You are on page 1of 6

2017 9th IEEE-GCC Conference and Exhibition (GCCCE)

An Advanced Control Strategy for Magnetic


Levitation Train System Based on an Online
Adaptive PID Controller
Umer Akram, Muhammad Khalid and Saifullah Shafiq
Electrical Engineering Department, King Fahd University of Petroleum & Minerals
Dhahran 31261, Saudi Arabia
Emails: g201512930@kfupm.edu.sa, mkhalid@kfupm.edu.sa, g201410440@kfupm.edu.sa

Abstract—Magnetic levitation train is an unstable system;


state feedback and PID controllers can be used to stabilize the
system. Traditional PID and state feedback controllers are simple
in design and easy to tune but they cannot give satisfactory
performance under larger variations. This paper proposes a new
control strategy based on adaptive PID (APID) for magnetic
levitation train system. The proposed APID is robust against the
parameter variations as it is capable of adjusting its parameters
online according to variations in the system. Mathematical
model of magnetic levitation train system, adaptive PID, state
feedback and PID are derived and implemented in MATLAB.
The proposed APID control strategy is assessed in presence of
small and large disturbances and simulation results depict the
effectiveness of the proposed methodology.

I. I NTRODUCTION
Fig. 1. Magnetic levitation train system [6].
Magnetic levitation (Maglev) is a technique which levitates
an object in air by overcoming the gravitational pull. Elec-
tromagnetic suspension (EMS), Electrodynamic suspension
(EDS) and Hybrid electromagnetic suspension (HEMS) are the with very small error. A Kalman filter based roboust control
commonly used levitation techniques [1]. Maglev techniques scheme is also proposed for magnetic levitation system in [11].
provide contactless, low noise and low friction properties These controller parameters tuned using genetic algorithm
which gives them way to engineering applications, i.e., high gives better performance as compared to conventional PID for
speed trains, wind tunnels, frictionless bearings and levitation the control of magnetic levitation system [12]. A performance
of metal slabs during manufacturing [2]. comparison of PID and Fuzzy controllers when employed to
Maglev train is one of the most promising applications of control the magnetic levitation system is presented in [13]. It
magnetic levitation systems [3]. Maglev train provides fast, is concluded that PID has small steady state error while fuzzy
noise free, safe and low cost journey as compared to conven- has small overshoot.
tional rail system [4], [5]. In EMS based Maglev train cart This paper proposes an Adaptive PID (APID) controller for
is levitated above the track by making use of electromagnetic Maglev train system [14]. In APID parameters are updated
forces, which makes system unstable shown in Fig. 1. Hence online continuously with any variation in the system so that
it is necessary to have an external controller to stabilize the to give satisfactory performance under varying conditions.
system. The guide magnets keeps the rail on track while train Maglev train model and controllers are implemented in the
magnets raise the train above the track. Matlab Simulink. Performance of APID is compared with
State feedback and nonlinear state feedback controller are state feedback and PID controller and it is found that APID
designed to stabilize the system [7] and it is found that non- gives better performance. As state feedback controller shifts
linear state feedback gives better performance. Similarly it is the poles of the system to the desired location in finite time in
suggested that PID gains should be redesigned with variations order to get the desired response. Once the parameters of the
in the system for satisfactory performance [8]. In [9], a model state feedback are set, they will not change and consequently
reference adaptive control based PID is proposed to control it does not give satisfactory performance in case of a large
the magnetic levitation system and due to online tuning of disturbance. Same is true for PID. While in case of APID, all
control parameters controller gives satisfactory performance. control parameters are adaptive and are continuously tuned
In [10], a roboust control scheme is applied to magnetic accordingly with the system variations. Due to continues
levitation system and it is found that it tracks the position update of parameters APID performs well under small as well

978-1-5386-2756-3/17/$31.00 ©2017
Authorized licensed use limited to: IEEE de Talca. Downloaded on April 29,2021 at 00:36:06 UTC from IEEE Xplore. Restrictions apply.
Universidad
2017 9th IEEE-GCC Conference and Exhibition (GCCCE)

TABLE I
dƌĂĐŬ S YSTEM PARAMETERS [16]

C M xo fo g
0.00545 725kg 0.012m 7288.75N 9.8N/kg

dž
Also, the mechanical system is obtained using free body
dƌĂŝŶ diagram as follows
& DĂŐŶĞƚ
mẍ = f + mg − F (2)
where F and L are electromagnetic force and disturbance
ŝ respectively.
ŵŐ The dynamic model of the system can be expressed as
Ĩ

