You are on page 1of 8

RINKO M1 , JUNE 2015

Advanced research in state estimation of sensor less


air gap electromagnetic levitation control
Ahmed Salman Student ID 37-145015 , Takafumi Koseki

AbstractThis survey has been conducted to review the


latest advances in the field of sensorless air gap estimation for
electromagnetic levitation control. Sensorless magnetic levitation
can generally be classified into 2 types. 1)Using state observers
and 2) Modulation based. This paper highlights some of the
recent works in modulation based sensorless air gap estimation
and compares the pros and cons for the sub-techniques in the
modulation based approach.
Index TermsMagnetic levitation , sensorless , air gap estimation , self sensing , active magnetic bearings , current change
rate

I. I NTRODUCTION

Fig. 1. Electromagnetic Suspension

AGNETIC levitation, Maglev or magnetic suspension,


is a method by which an object can be suspended in
mid air without any physical support or contacts other than
the magnetic field. The gravitational pull and other external
disturbances are countered only by the magnetic force which
is the reluctance force in this case. Magnetic suspension
boasts significant advantages over conventional mechanical
machines such as high speed ,frictionless, less wear and tear
and durability. Magnetic levitation has found its way into the
industrial applications as, most commonly known, the high
speed train. Active magnetic bearings and bearingless motors
are also a significant part of this field. Other applicataions
include vibration isolation platforms and conveyer systems.
Magnetic levitation can be classified, generally, into two
types. 1) Electromagnetic suspension (EMS) and 2) Electrodynamic suspension (EDS). Though, similar in consequence,
they operate under a different mechanism. EMS uses electromagnets to produce attractive force. This results in an inherently unstable system and thus requires active control system
(Fig. 1). EDS, on the other hand, uses superconductors to
produce repulsive force for suspension resulting in a relatively
stable system (Fig. 2)
This paper reviews some of the recent works related with
the sensor less air gap estimation. Firstly a general overview
of theory of electromagnetic suspension is given in section
II. Section III describes the fundamental requirements for
closed loop control and explains the motivation for sensorless air gap estimation. Section IV introduces the basics of
sensorless estimation and its classification, and highlights the
recent advances to counter the technical problems in the afore
mentioned methods. In Section V a comparison between latest
estimation techniques is shown and finally in section VI a
conclusion is drawn from this review.

Fig. 2. Electrodynamic Suspension: a) Front view b) Side view

II. T HEORY: E LECTROMAGNETIC SUSPENSION AND


RELUCTANCE FORCE

Electromagnetic suspension,as previously mentioned, uses


electromagnets to generate force. This force is called reluctance1 force. Reluctance force is derived from the energy
stored in the magnetic field which can be converted to mechanical energy by principle of virtual work. This force arises
between surfaces with different permeabilities for example
iron core and air. The direction of the force is perpendicular to
the surface of the core and it tends to decrease the reluctance
of the over all system. Fig.3 shows an iron C-I core with
copper wire wound around one of its legs. Voltage is applied
1 Magnetic

resistance

RINKO M1 , JUNE 2015

to the coil resulting in current which produces magnetic flux


in the C-I core. This flux produces reluctance force causing
the I core to be lifted.

Fig. 4. Linkage between electrical, magnetic and mechanical terminals.

according to some predefined control algorithm, computes the


output voltage. This voltage is fed to the power amplifier
which amplifies it and inputs it to the coil resulting in
current and consequently magnetic force. Generally, in an
EMS system, current sensor and displacement sensors are
used. Displacement sensors are based on different physical
phenomena , such as Inductive Displacement sensors, Eddy
current sensors, Capacitive displacement sensors, flux sensors
and optical sensors [1].

Fig. 3. Electromagnet C-I core.

