See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281861536

DYNAMICS AND CONTROL DESIGN VIA LQR AND SDRE

METHODS FOR A MAGLEV SYSTEM

Article  in  Global Journal of Pure and Applied Mathematics · January 2015

DOI: 10.12732/ijpam.v101i2.13

CITATION READS
1 4,271

5 authors, including:

Fábio Roberto Chavarette

São Paulo State University
114 PUBLICATIONS   300 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Energy Harvesting View project

All content following this page was uploaded by Fábio Roberto Chavarette on 18 September 2015.

The user has requested enhancement of the downloaded file.

International Journal of Pure and Applied Mathematics
Volume 101 No. 2 2015, 289-300
ISSN: 1311-8080 (printed version); ISSN: 1314-3395 (on-line version)
url: http://www.ijpam.eu
doi: http://dx.doi.org/10.12732/ijpam.v101i2.13
AP
ijpam.eu

DYNAMICS AND CONTROL DESIGN VIA

LQR AND SDRE METHODS FOR A MAGLEV SYSTEM

Thalles Denner Ferreira Cabral1 § , Fábio Roberto Chavarette2

1 UNESP- Universidade Estadual Paulista
Department of Mechanical Engineering
Brazil Avenue, 56, 15385-000, Ilha Solteira, SP, BRAZIL
2 UNESP – Universidade Estadual Paulista

Department of Mathematics
Brazil Avenue, 56, 15385-000, Ilha Solteira, SP, BRAZIL

Abstract: Several experimental maglev systems all around the world, mainly
in Germany and Japan have demonstrated that this mode of transportation
can profitably compete with air travel. However, a system such as the German
maglev train (called Transrapid) is inherently unstable. This instability is be-
cause the electromagnetic suspension (EMS) uses attractive force to levitate the
train. So, the electromagnets of the vehicle must be actively controlled to make
safe operation. Herewith, from a simplified model for the German Transrapid
experimental system, we propose two control designs and, then we compare
them. The linear quadratic regulator (LQR) is used to design the linear con-
troller and the state-dependent Riccati equation (SDRE) is used to design the
nonlinear controller. The simulation shows that the SDRE controller allows the
maglev train to operate with much larger disturbances in the air gap than the
LQR controller does.

AMS Subject Classification: 93C10

Key Words: computer simulation, dynamics systems, mathematical model,
maglev system, optimal control
c 2015 Academic Publications, Ltd.

Received: January 19, 2015 url: www.acadpubl.eu
§ Correspondence author
290 T.D.F. Cabral, F.R. Chavarette

1. Introduction

Maglev trains (Magnetic levitation transport) are transport systems that can
achieve high speeds with low friction compared to conventional wheel-rail trains.
Other factors that become the very interesting maglev trains technology is
that it can ease traffic congestion and helps reduce negative environmental
impacts. Several engineers around the world, have developed projects related
with the maglev trains technology, being that some these projects are in a
relatively advanced stage. Several experimental maglev systems all around the
world, mainly in Germany and Japan have demonstrated that this mode of
transportation can profitably compete with air travel [1] and [2].
The German maglev train is the first system that operates completely non-
contact with the guideway and without wheels. German scientists have de-
signed a system called Transrapid which uses the technology of electromagnetic
suspension (EMS) to levitate a train (see Fig. 1a). The Transrapid maglev
train has a 31.5km test track in Germany, Emsland (TVE) and is commercially
operated in Shanghai, China, since 2004 [3].
The levitation occurs through electromagnets that are strategically posi-
tioned on a series of C-shaped arms. Thus, these electromagnets exert a force
of attraction in the ferromagnetic plates installed on the flange of a T-Beam [2].
This configuration allows the maglev train levitate and has lateral stability.
An electromagnetic suspension (EMS) system such as of the Transrapid ma-
glev train is inherently unstable. Therefore, the electromagnets of the vehicle
must be actively controlled to make safe operation [3] and [4]. So, from a sim-
plified model for the German Transrapid experimental system, we propose two
control designs and, then we compare them. The paper is organized as follows.
In Section 2, we present two simplified models for the German Transrapid ex-
perimental system. In Section 3, we discuss some issues about dynamics of the
electromagnetic suspension (EMS) system. In Section 4, we present two control
designs and the results of computer simulations. In Section 5, the conclusions
are presented.