Ƶ ⎨ ẋ = y
f  2
ẏ = m + g − Cm xi (3)

Fig. 2. Magnetic levitation train system model. u = Ri + 2C x i̇ − 2Ci
x 2 ẋ
The system can be reduced to second order by taking current
as large disturbances. This adaptive nature makes the APID as input, which reduces the Maglev train system model to
superior than conventional PID and state feedback controllers. second order [17]. Hence
The air gap between track and train is variable and affected by ⎧
the disturbances, and its range is from 8mm to 10mm with an ⎨ ẋ = y
 2
allowable error as in [15]. Also, it is important to mention that ẏ = mf + g − Cm xi (4)

both external disturbances and model uncertainties are lumped
and considered as disturbance in this study and the proposed Using linearization, we get
APID is capable of achieving the recommended air gap.

The remainder of the paper is structured as follows: Sec- Δẋ = Δy
tion II presents the system modelling. The control system 2Ci0 2Ci0 (5)
Δẏ = mx 3 Δx − mx 2 Δi
design including description of traditional controllers and the 0 0

proposed APID have been described in Section III and its In control theory, a state–space representation is a mathe-
sub sections. Section IV presents the simulation results. The matical model of a physical system as a set of input, output
Section V concludes the paper. and state variables related by first-order differential equations.
Hence the overall state space representation of the system is
II. M ATHEMATICAL M ODELLING    
ẋ1 0 1 x1 0
Electromagnetic suspension system has three sub-systems, = − v (6)
air spring system, track system and electromagnet system. ẋ2 3376 0 x2 2
We can ignore air spring system and also the flexibility of where x1 is Δx, x2 is Δẋ and v is Δi. Note that the values
track is not considered in the system under study. In this of the system parameters are taken from [16] and are given in
study, we consider the electromagnet system only. Simplified Table I.
model of the system under study is shown in the Fig. 2. It
consists of train magnet and track. When electric current i III. C ONTROLLER D ESIGN
flows through the coil, it magnetizes it and this results in a As it is clear from state space model that the system is
force of attraction F between track and train. This force F unstable, a controller is required to stabilize the system. We
is called electromagnetic force and it lifts the train above the give an overview of traditional control approaches and the
track. Weight and disturbance force are in opposite direction proposed control scheme in the following sub–sections.
of electromagnetic force.
Using Kirchhoff’s voltage law (KVL), the corresponding A. Linear State Feedback Controller
electrical system can be written as In general, linear state feedback controller begins with the
linear time–invariant state equation, which represents the open-
d (Li)
u = Ri + i (1) loop system or plant to be controlled. The main focus is
dt on the application of state feedback control laws with the
where R and L are resistance and inductance of the coil; goal of achieving desired performance characteristics for the
note that L is the function of the frequency of applied voltage closed–loop state equation. In fact, the state feedback control
and the separation [16]. law features a constant state feedback gain matrix K and

Authorized licensed use limited to: Universidad de Talca. Downloaded on April 29,2021 at 00:36:06 UTC from IEEE Xplore. Restrictions apply.
2017 9th IEEE-GCC Conference and Exhibition (GCCCE)

W/ ŝƐƚƵƌďĂŶĐĞƐ
ĚĂƉƚŝǀĞ
DĞĐŚĂŶŝƐŵ
ZĞĨĞƌĞŶĐĞ <Ɖ <Ɖ͕<ŝ͕<Ě
н
/ŶƉƵƚ н н KƵƚƉƵƚ
Ğ Ƶ WůĂŶƚ
ɇ <ŝ ɇ ɇ
Ͳ W/ ŝƐƚƵƌďĂŶĐĞƐ
н
<Ě ZĞĨĞƌĞŶĐĞ
/ŶƉƵƚ <Ɖ
н
н Ğ н Ƶ
ɇ <ŝ ɇ ɇ WůĂŶƚ
Ͳ н
<Ě KƵƚƉƵƚ

Fig. 3. PID control based Maglev train system.