N : N umber of Coil turns.


x(t) : Air gap.
l + 2x(t) : M agnetic path.
fmag : Reluctance f orce
Mathematically, this system can be described by considering
three terminals 1)Electrical 2) Magnetic energy storage and 3)
Mechanical terminal. With each terminal there are associated
variables which govern its side of the dynamics. Fig.4 shows
the linkages in electromagnetic system. The equations governing the dynamics of an electromagnetic suspension system are
given as follows:
d
dt
d((L(x)i(t))
v(t) = i(t)R +
dt

v(t) = i(t)R +

Where L(x) =

(1)
(2)

N 2 A0
2(x(t)+xg )+ lr

fmag = N 2 A0

i2
(2(x(t) + xg ) +

l 2
r )

A. Motivation for sensorless air gap estimation


Over the past few decades, much of the research has been
conducted to achieve stable levitation by applying different
control algorithms. Linear and nonlinear control algorithms
have been applied. Linear control algorithms include conventional lag lead, PID, state feedback controllers. Recently more
advanced and complex controllers, for instance Hinf have also
been applied to demonstrate relatively higher performance in
terms of accuracy, stability, power consumption etc. Being a
nonlinear system, researchers have also demonstrated the use
of nonlinear control algorithms, for instance state feedback
linearization etc, to compensate magnetic nonlinearity and
increase the operating range for the levitation. To provide
logical flow to the conlusion, we turn to a simple example of
state feedback control which requires the expression of system
dynamic model in state space as shown (5).

(3)

dx
= fmag M g
(4)
dt
Using (1),(3),(4) entire magnetic suspension system is defined and analyzed. This set of equations is generally used to
design control system for attaining stable levitation. From the
above set of equations, it is evident that the system is nonlinear.
M

III. C LOSED LOOP CONTROL


Electromagnetic suspension is inherently an unstable system
and that is why it requires an active control system to achieve
stable levitation. A basic control loop is shown in Fig.III.
Key components are a controller, power amplifier and a
displacement sensor. The displacement sensor measures the
air gap and feeds this information to the controller which,

x = Ax + Bu

(5)

y = Cx + Du
The state vector x in (5) has important dynamics variables
which define the complete system. Some common state vectors
used in magnetic levitation control are given in (6).

x air gap
x air gap
x velocity x velocity
i current
x
acceleration

(6)

It is important for control algorithms which use state space


model or its variant to estimate the entire state vector. For
that, system need to be observable i.e from measured output,
all the states can be estimated. Unfortunately this is not the
case with every measured output. In designing control system,
care is taken in choosing outputs which are to be measured
using sensors.

RINKO M1 , JUNE 2015

Fig. 5. Basic control loop for stable levitation

Fig. 6. Changing from a conventionally sensed AMB configuration to


sensorless configuration [1]

For the case of EMS, air gap holds that significance of


providing such facility. By measuring air gap as an output, not
only is the system observable but it also induces robustness.
Intuitively it seems more acceptable as the controlled physical
quantity when the final goal is suspension. But as previously
mention, displacement sensors are required to measure air gap
which add substantial cost to EMS applications. Therefore,
drastic reduction in the cost of an EMS system, is a significant
motivation for sensorless air gap estimation.
IV. S ENSOR LESS AIR GAP ESTIMATION
Sensorless air gap EMS or self-sensing EMS, as the name
suggests,is a technique that uses the voltage and current histories to estimate the air gap without the help of displacement
air gap sensors. The function of the displacement sensor is
replaced by some sort of signal processing which can extract
air gap information hidden inside the current and voltage
of the electromagnetic coil as shown in the Fig. 6. There
are several advantages of this technique, the most obvious
being the drastic reduction in expense due to the higher
cost of the industrial grade displacement sensors. Apart from
that, this also results in reduced hardware, less wiring , can
add redundancy in case of failure of displacement sensor.
Especially for the case of Active magnetic bearings (AMB),
no sensor can allow reduced rotor length with the consequence
of higher bending mode and thus higher rotating speeds. This
technique can also be used in hostile conditions when the
medium of levitation is not air e.g water pumps etc.
Nonetheless, dispensing the air gap sensor comes with
certain technical disadvantages. The EMS becomes less robust
due to the fact that system parameters tend to shift from their
mean values. This makes air gap estimation lose accuracy. Sensorless estimation is also computationally demanding because
of the use of high speed processing units for signal processing.
This method is significantly affected by the eddy currents and
magnetic saturation.
As mentioned previously, air gap information is hidden in
the voltage and current of the EMS coil. This can be seen by
manipulating (2) and using expression of L(x). Note that since
the electrical time constant and mechanical time constants vary
dx
by order of 10s, it is justifiable to neglect
resulting in (7).
dt