2. The Maglev Model

Figure 1b shows the magnetic levitation system of the Transrapid maglev train,
which is simplified as a single mass system on a rigid guideway. The primary
suspension consists of electromagnets. Suspension systems are dominant in de-
termining the basic dynamic and vibrational behavior of the maglev vehicle. A
DYNAMICS AND CONTROL DESIGN VIA... 291

two-degree-of-freedom model considering coupling effects of primary and sec-

ondary suspensions was studied by [5]. Note that the model presented in this
paper hasn’t got a secondary suspension, however, we can use it to simulate
the vehicle vertical dynamics [1], [5] and [6].

(a) Electromagnetic suspension. (b) Simplified model.

Figure 1: Magnetic levitation system of the Transrapid maglev train.

where, mg is the vehicle weight (including the electromagnets), sM is the ver-

tical gap between the guideway and the vehicle, S0 is the vertical gap between
the guideway and the equilibrium position of the vehicle, L(sM ), R, iM and uM
are the coil inductance, resistance, current and voltage, respectively.
The inductance L(sM ) is a nonlinear function of vehicle position (sM ).
According to [7] various approximations have been used to determine the in-
ductance. Such as [6], [8] and [9] we shall neglect the leakage flux and eddy
current effects, so that the inductance varies with the inverse of vehicle position
as follows:
c µ0 N 2 A
L(sM ) = wherein, c= (1)
sM 2
where, µ0 is the inductance constant, A is the pole area and N is the number
of coil turns.

2.1. Nonlinear Model Maglev

The nonlinear motion equations of the model shown in Fig. 1b are:

c 2
ms̈M = − i + mg
2s2M M
iM RiM uM
i̇M = ṡM − sM + sM (2)
sM c c
292 T.D.F. Cabral, F.R. Chavarette

The nonlinearities of the system come from the nonlinear inductance due
to the geometry of the magnet and the inverse square magnetic force law [1].
Details of the derivation of the model are discussed in [4] and [6].

2.2. Linear Model Maglev

For small deviations from steady-state as shown in Fig. 1b, we can write:

sM = S0 + δs, iM = I0 + δi, uM = U0 + δu, Ṡ0 = 0, I˙0 = 0, U0 = I0 R (3)

Equation 1:
The electromagnetic force of the magnetic levitation system (useful for next
calculations) is found using the concept of co-energy [8]. To calculate it, we
consider a linear relationship between the flux and the current, so that the
force electromagnetic can be written as follows:
 2
1 dL(sM ) c i
f (sM , i) = − i2 ⇒ f (sM , i) = (4)
2 dsM 2 sM

The system has one equilibrium state at which the force electromagnetic
exactly counterbalances the force due to gravity, i.e., mg = f (S0 , I0 ). So,
substituting (3) into Eq. (2) and using (4), we have:
 2  2
c I0 + δi c I0
mδs̈ = − +
2 S0 + δs 2 S0

Equation 2:
Substituting (3) into Eq. (2), we have:

I0 + δi R(I0 + δi)(S0 + δs) (U0 + δu)(S0 + δs)

δi̇ = δṡ − +
S0 + δs c c

Finally, neglecting the terms of higher order and making some rearrange-
ments, we obtain the linearized motion equations:

cI0 δi cI02 δs
δs̈ = − +
mS02 mS03
I0 RS0 S0
δi̇ = δṡ − δi + δu (5)
S0 c c
DYNAMICS AND CONTROL DESIGN VIA... 293

3. Dynamics of Electromagnetic Suspension System

The nonlinear system (2) can be rewritten in state space as follows:

x3 (uM − Rx1 ) x1 x2
ẋ1 = +
c x3
2
cx1
ẋ2 = − +g (6)
2mx23
ẋ3 = x2

Similarly, the linearized system (5) can be rewritten in state space as follows:

S0 (δu − Rx1 ) I0
ẋ1 = + x2
c S0
 
cI0 I0
ẋ2 = x3 − x1 (7)
mS02 S0
ẋ3 = x2

The state variables x1 , x2 and x3 represent coil current, vertical velocity of

the vehicle, and vertical gap between the guideway and the vehicle, respectively,
for both systems (6) and (7). Rearranging (1), we can calculate the constant c
as follows: c = L(sM )sM , however, when the vehicle is in the equilibrium state,
we have sM = S0 and L0 = L(S0 ), therefore, c = L0 S0 .
As we’ve seen previously, the system has one equilibrium state when the
electromagnetic force exactly counterbalances the gravitational force and the
vehicle hasn’t got vertical velocity and acceleration. However, the equilibrium
is a saddle node which is unstable in the sense of Lyapunov (see Fig. 2a).
Details of the stability analysis are discussed in [1] and [4]. Table 1 shows the
system parameters we used in our simulation.

m[kg] g[m/s2 ] S0 [m] L0 [h] R[Ω] uM [v] I0 [A] δu[v]

10000 9.8 0.01 0.1 1 140 uM /R uM − RI0

Table 1: Model parameters.