Fig. 4. Adaptive PID control based Maglev train system.


a new external reference input particularly having the same
dimension as the open–loop input with the same physical units
(see, e.g., [18] and references therein for further details). the controller robust against system variations. As mentioned
In our case, the control input (see, e.g., [7]) is earlier, state feedback controller and traditional controllers like
PID shift the poles of the system to the desired location in
v = K p x1 + Kd x2 (7) finite time in order to get the desired response. Once the
parameters of the state feedback are set, they remain fixed
Hence the characteristic equation of the system is
throughout the simulation period. While in case of APID,
all control parameters are updated adaptively in accordance
s2 + 2Kd s + 2K p − 3376 (8)
with with the system variations. This adaptive nature makes
The values of Kd and K p can be determined from this the APID superior than conventional PID and state feedback
characteristic equation for any rise time and settling time. For controllers.
example, for Ts = 0.1 sec and 5% percentage overshoot; the Gradient descent algorithm is one of the most popular
parameters are Kd = 40, K p = 2625. algorithms in optimization theory and it is the most common
way to optimize the system’s performance. Gradient descent
B. PID Controller
minimizes an objective function J parameterized by the param-
As mentioned earlier, an overview of traditional control eters of the model by updating them in the opposite direction
approaches is given for comparison purposes. A proportional- of the gradient of the objective function w.r.t. to the system
integral-derivative controller (PID controller) is a well known parameters. We have used gradient decent algorithm for our
control scheme widely used in industrial control systems system optimization; full technical details and optimization
because of its simple design and good performance [19], [20]. procedure of the algorithm is out of scope of this paper,
A PID controller continuously calculates an error value as the however, the interested reader is referred to [21], [22] and
difference between a desired reference and a measured process reference therein for further details.
variable. PID controller can be easily employed to control the The error e between desired and reference output as used
Maglev Train system. Block diagram of PID controller for in gradient descent algorithm is given as
Maglev system is shown in Fig. 3. Output of the system is
compared with the reference and error is generated, this error e = xre f − x (10)
served as input to PID which in turn generates the control
signal in order to minimize the error. with an objective function J as
The governing PID equation can be written as follows 1
J = e2 (11)
de (t) 2
u = K p e (t) + Kd + Ki ∫ e (t) dt (9)
dt In order to derive the governing equations for our proposed
where K p , Kd and Ki denote the coefficients for the propor- APID, consider the following
tional, integral, and derivative terms, and are all non-negative. ∂J
K p (t + 1) = K p (t) − γ (12)
C. Adaptive PID Controller ∂ Kp
APID is a controller in which parameters: K p , Ki and Kd
∂J
are tuned online according to system conditions. In the pro- Kd (t + 1) = Kd (t) − γ (13)
posed controller parameters are updated using gradient descent ∂ Kd
algorithm. Block diagram of Maglev system with APID is
∂J
shown in Fig. 4. Output of the Maglev is measured and fed Ki (t + 1) = Ki (t) − γ (14)
to adaptive system. The system parameters are updated based ∂ Ki
on the deviation in output control, this online tuning makes Using chain rule, we get

Authorized licensed use limited to: Universidad de Talca. Downloaded on April 29,2021 at 00:36:06 UTC from IEEE Xplore. Restrictions apply.
2017 9th IEEE-GCC Conference and Exhibition (GCCCE)

×10 -3 ×10
-4
1 4

0
2

-1

0
-2
Air gap variation (m)

Air gap variation (m)


-2
-3

-4 -4

-5
-6

-6

-8
-7

-8 -10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (sec) Time (sec)

Fig. 5. Air gap variation with state feedback. Fig. 6. Air gap variation with PID.

∂J ∂J ∂x ∂u ×10
-4

= (15) 4

∂ Kp ∂ x ∂ u ∂ Kp
2

∂J ∂J ∂x ∂u
= (16)
∂ Kd ∂ x ∂ u ∂ Kd 0
Air gap variation (m)

∂J ∂J ∂x ∂u -2
= (17)
∂ Ki ∂ x ∂ u ∂ Ki
-4
Finally, parameters update equations are
2  2 -6

K p (t + 1) = K p (t) − γ xre f − x (18)


3376 -8

2 de (t)
Kd (t + 1) = Kd (t) − γ e (t) (19) -10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
3376 dt Time (sec)

2
Ki (t + 1) = Ki (t) − γ
e (t) ∫ e (t) dt (20) Fig. 7. Air gap variation with adaptive PID (APID).
3376
The above equations are the main governing equations for
the proposed adaptive PID controller for Maglev train system.
with APID is shown in Fig. 10. It is clear that in case of
IV. R ESULTS AND D ISCUSSIONS
state feedback and PID, changes in the air gap is more than
Maglev train system and the controllers are implemented in 10mm making the system unstable, while APID successfully
MATLAB and system is subjected to a variety of disturbances stabilizes the system.
with increasing intensity in order to check the performance
As mentioned earlier, it is important to highlight that both
of controllers. The results are then compared using small and
external disturbances and model uncertainties are lumped and
large disturbances and robustness of the controllers is reported.
considered as disturbance. The air gap between track and train
Fig. 5 and Fig. 6 represent the response of the system when
is variable and affected by the disturbance. This air gap range
state feedback and PID are used to stabilize the system. It
is from 8–10mm with an allowable error.
is clear that system can be stabilized using these controllers.
Fig. 7 depicts the response of the system when APID is used. Table II gives a numerical comparison of the controllers per-
It can be observed that change in air gap is minimum in formance for small and large disturbance. It can be observed
case of APID. Consequently, larger disturbance is applied to that the proposed APID performs well with minimum air gap
the system and the response of the system with the given variation for both small and large disturbance as compared
controllers has been recorded. Fig. 8 and Fig. 9 show the with PID and state feedback controller. The proposed APID
response of the system with state feedback and PID subjected is capable of achieving the recommended air gap and is fully
to large disturbance respectively. Response of the system robust against system uncertainties and variations.

