers.
ADC Analogue to Digital Converter:
- 8 analogue input channels.
- Anti-alias filters.
- Multiplexer selection.
- 12 bit, bipolar ADC, 10 µs conversion time.
- 16 bit, 10 MIPS integer processor (T212).
- Front panel switch input.
TMB Tram MotherBoard:
- Motherboard with 4 size1 TRAM sites.
DAC Digital to Analogue Converter:
- 4 buffered analogue output channels.
- 12 bit unipolar DACs.
- 16 bit, 10 MIPS integer processor (T212).
- Watchdog timer.
- Front panel switch input.
IFIN InterFace INput:
- Optical chassis input interface.
- 1 bidirectional transputer data link.
- Transputer reset & analyse control inputs.
- Transputer error control output.
IFOUT InterFace OUTput:
- Optical chassis output interface.
- 1 bidirectional transputer data link.
- Transputer error control input.
- Transputer reset & analyse control outputs.
The schematic design of the analogue to digital converter card is illustrated in
Figure 6.3. Eight analogue input channels are provided, each of which is buffered and
filtered to remove high frequency noise which would otherwise be aliased107 down to
the frequency bandwidth used by the control system signals. The anti-alias filters have
a critically damped, second-order response, with a corner frequency of 1.6 kHz. The
Control system implementation 127
electromagnet current controllers switch a total of up to 80 A at a frequency of 40 kHz.
They can therefore be expected to be a source of noise which will be inductively and
capacitively coupled to the feedback signals. Such noise (at >40 kHz) is reduced by
56 dB by the anti-alias filters. The anti-aliased signals are selected by an analogue
multiplexer which routes the selected signal to the 12 bit analogue to digital converter,
where it is sampled and converted to the digital domain. A parallel input port is also
provided which is used solely to read the state of a switch mounted on the front panel
of the card.
The sensor analogue processor card applies the appropriate signal conditioning and
Figure 6.3 Schematic diagram of the analogue to digital converter.
amplification to the sensor signals to give a full scale deflection of 0-5 V for the air gap
sensor and ±5 V for the accelerometer. In addition, the accelerometer signals are
filtered by a first-order high pass filter with a corner frequency of 0.1 Hz to remove the
1 g measurement offset due to gravity.
The 12 bit analogue converter produces a channel quantisation amplitude of 2.5 mV
which in turn gives an air gap resolution of 5.0 µm and an acceleration resolution of
5 mm/s2.
The digital to analogue converter card employs four 12 bit converters to drive four
electromagnet current controllers. The schematic design of this card is illustrated
Figure 6.4. In addition, a parallel input port is used to monitor a panel mounted switch.
This card also carries the system watchdog timer which can disable the electromagnet
Control system implementation 128
current controllers. If the watchdog is not triggered at least once every 10 ms, all
connected current controllers will shut down immediately. This feature helps to provide
fail-safe operation and a controlled system startup. The card also includes a 40 kHz
clock generator which is used to synchronise the PWM switching of the electromagnet
current controllers. An optical interface is provided on one of the transputer’s links to
provide electrical isolation for the vehicle chassis containing the DAC card and the
electromagnet current controllers.
The 12 bit digital converters are configured for a unipolar, full scale deflection of 10 V
Figure 6.4 Schematic diagram of the digital to analogue converter.
which produces an output quantisation level of 2.5 mV, and gives a current resolution
of 5 mA. The conversion resolutions for the air gap, acceleration and current signals
are thus at least twice the values specified in Table 6.2 for the end-to-end system signal
resolutions.
6.4.4 Digital signal processing and communication
The TRAM motherboard carries commercial plug-in transputer modules which perform
the signal processing functions. The ADC and DAC cards are provided with 16 bit
integer processors in order to facilitate intelligent analogue interfaces, rather than to
perform signal processing. Figure 6.5 illustrates the schematic design of the TRAM
motherboard. It simply connects four TRAM sites in a pipelined configuration, and
Control system implementation 129
provides buffered interfaces between the processor control signals and the chassis
backplane.
To provide optical fibre connections between the two chassis and the PC, a pair of
Figure 6.5 Schematic diagram of the TRAM motherboard.
optical interface cards are used. One provides a chassis input interface, and the other
is configured as a chassis output interface. Each card provides optical interfaces for one
bi-directional transputer data link, and the processor reset, error and analyse control
signals.
6.5 Software design
The design of the control system software is described in five main parts. The reasons
for using occam are explained first, and some salient features of the language are
outlined. The high-level structure of the control system software is then described.
Next, a method for converting the continuous-time algorithms to the discrete time
domain is selected. The level of numerical precision required for the signal processing
is then calculated. Finally, the configuration of the constituent processes of the control
system software onto the hardware is described.
6.5.1 Occam
The occam programming language108 is a high level language, designed to express the
sequential and concurrent components of algorithms, and their configuration on a
network of processors. Since the transputer has a built-in multi-tasking scheduler,
parallel algorithms can be run on a single processor as well as being distributed over
a network of processors. The strength of this facility is that a parallel algorithm can
Control system implementation 130
be directly coded into a parallel program, with little regard to the processor
configuration. The program can then be configured to execute on one or more
processors, parallelism and communication permitting.