Figure 2 shows the phase portrait and a trajectory of the system (6) and
(7) in the x2 − x3 projection.
Figure 2a shows the phase portrait of the linearized model (7) around equi-
librium point, which is topologically orbitally equivalent to nonlinear model (6).
However, in regions far from the equilibrium point, the linearized model doesn’t
294 T.D.F. Cabral, F.R. Chavarette

10 250
Nonlinear Nonlinear
8 Linear Linear

6 200

2 150
X2 [mm/s]

x [mm/s]
0

2
−2 100

−4

−6 50

−8

−10 0
9.7 9.8 9.9 10 10.1 10.2 10.3 10 12 14 16 18 20
X3 [mm] x3 [mm]

(a) Phase portrait. (b) Trajectory of the system.

Figure 2: Simulation computer system (6) and (7).

give good approximation the nonlinear model (see Fig. 2b). Therefore, linear
controllers tend to saturate at the beginning of large initial displacements of
the system equilibrium point [1].

4. Optimal Control Design

The control objective is to stabilize the maglev vehicle traveling above a guide-
way and, to maintain a constant distance between the vehicle and the guideway.
The control parameter is the coil input voltage.

4.1. Control Design via Linear Quadratic Regulator

The LQR approach for obtaining an optimal solution of the control problem
has the following procedure:

1. Represent the model in state-space form, and rewrite it as follows:

ẋ = Ax + Bu
(8)
y = Cx

where, A ∈ ℜnxn is the dynamic matrix, B ∈ ℜnxm is the input matrix,

C ∈ ℜsxn is the output matrix, x ∈ ℜn is the state vector, u ∈ ℜm is the
control law, y ∈ ℜs is the output vector.
DYNAMICS AND CONTROL DESIGN VIA... 295

2. Define the initial conditions x(0) = x0 , and choose the coefficients of

positive definite weighting matrices Q and R, which determine the relative
importance of state x(t) and control effort u(t), respectively.

3. Solve the Riccati equation given by:

AT P + P A − P BR−1 B T P + Q = 0 (9)

4. Construct the linear feedback control via:

u = −R−1 B T P x (10)

The
R ∞ control law (10) is calculated so that the performance index given by,
J = t0 (xT Qx + uT Ru)dt is minimized.
The technique LQR requires that the linear system is controllable. Details
about the technique LQR can be found in [10] and [11].
Applying the above procedure in the linearized system (7), we obtain:
 RS0 I0 
− c S0 0 x1   S0 
c
 cI0 cI 2 
ẋ = − mS 2 0 mS03  x2  +  0  u (11)
0 0
0 1 0 x 3 0

where, u = δu.
The system output and the coefficients chosen for the matrix Q and R are:
    
1 0 0 x1 1 0 0  
y = 0 1 0 x2  Q = 0 1 0  R= 1 (12)
0 0 1 x3 0 0 3.1x1012

4.2. Control Design via State-Dependent Riccati Equation

The SDRE nonlinear regulator has the same structure of the linear quadratic
regulator (LQR), except that all the matrices are state-dependent. The SDRE
approach for obtaining a suboptimal solution of the control problem has the
following procedure [12] and [13]:
1. Represent the model in state-space form. Use direct parametrization to
bring the nonlinear dynamics ẋ = f (x) + g(x) to the state-dependent
coefficient (SDC) form, as follows:

ẋ = A(x)x + B(x)u
(13)
y = C(x)x
296 T.D.F. Cabral, F.R. Chavarette

where, f (x) = A(x)x and g(x) = B(x), A(x) ∈ ℜnxn is the dynamic
matrix, B(x) ∈ ℜnxm is the input matrix, C(x) ∈ ℜsxn is the output
matrix, x ∈ ℜn is the state vector u ∈ ℜm is the control law, y ∈ ℜs is
the output vector.

2. Define the initial conditions x(0) = x0 , and choose the coefficients of

positive definite weighting matrices Q(x) and R(x), which determine the
relative importance of state x(t) and control effort u(t), respectively.