Fig. 7. Classification of sensorless air gap estimation.

x(t) + xg =

l
N 2 A0
di(i)

2(v(t) i(t)R) dt
2r

(7)

From (7), if the parameters of the EMS system, the applied


di(t)
voltage v(t) , coil current i(t) and current slope
are
dt
known, air gap can be calculated accurately. However, in
practical implementation, much of the system parameters are
prone to changes which leads to lose of robustness. Over the
past two decades, research has been conducted to over come
such problems related to robustness and different techniques
have surfaced.
A. Classification of sensorless air gap estimation
Sensorless EMS can be classified into two types. 1)Using
State observers and 2) Modulation/Demodulation which is
further divided into two types. The classification is shown in
Fig.7.
The first type which uses state observers is demonstrated
in [2] in which a hybrid observer is used to show the self
sensing EMS. The results are experimentally proven and shows
the difference between measurements by displacement sensor
and estimated air gap. However, it lacks stability and robust
analysis. Such task has been demonstrated in [3] which is
based on the simplified LTI model of the EMS system. This

RINKO M1 , JUNE 2015

Fig. 9. Demodulation of current ripple using HPF and LPF

Fig. 8. Basic closed loop for modulation based self-sensing.

research had concluded the inherent lack of robustness in


the sensorless EMS system when modeled as an LTI system.
However research in this field had not stopped because the
experimentally obtained results by contemporary researches
seem to be in contradiction to it. This paradox has been recently clarified by a comprehensive robustness analysis of selfsensing EMS system by Eric. H.Maslen in [4]. The research
conducts senstivity analysis for both LTI based observers and
linear periodic (LP) based approach in modelling of system.
The results have shown that sensorless EMS shows robustness
if modeled as LP system.
The second category for air gap estimation actually uses
the same approach i.e it involves perturbing the EMS system
continuously. Thus making it an LP system and consequently
more robust. Following is an overview of the different techniques employed in the modulation/demodulation approach.
The Modulation approach involves injecting high frequency
signals along with the control current into the coil. This high
frequency signal injection acts as an interrogation signals to
the EMS system. This signal is inherently present in cases
when the applied voltage is via PWM switching amplifier.
The system dynamics of the EMS cause the high frequency
signals amplitude to be changed based on the value of the
air gap as shown in Fig.8. This current is obtained via signal
processing techniques which is in direct proportion to air gap.
Further this approach is sectioned into two types based on how
this high frequency signal (current ripple) is obtained.
B. Current ripple extraction: Signal processing using
HPF/LPF
As mentioned, the aim is to some how extract the current
ripple which in the frequency spectrum resides at a relatively
high frequency. Signal processing concepts are applied to first
remove the control current spectrum by a high pass filter
(HPF). Then the remaining spectrum is brought down to lower
frequencies by demodulation. This process is depicted in Fig.9.
This process of strongly dependent on the duty rate of the
power amplifier and the magnetic nonlinearity. Phase delays
are caused by the use of HPF and LPF. The drifting parameters
of the EMS also effect the estimation. In the recent trends,
these issues have been dealt with. Some recent research on
mitigation of these issues follow:
1) New results for self-sensing AMB using modulation
approach [5]: To compensate the duty rate variation, this

Fig. 10. Estimation algorithm with permeability compensation.

research proposes the use of voltage and current signals


simultaneously. Fig.10 shows the estimation algorithm. The
applied voltage is also band pass filtered and demodulated to
obtain ud . Equation (8) shows the dependency of demodulated
coil current on air gap x, magnetic offset lmr and demodulated
voltage vd (Duty rate). The effects of duty rate and magnetic
offset are compensated. The magnetic nonlinearity is compensated by estimation of magnetic field from iL and subtracting
this offset from uidd .
1
lm
(2(x + x0 ) +
)ud
Ks
r
id = (kx x + kb2 B 2 + kb1 B + kb0 )ud
1 id
xge =
kx ud
id =

(8)

The algorithm is implemented on an experimental test rig. A


linear curve between the measured position and estimation position is achieved as depicted in Fig.11. The research concludes
by stating that systems robustness can further be improved
by modeling the systems nonlinearities more accurately. A
critical drawback for this technique is the use of an additional
voltage sensor for compensating duty rate effect.
2) Magnetic displacement of one DOF using simultaneous actuation and displacement sensing technique[6]: The
research deals with demonstration of removing the effect of
variable duty rate of PWM switching power amplifier from
the air gap estimation. The concept is based on the the idea of
D.Noh [7]. The concept is similar to the use of state observers.
Difference lies in the fact that the instead of the output i(t)
to be compared to generate error for
and simulated output i(t)
observer, the current ripple is used. The algorithm is shown
in Fig. 12.