Occam has been used for this work for three main reasons. Firstly, the use of
concurrent processing hardware introduces high level concurrency within the software
whilst control systems typically contain low level concurrency as well.109 Occam is
a rare example of a programming language specifically designed to implement
concurrency at all structural levels in a natural and efficient manner. Secondly, when
using concurrent algorithms, occam can provide a degree of security unknown in
conventional sequential programming languages. Finally, the transputer reflects the
occam structural model and may be considered an occam machine. Programming in
occam is thus almost as efficient as using assembly language on conventional
processors.
The features which differentiate occam from conventional languages are outlined in
Appendix E prior to the listings of the control system software programs.
6.5.2 Control system software structure
The structure of the control system software has been designed using a combination of
functional decomposition as advocated by Wirth,110 and functional partitioning to
minimise the data flow between processes as proposed by DeMarco.111 Figure 6.6
illustrates the structure of the system in terms of tasks which are connected via channels
providing synchronised communication of data signals, control messages and exception
reports. The real-time data flows through the input signal interface, the state
calculators, the control algorithms and the output signal interface. These components
are supported by the sample scheduler, the exception handler, the data monitor and the
user interface.
Table 6.4 gives a functional overview of the software tasks for the vehicle control
system. The specification for the single electromagnet suspension is the same except
for the use of only one electromagnet rather than four.
The control algorithms block depicted in Figure 6.6 is partitioned into two sub-blocks,
which perform the suspension mode force calculation and the electromagnet current
calculation. Figure 6.7 illustrates this arrangement for the single electromagnet
Control system implementation 131
suspension system. The air gap and acceleration signals are supplied by the input signal
Figure 6.6 Control system task structure, and data and control flows.
interface and the current demand is fed to the output signal interface. The parallelism
within the state calculation and the control blocks is fine grained, and is therefore not
well suited to parallel processing using transputers.
The vehicle suspension system illustrated in Figure 6.8 uses four independent
Figure 6.7 Single electromagnet controller data flow
suspension controllers, one for each of the vehicle motions, heave, pitch, roll and
torsion. Each suspension controller, plus its associated transformation and state
calculation, involves a significant amount of computation. This results in a
Control system implementation 132
computational granularity which is sufficiently large to utilise parallel processing
Table 6.4 Vehicle control software functional overview
Input tasks:
• Read in acceleration and air gap sensor signals using ADC.
• Scale signals, check and report errors.
State calculation:
• Calculate velocities and positions from acceleration signals (Chapter 4).
• Calculate track position from magnet positions and air gaps (Chapter 4).
Control tasks:
• Apply the vehicle suspension control algorithm (Chapter 5) and magnet force
control algorithm (Chapter 3) to the input and calculated data sets to produce
a set of current demands for output. Check signals and report errors.
• Send user selected data to PC via asynchronous monitor process.
Output tasks:
• Scale current demand signals and output to the DACs.
• Check signals and report errors.
• Reset watchdog timer if the system is operational.
Miscellaneous:
• Permit user modification of controller parameters (including sample rate).
• Provide user selectable test reference signals (d.c., sine wave, square wave).
• Generate smooth startup and shutdown under normal and error conditions.
effectively using one transputer per suspension controller. After the vehicle motion
force demands have been calculated, they are transformed to electromagnet force
demands which are fed to the electromagnet controllers. The four independent
electromagnet force controllers involve a level of computation similar to that of the
vehicle suspension controllers, and so parallel processing using one transputer per
controller is again efficient.
For the sake of convenience in terms of monitoring and comparing vehicle mode
signals, the input and output decoupling transformations also normalise the signal
amplitudes for the vehicle pitch, roll and torsion modes. This normalisation is
described in Chapter 5, and it converts the angular signals and torques to the equivalent
linear signals and forces at the electromagnet centres. Normalising the signals also
permits a single data monitoring configuration to handle signals for suspension control
algorithms using different decoupling transformations.
The sequencing of the operations of the vehicle control system can be summarised in
terms of the following eight major phases:
Control system implementation 133
1. Delivery of all magnet air gap and acceleration signals to each vehicle
section.
2. Transformation of magnet signals to vehicle mode motions.
3. Calculation of unmeasured vehicle motion states.
4. Computation of each vehicle mode force demand.
5. Distribution of all vehicle mode forces to each magnet control section.
6. Transformation of vehicle mode force demands to magnet force demands.
7. Computation of each magnet current demand.
8. Delivery of each magnet current demand to the current demand signal pool.
These phases occur irrespective of the number and configuration of processors used to
Figure 6.8 Vehicle suspension controller data flow
run the control system. For a balanced processor load distribution, either 1, 2 or 4
processors can be used, each computing either 4, 2 or 1 vehicle motion sections
followed by the same number of electromagnet control sections. The number of
processors required depends on the individual processor power, the algorithm
complexity and the required speed of execution. The multi-processing overheads for
Control system implementation 134
the transputer, including the inter-processor communication overheads are small which
gives a high multi-processor utilisation.
6.5.3 Discrete time domain integration
The control algorithms developed in Chapters 3, 4 and 5 have been designed in the
continuous time domain. Therefore, they must be converted to discrete time domain
representations before they can be implemented in software. The selection of a suitable
technique for implementing discrete-time integrators and filters is considered next.