3. Solve the state-dependent Riccati equation given by:

AT (x)P (x) + P (x)A(x) − P (x)B(x)R−1 (x)B T (x)P (x) + Q(x) = 0 (14)

4. Construct the nonlinear feedback control via:

u = −R−1 (x)B T (x)P (x)x (15)

TheR ∞control law (15) is calculated so that the performance index given by,
J = 21 t0 [xT Q(x)x + uT R(x)u]dt is minimized.
In the multivariable case, there always exists an infinite number of SDC
parameterizations. Therefore, the choice of the matrix A(x) isn’t unique [12].
The pair {A(x), B(x)} is a controllable parametrization of the nonlinear
system in a region Ω if {A(x), B(x)} is pointwise controllable in the linear
sense for all x ∈ Ω. Therefore, the choice of A(x) must be such that the state-
dependent controllability matrix [B(x) A(x)B(x) ... An−1 (x)B(x)] has full
rank [13].
The SDRE technique has been used to control various systems, such as
agroecosystems [14], non-ideal systems with chaotic behavior [15], etc. Details
about the technique SDRE can be found in [12] and [13].
Applying the above procedure in the nonlinear system (6), we obtain:
 x2 2x1
− Rxc 1 x1  0  x3 

− x3 x3 c
 cx1 cx21   
ẋ = − mx 2 0 2mx3 3  x 2 + g  +  0  u (16)
3
0 1 0 x3 0 0

where, u = uM .
Assuming C(x), Q(x) and R(x) are constant matrices, we have:
    
1 0 0 x1 1 0 0
 R = 1x10−1
 
y = 0 1 0 x2
    Q= 0 1 0 (17)
0 0 1 x3 0 0 1.0085x108
DYNAMICS AND CONTROL DESIGN VIA... 297

4.3. Computer Simulations

Assume that u is produced by a buck converter capable of delivering any voltage

0 ≤ u ≤ 450[V ], and that the power supply delivers 140 volts at the equilib-
rium position of the vehicle. With respect to the state variables, consider the
following domain of discourse for the analysis: {(x1 , x2 , x3 ) ∈ ℜ | 0 ≤ x1 ≤
300[A], −0.2 ≤ x2 ≤ 0.2[m/s], 0 < x3 ≤ 0.02[m]}.
We emphasize that the control designs are in standard form. Therefore, we
haven’t considered constraints on the state variables x(t) and control signal u(t).
Obviously, constraints such as the control signal saturation can be implemented,
so, the results below may be different.
The responses of the controlled system (11) and (16) considering two dif-
ferent initial conditions are shown in Figs. 3, 4 and 5.

LQR 10.45
SDRE LQR
10.4 SDRE

10.5
10.35
10.4
10.3
10.3
x3 [mm]

10.25
x3 [mm]

10.2
10.2
10.1 X: 140
Y: 0
Z: 10.2
10.15
10
5 10.1
148
0 146
−3
10.05
x 10 144
−5 142
140 10
x [mm/s] −10 138 0 0.5 1 1.5 2
2 x [A]
1 tempo [s]

(a) Trajectory of the controlled system. (b) Time history of the vertical gap.

Figure 3: Response of the controlled system for x(0) = (140, 0, 0.0102).

Table 2 shows some transient-response characteristics.

x(0) Control x1max [A] x2max [mm/s] x3max [mm] umax [V ] te[s]
(140,0,0.0102) LQR 147.87 -6.60 10.20 427.96 0.1
- SDRE 147.01 -5.05 10.35 157.62 1.5
(140,0,0.015) LQR 431.17* -249.79* 15.00 10699* 0.1
- SDRE 281.16 -80.73 16.64 448.39 1.5

Table 2: Some transient-response characteristics

In Tab. 2 the values with superscript (*) are outside of the domain of
298 T.D.F. Cabral, F.R. Chavarette

LQR 17
SDRE LQR
16 SDRE

18
15
16
14
X: 140
14
x [mm]

Y: 0
13

x3 [mm]
Z: 15
12
3

12
10
11
8
0.1 10
0 600
−0.1 400 9
200
−0.2 0 8
x [mm/s] −0.3 −200 0 0.5 1 1.5 2
2 x [A]
1 tempo [s]

(a) Trajectory of the controlled system. (b) Time history of the vertical gap.

Figure 4: Response of the controlled system for x(0) = (140, 0, 0.015).

450 500
LQR LQR
SDRE 450 SDRE
400
400
350
350
300
300
u(t)

u(t)

250 250

200
200
150
150
100
100
50

50 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
tempo [s] tempo [s]

(a) x(0) = (140, 0, 0.0102). (b) x(0) = (140, 0, 0.015).

Figure 5: Control effort u(t).

discourse.