RINKO M1 , JUNE 2015

Fig. 11. Estimator static performance compared with the position sensor.
Estimated position (solid)- measured position (dotted)

Fig. 13. Closed loop employed for direct measurement of current slope using
transformer.

C. Direct Measurement of current ripple change rate

Fig. 12. Self sensing parameter estimation scheme for determining air gap.

Both the forward path filters consist of a BPF , a full wave


rectifier and a low pass filter. Its purpose is to generate a signal
proportional to the air gap but at a low frequency. The outputs
of each forward path filter are subtracted to generate an error
which is fed to a controller, Proportional-Integral (PI) in this
case. The output is the estimated air gap. The controller serves
to keep the error e zero. As long as this holds, the estimated
air gap is well within acceptable bounds for stable levitation.
A critical drawback for this method is its dependency on
the simulated model of the EMS. Since actual hardware can
shift from mean values of its parameters, the robustness of
this method is questionable. This can also be attributed to
the simplified EMS model. If EMS system is modeled by
taking the neglected phenomena of magnetic saturation, cross
coupling and eddy currents, as has been done in [8] , this
approach can be improved. Use of adaptive control algorithms
to compensate the changing parameters can be a good solution
for this issue.

Unlike the demodulation approach which appears mathematically challenging, this approach is somewhat simple. As
the name suggest, it involves direct measurement of the change
di(t)
of current ripple. It is evident from (7) that measuring
dt
holds significance in air gap estimation. Since, the current ripples frequency spectrum lies at high frequencies, measuring a
rapidly changing signal demands fast computational processors
and samplers. Nonetheless, due to availability of high speed
processors, this approach has been pursued by many. Initially,
the idea was first proposed and demonstrated by Lichuan Li in
di(t)
) is measured by using a transformer
[9]. Current slope(
dt
like coil whose output voltage is directly proportional to the
current slope as shown in Fig.13. Though the approach is new,
it involves yet use of another sensor, (extra coil) , side by
side current sensor, which could be used for the purpose of
displacement sensing as well. Another drawback stated for
this method is that it required exact timing for sampling the
transformers output. This is attributed to the transient in the
current ripple due to the fast switching PWM. An improvement
in for this case would be to measure current slope without such
extra hardware. This problem has been addressed the following
way by some recent works.
1) Direct current measurement approach [10]: This is a
recent work based on the original work of [5]. However, this
is an improvement from it as it doesnt use extra voltage
sensors to eliminate the duty rate effects. The idea is based on
the measurement of current ripple amplitude to approximate
current slope and consequently estimate the air gap. The
algorithm scheme in shown in Fig. 14. The sampled current
is averaged out and subtracted from within itself to extract
out the current ripple. The maximum value of ripple is taken
which is when multiplied with a constant kx gives estimated
air gap. Further magnetic offset is subtracted as had been done
in [5].
Since the operating frequency of the PWM switching power
amplifier is high at 20Khz, the nonlinear dependency of the

RINKO M1 , JUNE 2015

Fig. 14. DCM Self-sensing approach

Fig. 16. Position errors for Simulation and experiment

just rising slope, falling slope is also calculated. This adds


additional advantage which can be inferred from the final
equation (10) derived for the position estimation. The air gap
in this case doesnt depend on the resistance of the coil or the
velocity of the levitating rotor. These two quantities have been
canceled out by the use of double detection of current slopes.
PM 1
i n, =

(i n,,j in, )(tn,,j tn, )


PM 1
2

j=0 ((tn,,j tn, ) )

j=0

(9)