Filters are used by the state calculation algorithm, the suspension mode control
algorithm and the electromagnet force control algorithm. The first two algorithms use
low pass and high pass filters, with time constants ranging from 1.6 s to 40 ms, whilst
the electromagnet force control algorithm uses a phase lead compensator with a pole
time constant of 1.5 ms. The time constant of the phase lead compensator pole is thus
close to the required control system sampling interval of 0.5-1.0 ms. The discrete-time
representation is therefore a fairly crude approximation to the continuous-time response.
However, the effect of this misrepresentation is to increase the effective phase lead
slightly which is not detrimental to the location of the closed-loop system poles. The
simulation model listed in Appendix C was used to verify this assumption. The high
pass filter and phase lead compensator are formulated in terms of a first-order low pass
filter as described in Table 6.5.
Various integration algorithms are available for approximating the continuous time
Table 6.5 Filter formulations
LowPass(s,T )1
1 sT
HighPass(s,T )sT
1 sT1 LowPass(s,T )
PhaseLead(s,T,N )1 NsT
1 sTN (N 1)LowPass(s,T )
domain in the discrete time domain and each method has its own merits.112 A
primary requirement for this development is that the approximation algorithm must be
cascadable so that each independent functional block within the control system can be
designed and implemented independently. The algorithm must also produce an accurate
Control system implementation 135
d.c. gain, and map stable continuous-time poles to stable discrete-time poles. These
requirements are effectively met by the Tustin algorithm (trapezoidal integration) and
by the first difference algorithm (Euler integration). Since all the filters except for the
phase lead compensator have time constants which are orders of magnitude greater than
the sample interval, the simpler first difference algorithm is sufficiently accurate. The
lower quality approximation that results for the phase lead compensator is acceptable.
Prewarping113 of the filter time constants is unnecessary for the normal control system
parameter settings since the time constants are much larger than the sampling interval,
and so their pole placement accuracy is normally very good. However, frequency
prewarping is required for the phase lead compensator time constant since it is so small.
It is also employed for the other time constants to allow for experimentation with
extreme settings for the filter frequencies and the sampling interval. Table 6.6 lists the
discrete time algorithms which are used for integration and low-pass filtering, and the
prewarping correction factor.
Table 6.6 Discrete-time integrator and filter implementation
Integrator: yk
yk 1
xk
Tsample
Low pass filter: yk
yk 1
1T
sample
Tcutoff
xk
Tsample
Tcutoff
Prewarp correction:
Tsample
Tcutoff corrected
≈ 1 exp
Tsample
Tcutoff
6.5.4 Numerical accuracy
Having determined the discrete time integration algorithm, the level of numerical
accuracy required to implement the control algorithms is now calculated. A
floating-point representation is assumed for all operations in order to have precision
independent of scaling. This eliminates the implementational overhead associated with
the use of fixed-point representations where signals must be re-scaled as necessary to
maintain sufficient precision.
Control system implementation 136
The resolution required for the transducer signals calls for the use of 12 bit
analogue-to-digital and digital-to-analogue converter hardware. The calculation of each
current demand signal involves a total of about 80 mathematical operations, and in
general, a 32 bit floating-point representation (which has a 23 bit mantissa) provides
sufficient numerical accuracy for 12 bit data. However, loss of accuracy can occur
where two values which differ by many orders of magnitude are added or subtracted.
The worst case of this behaviour occurs in the state integration filters for velocity and
position, where the required sample interval of 0.5-1.0 ms can result in very small
values being summed onto the very much larger value of the state integrator.114
Table 6.7 shows the calculation of the relative size of the input and integrator values
for the velocity and position state calculators. The sampling interval, Tsample, is assumed
to be 0.5 ms, and the integration filter corner frequency of 0.1 Hz gives a filter time
constant, Tfilter, of 1.6 s. The decay product term is neglected for this analysis since it
is approximately unity. Accmin is set to half the required quantisation amplitude for the
acceleration measurement and velmin represents the corresponding velocity quantisation.
Velmax and posmax represent the maximum practical values that the velocity and position
signals can have.
The position integration filter provides the most exacting requirement since the
Table 6.7 Signal magnitude calculation for the state integration filter
Euler integration for velocity integration filter (worst case scenario) gives:
vel = velmax + accmin Tsample / Tfilter
= 10-1 m/s + 5x10-2 m/s2 5x10-4 s / 1.6 s
= 10-1 m/s + 1.6x10-5 m/s
Relative size: 10-1/1.6x10-5 = 6.4x103
Euler integration for position integration filter (worst case scenario) gives:
pos = posmax + velmin Tsample / Tfilter
= 5x10-3 m + 1.6x10-5 m/s 5x10-4 s / 1.6 s
= 5x10-3 m + 5x10-9 m
Relative size: 5x10-3/5x10-9 = 106
maximum accumulated position signal can be 106 times larger than the minimum added
value. A 32 bit floating-point representation with a 23 bit mantissa (plus the sign bit)
Control system implementation 137
has a precision of just under 7 decimal digits. Under the worst case scenario of
maximum position signal and minimum velocity, the input value would be summed into
the integrator with a precision of less than one decimal digit. This would cause severe
pole placement inaccuracy, and a significant loss of the summed signal accuracy. The
state integration filters therefore use a 64 bit floating-point representation which has a
52 bit mantissa giving a precision of about 15.5 decimal digits.