5. Conclusion

The simulation results shows that both designed controllers are able to sta-
bilize the vehicle traveling above a guideway. However, in the case of the
LQR controller, the control signal amplitude is very high if we consider initial
displacements δs > 0.2mm, so that the voltage required to bring the vehicle
DYNAMICS AND CONTROL DESIGN VIA... 299

back to the equilibrium point is larger than the maximum voltage produced
by buck converter, i.e., u > 450V . This is because the LQR controller design
uses the linearized model (7) which doesn’t give good approximation the non-
linear model (6) in regions far from the equilibrium point (as seen in Section
3). On the other hand, the SDRE controller can bring the system back to the
equilibrium with an initial displacement of up to 5mm from the equilibrium
position. Furthermore, the control effort is less than in the LQR. With respect
the settling time and maximum overshoot, the LQR controller is better than
the SDRE controller. However, considering the domain of discourse of x(t) and
u(t) we can said that the SDRE controller has better overall performance than
the LQR controller. Moreover, the simulation shows that the SDRE nonlinear
controller outperforms the linear controller LQR by a factor of 25 times with
respect the maximum recoverable displacement.

Acknowledgment

The authors thanks Conselho Nacional de Pesquisas CNPq for financial sup-
ports (Proc. n◦ 132786/2013-3, Proc. n◦ 301769/2012-5).

References

[1] F. Zhao, R. Thornton, Automatic design of a maglev controller in state

space, Proceedings of the 31st Conference on Decision and Control, 3
(1992), 2562-2567, doi: 10.1109/CDC.1992.371062.

[2] E. G. David, O Futuro das Estradas de Ferro no Brasil, Portifolium, Brasil

(2009).

[3] H.-W. Lee, K.-C. Kim, J. Lee, Review of maglev train technologies,
IEEE Transactions on Magnetics, 42, No. 7 (2006), 1917-1925, doi:
10.1109/TMAG.2006.875842.

[4] T. D. F. Cabral, F. R. Chavarette, Anlise de estabilidade e projeto de

controle pelo mtodo SDRE para um sistema maglev simplificado, Anais
do Congresso Nacional de Matemtica Aplicada a Indstria, (2014).

[5] Y. Cai, S. S. Chen, Control of maglev suspension systems, Jour-

nal of Vibration and Control, 2, No. 3 (1996), 349-368, doi:
10.1177/107754639600200305.
 300 T.D.F. Cabral, F.R. Chavarette

[6] G. Shu, R. Meisinger, State estimation and simulation of the magnetic

levitation system of a high-speed maglev train, International Conference
on Electronic & Mechanical Engineering and Information Technology, 2
(2011), 944-947, doi: 10.1109/EMEIT.2011.6023250.

[7] I. Ahmad, M. A. Javaid, Nonlinear model and controller design for mag-
netic levitation system, Proceedings of the 9th WSEAS International Con-
ference on Signal Processing, Robotics and Automation (2010), 324-328.

[8] T. T. Salim, V. M. Karsli, Control of single axis magnetic levita-

tion system using fuzzy logic control, International Journal of Ad-
vanced Computer Science and Applications, 4, No. 11 (2013), 83-88, doi:
10.14569/IJACSA.2013.041111.

[9] A. Suebsomran, Optimal control of electromagnetic suspension EMS sys-

tem, The Open Automation and Control System Journal, 6 (2014), 1-8,
doi: 10.2174/1874444301406010001.

[10] K. Ogata, Designing Linear Control Systems with MATLAB, Prentice Hall,
United States (1993).

[11] R. C. Dorf, Sistemas de Controle Modernos, Livros Tcnicos e Cientficos,

Brasil (2001).

[12] J. R. Cloutier, C. N. D’ Souza, C. P. Mracek, Nonlinear regulation and

nonlinear H∞ control via the state-dependent Riccati equation technique,
Proceedings of the International Conference on Nonlinear Problems in Avi-
ation and Aerospace (1996), 117-142.

[13] C. P. Mracek, J. R. Cloutier, Control designs for the nonlinear

benchmark problem via the state-dependent Riccati equation method,
International Journal of Robust and Nonlinear Control, 8 (1998),
305-461, doi: 10.1002/(SICI)1099-1239(19980415/30)8:4/5<401::AID-
RNC361>3.0.CO;2-U.

[14] A. Molter, M. Rafikov, Controle timo em agroecossistemas usando SDRE,

Tendncias em Matemtica Aplicada e Computacional, 12, No. 3 (2011),
221-232, doi: 10.5540/tema.2011.012.03.0221.

[15] F. R. Chavarette, Control design applied to a non-ideal structural system

with behavior chaotic, International Journal of Pure and Applied Mathe-
matics, 86, No. 3 (2013), 487-500, doi: 10.12732/ijpam.v86i3.3.

View publication stats