0 AN 2
(i1 ,n i 2 ,n )
(10)
2Udc
For the rejection of duty cycle effects, this method follows
the same approach of fixing the duty cycle at 0.5. This on one
hand doesnt disrupt the current control but it does decrease is
performance in terms of slew rate for current rise and current
fall as shown in Fig.18. Thus a compromise is made between
the accuracy of the position estimation and the dynamics of
the current control.
This estimation method is also experimentally verified by
using it for an Active magnetic bearing. The block diagram
for the control system is shown in Fig.19. The hardware
consists just of power amplifiers and a current sensor. This
is a significant edge over the previously mentioned estimation
methods. The experimental results are shown in 20 which are
acceptable for the application of AMB.
xn =

Fig. 15. Raw coil voltage and current showing measurement and control
cycles.

air gap estimation on the duty cycle is restricted to be the


same, 0.5 at the time current ripple is measured. This divides
the PWM cycle into a sensing cycle and a measurement
cycle as shown in Fig.15. By doing this, the current control
performance is degraded or to achieve the same current slope
slew rate, the applied voltage is required to be doubled.
The proposed method is compared with the results of [5]
as shown in Figure.16. The position error for the case of
simulated DCM remains approximately zero for a range of
about 400s unlike that of Schammass [5]. However, the
experimental results show some odd variation throughout the
operating range. Yet, the error still remains quite low.
2) Detection of current slope using multiple current samples [11][12][13]: This works has proposed a new improved
method of direct measurement of current slope by using high
speed ADC to sample current multiple times in one duty cycle
as shown in Fig.17. M samples are then used to estimate
the current slope using least square method as given in (9).
It should be noted that in the latest work [13], instead of

V. C OMPARISON
After a short review of some of the recent works in the
field of sensorless air gap estimation, Table.V sums up the
comparison between different techniques. General trend observed is that the modulation approach using filters somewhat
lacks good duty rate rejection and often involves an additional
sensor apart from current sensor thus increasing hardware. The
mathematical analysis also is relatively complex unlike the
direct current slope measurement which is straight forward.
Duty cycle rejection is seen compensated significantly by the

RINKO M1 , JUNE 2015

Fig. 17. Principle of multiple sampling [11]

Fig. 20. Estimation performance of step change of the real position in y-axis.
a)Position of x-axis during the step change in the y-axis. b)Zoom of a. c)Step
change of real position in y-axis, d)zoom of c.

Fig. 18. PWM pattern with split sensing and control cycles.

[10] and [13]. But this is done at the expense of reduced


current control performance.
It should also be noted that comparison is based on the
commonly touched topics between these researches. However,
a comprehensive comparison involving additional comparison
parameters for instance stability, current control performance
etc could be incorporated.
Modulation approach using filters is observed to be mathematically more complex and involved relatively higher degree
of mathematical analysis e.g the derivation of the current
ripple equation after signal processing. Unlike this, the direct
current slope measurement is straigh forward in sense, the the
original equation for air gap estimation is implementedw with
an increased effort in calculation of the current ripple slope.
Magnetic saturation is seem to be compensated only in [5]
and its improved method [10]. Apart from these, the remaining
presented methods do not explicitly deal with this issue.
In case of hardware complexity, most of the works presented
included either additional sensors or filters. The work done by
[13] employs just current sensors while most of the algorithm
is implemented digitally.
As mentioned before in IV, eddy currents can affect the
accuracy of the estimation. Most of the researches have
sufficed by avoiding it by fixed timing or limitation of the duty
cycle since the such effects are predominant at the transition
times between positive and negative pulses.
VI. C ONCLUSION AND FUTURE PROSPECTS

Fig. 19. Block diagram of the control system

A comprehensive survey in the field of sensorless air gap


estimation has been conducted. Two commonly researched approaches, modulation approach using signal processing filters
and direct current slope measurement, have been dealt with.
Each approach has its own pros and cons, however the latter

RINKO M1 , JUNE 2015

TABLE I
C OMPARISON BETWEEN DIFFERENT SENSORLESS AIR GAP ESTIMATION TECHNIQUES
+ Good
bad
Duty Cycle variation effects
Less Complexity
Magnetic
Saturation
Compensation
Reduced
Hardware
components
Eddy
current
compensation

Signal processing using filters(demodulation)