A similar argument applies to the guideway following filter which consists of two
Table 6.8 Signal magnitude calculations for the guideway following filter
Euler integration for guideway filter (worst case scenario) gives:
trk = trkmax + gapmin Tsample / Tfilter
= 5x10-3 m + 5x10-6 m 5x10-4 s / 4x10-2 s
= 5x10-3 m + 6.25x10-8 m
Relative size: 5x10-3/6.25x10-8 = 8x104
And similarly for the second (cascaded) guideway filter gives:
Relative size: 5x10-3/7.8x10-10 = 6.4x106
cascaded low-pass filters. Table 6.8 shows the magnitude calculations for the guideway
filter where gapmin is set to half the required air gap signal quantisation and trkmax is the
maximum air gap measurement. Once again, the large relative size between the two
filter terms, almost 7 decimal digits, requires the use of 64 bit floating-point numbers
for the guideway following filters.
The fast time constant used by the phase lead compensator used in the magnet force
control algorithm results in 32 bit floating-point accuracy being sufficient. The
remaining computation for the mode and force transformations and the suspension and
electromagnet control algorithms can all be satisfactorily performed using 32 bit
floating-point arithmetic.
6.5.5 Process configuration
Having determined the requirements for the control system software, the final stage is
to partition the software tasks onto a suitable transputer configuration. For the single
Control system implementation 138
electromagnet suspension controller, this is simple since a single 20 Mhz T800
processor is sufficient to provide a minimum sampling interval of 0.8 ms. Experimental
responses showed this sampling rate to be more than adequate. Responses were
therefore obtained with larger sampling intervals, and they showed that the response
started to become unacceptable for values larger than about 2 ms.
For the vehicle control system software, the execution time for all of the tasks identified
in Figure 6.8 was measured for a single 25 MHz T800 processor. This achieved a
minimum sampling interval of about 2.3 ms, compared with the target range of
0.5-1.0 ms, and experimental responses showed it to be unsatisfactory. It was estimated
that the use of two processors would reduce the sample interval to about 1.2 ms, whilst
employing four processors would reduce it down to about 0.65 ms.
For economic reasons, a two processor configuration was investigated. This
configuration executed with a measured sampling interval of 1.25 ms, and good
experimental responses were obtained. Figure 6.9 shows how the software tasks are
partitioned between the two processors. This configuration and sampling interval is
used for all of the experimental vehicle responses presented in this dissertation. The
lower processing load on the signal processor without the data monitor and its
associated real-time buffer permits the use of a 20 Mhz processor.
A fully controlled vehicle with four suspension electromagnets and four guidance
electromagnets could therefore be controlled by four transputers, since the additional
overheads are minimal. The code listings for all of the software described in this
chapter are listed in Appendix E.
6.6 Conclusions
This chapter has shown that the electromagnetic vehicle suspension strategy developed
in the earlier chapters can be implemented using readily available parallel processing
hardware. The use of transputers for the signal processing and intelligent converter
interfaces enabled a highly modular and inexpensive real-time processing platform to
be constructed. This was complemented by using the occam programming language
which enabled a fairly complex software implementation to be developed very rapidly.
Control system implementation 139
Figure 6.9 Vehicle control system process configuration
Conclusions 140
7
Conclusions
The aim of the research described in this dissertation has been to improve the
performance of electromagnetic secondary suspension for vehicles through the use of
improved control techniques.
The difficulties associated with the control of an electromagnetic vehicle suspension
accrue from the nonlinear and unstable nature of the electromagnetic force
characteristic. In addition, for systems which use electromagnets to provide both the
primary and the secondary suspension functions, the direct coupling of the
electromagnets to the vehicle chassis causes further complications by producing a
multivariable control problem.
The proposed structured design approach requires the electromagnets to be controlled
in such a way that they can be regarded as independent linear force actuators. The
electromagnet force characteristic has therefore been analysed, and a detailed nonlinear
model has been developed. Practical force control schemes employing only linear
feedback techniques have been shown to be unsuitable for providing independent linear
force actuation in an environment where a number of electromagnets are rigidly
coupled. A new force control algorithm has therefore been developed which employs
the detailed electromagnet model, in conjunction with electromagnet air gap and current
feedback, to provide force actuation which is dominantly linear and independent.
Having obtained a suitably linear force actuation, the vehicle suspension is controlled
by using linear transformations to translate between the electromagnet coordinate system
and a vehicle coordinate system based on the heave, pitch, roll and torsion of the
vehicle chassis. Each vehicle mode motion is then controlled by an independent
suspension controller.
Conclusions 141
A sophisticated control algorithm has been developed to control the independent
suspension modes. This consists of an absolute position controller, designed to achieve
the required disturbance force rejection, which receives its position reference from a
filtered version of the track position signal. All feedback signals are derived from
measurements of the absolute acceleration and the air gap of each electromagnet. The
structure of the new suspension control algorithm provides greater design flexibility
than existing algorithms. This is because it permits the algorithm which defines the
guideway following characteristic, to be designed largely independently of other
suspension design considerations.