Original
Schammass Tau
Idea
2005
2011

Direct current slope measurement


Lichuan
Niemann
Wang,
Li 2004
2013
2014

++

+++

+++

++

+++

++

++

++

++

++

+++

one has been observed to have few technical edges over the
former one. For instance, increased simplicity and reduced
hardware. See Table.V.
In section IV, inherent lack of robustness was mentioned
as a major disadvantage of sensorless approach. The recent
works have definitely proved otherwise. It has been shown that,
though not as robust as the displacement sensor incorporated
control, the sensorless estimation methods when use the high
frequency injected interrogative signal make the over all
system gain robustness.
It has been observed that the theoretical basis and stability
analysis, for the direct current slope measurement is not widely
presented. Most researches extract the sensitivity curves experimentally. The author is of the view that a logical and rational
theoretical basis is a necessity if further knowhow about the
behavior of the such techniques is aimed.
Since, none of the listed works deal with eddy current
compensation and most have sufficed by avoided it by fixed
timings, this could be a ripe area of research for sensorless air
gap estimation.
Generally, industrial EMS system multiple coils which
raises yet another issue of cross-coupling which consequently
effects the estimation accuracy. A crude side-solution to this
problem is presented in [8]. However focused research into
this issue might hold additional possibilities.
R EFERENCES
[1] M. E. H. E. Schweitzer, Gerhard, Magnetic Bearings, Theory, Design,
and Application to Rotating Machinery, 2009.
[2] Y.-K. Tzeng and T. Wang, A novel compensating approach for selfsensing maglev system with controlled-pm electromagnets, Magnetics,
IEEE Transactions on, vol. 31, no. 6, pp. 42084210, Nov 1995.
[3] N. Morse, R. Smith, B. Paden, and J. Antaki, Position sensed and
self-sensing magnetic bearing configurations and associated robustness
limitations, in Decision and Control, 1998. Proceedings of the 37th
IEEE Conference on, vol. 3, 1998, pp. 25992604 vol.3.
[4] D. T. Eric H. Maslen and T. iwasaki, Robustness limitations in selfsensing magnetic bearings, Journal of Dynamic Systems, Measurement,
and Control, 2006.
[5] A. Schammass, R. Herzog, P. Buhler, and H. Bleuler, New results
for self-sensing active magnetic bearings using modulation approach,
Control Systems Technology, IEEE Transactions on, vol. 13, no. 4, pp.
509516, July 2005.
[6] T. M. Lim and S. Cheng, Magnetic levitation of a one {DOF} system
using simultaneous actuation and displacement sensing technique,
Mechatronics, vol. 21, no. 3, pp. 548 559, 2011. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S0957415811000213

[7] M. Noh and E. H. Maslen, Self-sensing magnetic bearings using parameter estimation, Instrumentation and Measurement, IEEE Transactions
on, vol. 46, no. 1, pp. 4550, Feb 1997.
[8] E. Ranft, G. van Schoor, and C. du Rand, Self-sensing for electromagnetic actuators. part ii: Position estimation, Sensors and Actuators
A: Physical, vol. 172, no. 2, pp. 410 419, 2011. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S0924424711005516
[9] L. Li, T. Shinshi, and A. Shimokohbe, State feedback control for active
magnetic bearings based on current change rate alone, Magnetics, IEEE
Transactions on, vol. 40, no. 6, pp. 35123517, Nov 2004.
[10] A. C. Niemann, G. van Schoor, and C. P. du Rand, A self-sensing
active magnetic bearing based on a direct current measurement
approach, Sensors, vol. 13, no. 9, p. 12149, 2013. [Online]. Available:
http://www.mdpi.com/1424-8220/13/9/12149
[11] J. Wang and A. Binder, Self-sensing magnetic bearings using multiple
sampling of currents alone, in Power Electronics and Applications
(EPE), 2013 15th European Conference on, Sept 2013, pp. 110.
[12] , Current slope calculation in fpga for sensorless control technique and associated slope based predictive precise current control,
in Sensorless Control for Electrical Drives and Predictive Control of
Electrical Drives and Power Electronics (SLED/PRECEDE), 2013 IEEE
International Symposium on, Oct 2013, pp. 18.
[13] A. B. Jinou Wang, Position estimation for self-sensing magnetic bearings based on double detection of current slopes, in 14th International
symposium on Magnetic Bearings, 2014.