In order to obtain experimental responses in addition to simulated responses for the
proposed vehicle suspension control scheme, an experimental suspension control system
using digital signal processing has been developed. The signal processing hardware is
based around the transputer family of microprocessors, and the control algorithms have
been implemented using the occam parallel programming language. The simulated and
experimental results presented in this dissertation have indicated the success of the
proposed control method, to an extent where confidence is given in its potential for
development to a full scale system.
The proposed electromagnet force control scheme has been shown to possess significant
advantages compared with existing stabilisation techniques using flux derivative
feedback due to its dominantly linear force actuation. However, it suffers a slight
disadvantage due to an increased reliance on an accurate air gap measurement at small
air gaps (see Chapter 3). If desired, this drawback could be overcome by developing
a hybrid control approach combining flux derivative feedback, for stability, with the
proposed scheme, for force linearity.
In addition, a number of areas which could benefit from further research have been
identified. Firstly, an improvement in the modelling accuracy for the electromagnet
core eddy-current time constant is desirable (see Chapter 2). Secondly, lateral force
control through air gap modulation (see Chapter 4) may permit lateral damping for
research vehicles using only suspension electromagnets. Thirdly, analysis of vehicle -
guideway interaction and the development of suitable damping algorithms may be useful
to permit the use of flexible guideways (see Chapter 1). Finally, and most importantly,
guideway following algorithms need to be researched in order to minimise air gap
deviations at the entrance and exit of gradients (see Chapters 1 and 4).
Conclusions 142
The results of this work suggest that with some additional research and development
in the areas outlined above, electromagnetic suspension may in future provide an
effective method for providing high ride quality, and low cost, for vehicle suspensions
in urban transit applications. For wheel-on-rail transport applications, the wheel may
then become a historical curiosity.
References 143
8
References
1 Geary, P.J.: ‘Magnetic and electric suspensions’, SIRA Research Report No.
R314, Brit. Sci. Instr. Res. Assc., 1964.
2 Jayawant, B.V.: ‘Electromagnetic levitation and suspension techniques’, Edward
Arnold, London, 1981.
3 Sinha, P. K.: ‘Electromagnetic suspension: Dynamics & control’, Peter
Peregrinus, London, 1987.
4 Weh, H.: ‘Magnetic levitation technology and its development potential’, Int.
Conf. Maglev ’89, July, 1989, pp 1-9.
5 Earnshaw, S.: ‘On the nature of the molecular forces which regulate the
constitution of the luminiferous ether’, Trans. Camb. Phil. Soc., 1842, 7, pp 97-
112.
6 Graeminger, B.: ‘Improvements in or relating to electromagnetic suspension
devices’, British Patents: 24499 and 24541, 1912.
7 Kemper, H.: ‘Schwebende Aufhängung durch electromagnetische Kräfte: eine
Möglichkeit für eine grundsätzlich neue Fortbewegungsart’, Electrotech. Zeits.,
1938, 59, pp 391-5.
8 Hrovat, D.: ‘Influence of unsprung weight on vehicle ride quality’, J. Sound and
Vibration, 124, 1988, pp 497-516.
9 Messerschmitt-Bölkow-Blohm: ‘Method for controlling the energisation of the
supporting and guiding electromagnets of a magnetically suspended vehicle’,
British Patent: 1401313, 1972.
10 Gottzein, E., Crämer, W.: ‘Critical evaluation of multivariable control techniques
based on Maglev vehicle design’, Proc. 4th MVTS Symposium, IFAC ’77, 1977.
11 Rule, R.G. and Gilliland, R.G.: ‘Combined Magnetic Levitation and Propulsion -
The Mag-Transit concept’, Boeing Aerospace, Automated Guideway Transit,
Technical paper 6, 1979.
References 144
12 Jayawant, B.V., Sinha, P.K., Wheeler, A.R., Whorlow, R.J. and Willsher, J.:
‘Development of a 1-ton magnetically suspended vehicle using controlled d.c.
electromagnets’, Proc. IEE, Pt. A, 123, 1976, pp 941-8.
13 Goodall, R.M.: ‘Suspension and guidance for a DC attraction Maglev vehicle’,
2nd Conf. on Advances in Magnetic Materials and their Applications, London,
1976, pp 100-3.
14 Nenadovic, V. and Riches, E.E.: ‘Maglev at Birmingham Airport: from system
concept to successful operation’, GEC Review, 1, 1985, pp 3-17.
15 Nakamura, S.: ‘Development of high speed surface transport system (HSST)’,
IEEE Trans. on Mags., MAG-15, 1979, pp 1428-33.
16 Atherton, D.L. and Eastham, A.R.: ‘Limitations of levitation by iron-cored
electromagnets’, IEEE Trans. on Mags., MAG-10, 1974, pp 410-2.
17 Gottzein, E., Meisinger, R. and Miller, L.: ‘The "Magnetic Wheel" in the
suspension of high-speed ground transportation vehicles’, IEEE Trans. on Veh.
Tech., VT-29, 1980, pp 17-22.
18 Bohn, G. and Alscher, H.: ‘The magnetic train Transrapid 06’, Int. Conf. on
Maglev and Linear Drives, 14-16 May, 1986, pp 47-52.
19 Miller, L.: ‘Transrapid 06 II - Performance and characteristics’, Int. Conf. on
Maglev and Linear Drives, 19-21 May, 1987, pp 155-62.
20 Friedrich, R., Eikermann, J. and Ruppel, J.: ‘The synchronous long-stator
propulsion system for Transrapid - Present state of development’, Int. Conf.
Maglev ’89, 1989, pp 147-154.
21 Boldea, I., Trica, A., Papusoiu, G. and Nasar, S.A.: ‘Field tests on a Maglev
with passive guideway linear induction motor transportation system’, IEEE
Trans. on Veh. Tech., VT-37, 1988, pp 213-9.
22 Iwaya, M.: ‘Dynamic behaviour of the module and suspension system on
HSST-05’, Int. Conf. Maglev ’89, 1989, pp 263-8.
23 Nagata, K., Sekimoto, H. and Murai, M.: ‘Levitation magnet driver for HSST -
Controlled by SI thyristors’, Int. Conf. Maglev ’89, 1989, pp 257-62.
24 Minakata, T., Iwamoto, T, Shimada, Y., Hosada, M. and Kawashima, M.:
‘Magnetically levitated transportation system for automated production lines in
factories’, Int. Conf. Maglev ’89, 1989, pp 375-80.
25 Morishita, M. and Azukizawa, T.: ‘Zero power control of electromagnetic
levitation system’, Electrical Engineering in Japan, 108, 1988, pp 111-20.
26 Jayawant, B.V. and Wheeler, A.R.: ‘Electromagnetic Suspension Railway
Assembly’, British Patent: 1557864, 1979.
References 145
27 Sinha, P.K. and Jayawant, B.V.: ‘Analytical and design aspects of magnetically
suspended vehicles’, Automatica, 15, 1979, pp 539-52.
28 Nakagawa, T. and Yamamura, S.: ‘Electromagnetically levitated vehicle on rails
considering rail oscillation’, Electrical Engineering in Japan, 105, 1985,
pp 123-30.
29 Jones, D.I.: ‘Parallel architectures for real-time control’, IEE Electronics &
Communications Engineering Journal, Sep/Oct, 1989, pp 217-23.
30 Inmos: ‘The transputer applications notebook: Systems and performance’, Inmos
Databook Series, 1989.
31 Kortüm, W. and Utzt, A.: ‘Control law design and dynamic evaluations for a
Maglev with a combined lift and guidance suspension system’, Trans. ASME,
1984, 106, pp 286-92.
32 Sinha, P.K.: ‘Electromagnetic suspension: Dynamics & control’, Peter
Peregrinus, London, 1987, pp 176-216.
33 McLagan, N.S. and Wright, D.P.: ‘Magnetic levitation’, Internal report,
University of Warwick, 1985.
34 Duffin, W.J.: ‘Electricity and magnetism’, McGraw-Hill Book Co., London,
1980, pp 341.
35 Earnshaw, S.: ‘On the nature of the molecular forces which regulate the
constitution of the luminiferous ether’, Trans. Camb. Phil. Soc., 1842, 7,
pp 97-112.
36 Sabnis, A.V.: ‘Analysis of forces in rectangular-pole geometries using numerical
integration techniques’, IEEE Trans. on Mags, 1974, MAG-10, pp 447-50.
37 Brzezina, W. and Langerholc, J.: ‘Lift and side forces on rectangular pole pieces
in two dimensions’, J. of Applied Physics, 1974, 45, pp 1869-72.
38 Kaye and Laby: ‘Tables of physical & chemical constants’, Longmans, London.
39 Chemical Rubber Book Co.: ‘Handbook of chemistry & physics’, Chemical
Rubber Book Co., London.
40 Jayawant, B.V.: ‘Electromagnetic levitation and suspension techniques’,
Edward Arnold, 1981, pp 107-9.
41 Sinha, P.K.: ‘Electromagnetic suspension: Dynamics & control’, Peter
Peregrinus, London, 1987, pp 174.
42 Yamamura, S. and Ito, T.: ‘Analysis of speed characteristics of attracting
magnetic levitation of vehicles’, IEEE Trans. on Mags, 1975, MAG-11,
pp 1504-7.
References 146
43 Borcherts, R.H. and Davis, L.C.: ‘Lift and drag forces for the attractive
electromagnetic suspension system’, IEEE Trans. on Mags, 1974, MAG-10,
pp 425-8.
44 Bohn, G.H.: ‘The influence of eddy currents on an electromagnetic levitation
system’, Conf. on advances in magnetic materials and their applications’,
London, IEE Conf. Publn. No. 142, 1976, pp 104-7.
45 Starling, S.G.: ‘Electricity and magnetism for degree students’, Longmans,
London, 1941, pp 279.
46 Limbert, D.A., Richardson, H.H. and Wormley, D.N.: ‘Controlled dynamic
characteristics of ferromagnetic vehicle suspensions providing simultaneous lift
and guidance’, Trans. ASME, J. of Dynamic Sys., Meas. and Con., 1979, 101,
pp 217-22.
47 Bardell, P.R. ‘Magnetic materials in the electrical industry’, McDonald & Co.,
London, 1960, pp 33.
48 Bardell, P.R. ‘Magnetic materials in the electrical industry’, McDonald & Co.,
London, 1960, pp 108.
49 Jayawant, B.V., Sinha, P.K., Wheeler, A.R., Whorlow, R.J., and Willsher, J.:
‘Development of a 1-ton magnetically suspended vehicle using controlled d.c.
electromagnets’, Proc. IEE, Pt. A, 123, 1976, pp 941-8.
50 Kovach, L.D.: ‘Advanced engineering mathematics’, Addison-Wesley, London,
1982, pp 130.
51 Warwick, K.: ‘Control systems’, Prentice Hall, Hemel Hempstead, 1989, pp 94.
52 McLagan, N.S. and Wright, D.P.: ‘Magnetic Levitation’, Internal Report,
University of Warwick, 1985.
53 Ohno, E.: ‘Introduction to power electronics’, Clarendon Press, Oxford, 1988.
54 Mowry, S.: ‘Current mode control of switching power supplies’, Texas
Instruments Engineering Journal, Mar, 1989, pp 815.
55 Shearer, J.L., Murphy, A.T. and Richardson, H.H.: ‘Introduction to system
dynamics’, Addison-Wesley, London, 1967, pp 366.
56 Jayawant, B.V., Sinha, P.K., Wheeler, A.R., Whorlow, R.J. and Willsher, J.:
‘Development of a 1-ton magnetically suspended vehicle using controlled d.c.
electromagnets’, Proc. IEE, Pt. A, 123, 1976, pp 946.
57 Sinha, P.K.: ‘Electromagnet suspension: Dynamics & control’, Peter Peregrinus,
London, 1987, pp 270-4.
58 Bohn, G. and Alscher, H.: ‘The magnetic train Transrapid 06’, Int. Conf. on
Maglev and Linear drives, 14-16 May, 1986, pp 47-52.
References 147
59 Iwaya, M.: ‘Dynamic behaviour of the module and suspension system on
HSST-05’, Int. Conf. Maglev ’89, 1989, pp 263-8.
60 Phillips, C.L. and Nagle, H.T.: ‘Digital control system analysis and design’,
Prentice-Hall, 1984, pp 245.
61 Sinha, P.K.: ‘Flying low magnetically’, Electronics & Power, 30, 1984,
pp 797-802.
62 ISO: ‘Guide to the evaluation of human body exposure to whole body
vibration’, ISO/DIS 2631, Geneva, 1972.
63 Healey, A.J., Nathman, E. and Smith, C.C.: ‘An analytical and experimental
study of automobile dynamics with random roadway inputs’, J. Dyn. Sys., Meas.
and Cont., Dec. 1977, pp 284-92.
64 Zicha, J. H.: ‘Civil aspects of Maglev design’, Int. Conf. on Maglev and Linear
Drives, 14-16 May, 1986, pp 69-87.
65 Sinha, P.K.: ‘Electromagnet suspension: Dynamics & control’, Peter Peregrinus,
London, 1987, pp 44.
66 Rule, R.G. and Gilliland, R.G.: ‘Combined magnetic levitation and propulsion -
the Mag-Transit concept’, Boeing Aerospace (Automated Guideway Transit),
Technical Paper 6, 1979.
67 Gottzein, E. and Crämer, W.: ‘Critical evaluation of multivariable control
techniques based on Maglev vehicle design’, Proc. 4th Symposium on MVTS,
IFAC ’77, 1977, pp 1-15.
68 Pollard, M.G. and Williams, R.A.: ‘Electro-magnetic suspension’, 2nd Int. Conf.
on Advances in Magnetic Materials and their Applications, London, IEE Conf.
Publn. No. 142, 1976, pp 455-77.
69 Lafrance, P.: ‘Fundamental concepts in communication’, Prentice Hall, London,
1989, pp 68.
70 Marshall, S.A.: ‘Introduction to control theory’, MacMillan, London, 1978,
pp 175.
71 Marshall, S.A.: ‘Introduction to control theory’, MacMillan, London, 1978,
pp 140.
72 Mitchell and Gauthier Associates: ‘Advanced Continuous Simulation Language
(ACSL) Reference Manual’, Mitchell and Gauthier Associates, Concord, Mass.,
USA, 1987.
73 Leigh, J.R.: ‘Applied digital control: Theory, design and implementation’,
Prentice Hall, London, 1985, pp 180.
References 148
74 Leon-Garcia, A.: ‘Probability and random processes for electrical engineering’,
Addison-Wesley, London, 1989, pp 405.
75 Borcherts, R.H. and Davis, L.C.: ‘Lift and drag forces for the attractive
electromagnetic suspension system’, IEEE Trans. on Mags., MAG-10, 1974,
pp 425-8.
76 Sinha, P.K. and Jayawant, B.V.: ‘Analytical and design aspects of magnetically
suspended vehicles’, Automatica, 15, 1979, pp 543.
77 Bohn, G. and Alscher, H.: ‘The magnetic train Transrapid 06’, Int. Conf. on
Maglev and Linear Drives, 14-16 May, 1986, pp 47-52.
78 Wright, P.G. and Williams, D.A: ‘The application of active suspension to high
performance road vehicles’, Int. Congress on Transp. Elecs., 1986, 20-22 Oct,
pp 333-8.
79 Messerschmitt-Bölkow-Blohm: ‘Method for controlling the energisation of the
supporting and guiding electromagnets of a magnetically suspended vehicle’,
British Patent: 1401313, 1972.
80 Pollard, M. and Williams, R.A.: ‘Electro-magnetic Suspension’, 2nd Int. Conf.
on Advances in Magnetic Materials and their Applications, London, IEE Conf.
Publn. No. 142, 1976, pp 455-77.
81 Sinha, P.K.: ‘Electromagnetic suspension: Dynamics & control’, Peter
Peregrinus, London, 1987, pp 193.
82 Shearer, J.L., Murphy, A.T. and Richardson, H.H.: ‘Introduction to system
dynamics’, Addison-Wesley, London, 1971, pp 366.
83 Warwick, K.: ‘Control systems: An introduction’, Prentice Hall, Hemel
Hempstead, 1989, pp 139-59.
84 Marshall, S.A.: ‘Introduction to control theory’, MacMillan Press, London, 1978,
pp 162.
85 Kortüm, W. and Utzt, A.: ‘Control law design and dynamic evaluations for a
Maglev with a combined lift and guidance suspension system’, Trans. ASME,
1984, 104, pp 286-92.
86 Leigh, J. R.: ‘Applied digital control’, Prentice Hall, 1985, pp 12.
87 Leigh, J. R.: ‘Applied digital control’, Prentice Hall, 1985, pp 157.
88 Phillips, C.L. and Nagle, H.T.: ‘Digital control system analysis and design’,
Prentice-Hall, New York, 1984, pp 459-88.
89 McLagan, N.S. and Wright, D.P.: ‘Magnetic Levitation’, Internal Report,
University of Warwick, 1985.
References 149
90 Hnatek, E.R.: ‘Design of solid state power supplies’, Van Nostrand Reinhold,
New York, 1989.
91 Kilgenstein, O.: ‘Switch mode power supplies in practice’, John Wiley & Sons,
Chichester, 1989.
92 Ohno, E.: ‘Introduction to Power Electronics’, Oxford University Press, 1984,
pp 150.
93 Lander, C.: ‘Power electronics’, McGraw-Hill, London, 1987.
94 Asher, G.M. and Summer, M.: ‘Parallelism and the transputer for real-time
high-performance control of AC induction motors’, Proc. IEE, Pt. D, 137, 1990,
pp 179-88.
95 Fleming, P., Jones, D. and Jones, S.: ‘Parallel processing architectures and
applications for real-time control’, American Control Conf., 21-23 June, 3, 1989,
pp 2708-14.
96 Jones, D.I.: ‘Parallel architectures for real-time control’, IEE Electronics &
Communication Engineering Journal, Sep/Oct 1989, pp 217-223.
97 Sinha, P.K. and McLagan, N.S.: ‘Concurrent processing in real-time simulation’,
IEE Conf. on Control ’91, Edinburgh, Mar. 1991, pp 387-392.
98 Hinton, J. C. and McLagan, N. S.: ‘Data link processor system - Architecture
study’, University of Reading, Private contract report, 1989.
99 Geffin, S. and Furht, B.: ‘Transputer-based dataflow multiprocessor for robot
arm control’, Microprocessors and Microsystems, 13, 1989, pp 219-26.
100 Garcia-Nocetti, F., Thompson, H.A., Jones, C.M. and Fleming, P.J.:
‘Implementation of a transputer based flight controller’, Proc. IEE, Pt. D, 137,
1990, pp 130-6.
101 Megson, G.M.: ‘Transputer arrays and computer-aided control system design’,
Proc IEE, Pt. D, 137, 1990, pp 197-210.
102 Ponton, J.W. and McKinnel, R.: ‘Nonlinear process simulation and control using
transputers’, Proc IEE, Pt. D, 137, 1990, pp 189-96.
103 Sinha, P.K., Hinton, J.C. and McLagan, N.S.: ‘Control of an electromagnetic
vehicle suspension system using the Inmos transputer’, 3rd European Conf. on
Power Electronics and Applications, Aachen, Oct. 1989, pp 1595-600.
104 Inmos: ‘Transputer Databook’, Inmos Databook series, 1989.
105 Inmos: ‘The Transputer Applications Notebook: Systems and Performance’,
Inmos Databook Series, 1989.
106 Inmos: ‘Communicating Process Architecture’, Prentice Hall, 1988.
References 150
107 Leigh, J.R.: ‘Applied digital control’, Prentice Hall, London, 1985, pp 14.
108 Inmos: ‘Occam 2 Reference Manual’, Prentice Hall, 1988.
109 Bennett, S.: ‘Real-time computer control’, Prentice Hall, 1988, pp 144.
110 Wirth, N.: ‘Program development by stepwise refinement’, Communications of
the ACM, 14, April 1971, pp 221-7.
111 DeMarco, T.: ‘Structured analysis and system specification’, Prentice Hall, 1979.
112 Houpis, C.H. and Lamont, G.B.: ‘Digital control systems - theory, hardware,
software’, McGraw-Hill, 1985, pp 239.
113 Houpis, C.H. and Lamont, G.B.: ‘Digital control systems - theory, hardware,
software’, McGraw-Hill, 1985, pp 245.
114 Phillips, C. L. and Nagle, H. T.: ‘Digital control system analysis and design’,
Prentice-Hall, 1984, pp 447.