Authorized licensed use limited to: Universidad de Talca. Downloaded on April 29,2021 at 00:36:06 UTC from IEEE Xplore. Restrictions apply.
2017 9th IEEE-GCC Conference and Exhibition (GCCCE)

TABLE II
0.05
AIR GAP VARIATION OF THE CONTROLLERS

Disturbance Intensity State Feedback PID APID


0
Small 7.8mm 0.902mm 0.86mm
Large 249.6mm 28.92mm 2.7mm
-0.05
Air gap variation (m)

-0.1
V. C ONCLUSION

-0.15 An adaptive PID controller has been designed and im-


plemented to stabilize a magnetic levitation train system
-0.2 particularly in the presence of a large disturbance. The given
system is highly unstable and many control techniques exist
-0.25
in the literature to stabilize it, e.g., state feedback, traditional
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (sec) PID, adaptive control. These techniques perform well when
system is subjected to small disturbances, however, innovative
Fig. 8. Air gap variation with state feedback controller for large disturbance. procedures and control techniques are always welcomed to
stabilize the system in the presence of large disturbances
and parameter variations. We have implemented the proposed
adaptive PID controller coupled with online tuning of its
0.01
parameters in an innovative way and robustness of overall
0.005 system has been demonstrated. A mathematical model of
magnetic levitation train system has been derived and state
0
feedback and traditional PID control have been implemented.
-0.005
For comparison purposes, the system is subjected to compara-
Air gap variation (m)

tively large disturbances for all controllers, it has been shown


-0.01 that the proposed adaptive PID has shown better performance.
Simulation results prove the effectiveness of the proposed
-0.015
approach.
-0.02

-0.025
R EFERENCES

-0.03
[1] H.-W. Lee, K.-C. Kim, and J. Lee, “Review of maglev train technolo-
0 0.5 1 1.5 2 2.5
Time (sec)
3 3.5 4 4.5 5 gies,” IEEE Transactions on Magnetics, vol. 42, no. 7, pp. 1917–1925,
2006.
[2] N. Al-Muthairi and M. Zribi, “Sliding mode control of a magnetic
Fig. 9. Air gap variation with PID for large disturbance. levitation system,” Mathematical Problems in Engineering, vol. 2004,
no. 2, pp. 93–107, 2004.
[3] M. Ono, S. Koga, and H. Ohtsuki, “Japan’s superconducting maglev
train,” IEEE Instrumentation & Measurement Magazine, vol. 5, no. 1,
-3
pp. 9–15, 2002.
×10
1 [4] H. Nakashima, “The superconducting magnet for the maglev transport
system,” IEEE Transactions on Magnetics, vol. 30, no. 4, pp. 1572–
0.5 1578, 1994.
[5] J. Eastham, M. Balchin, P. Coles, and D. Rodger, “Comparison of
0 short primary linear machines for high speed maglev vehicles,” IEEE
Transactions on Magnetics, vol. 23, no. 5, pp. 2338–2343, 1987.
-0.5 [6] B. Courtney, “Cheshire ”Innovation”,” 2016. [Online]. Available:
Air gap variation (m)

http://www.cheshire-innovation.com/Transport%20internet.htm
-1 [7] H. Liu, X. Zhang, and W. Chang, “PID control to maglev train system,”
in International Conference on Industrial and Information Systems
-1.5 (IIS’09). IEEE, 2009, pp. 341–343.
[8] R.-J. Wai and J.-D. Lee, “Performance comparisons of model-free con-
-2 trol strategies for hybrid magnetic levitation system,” IEE Proceedings–
Electric Power Applications, vol. 152, no. 6, pp. 1556–1564, 2005.
-2.5 [9] B. Singh and V. Kumar, “A real time application of model reference
adaptive pid controller for magnetic levitation system,” in 2015 IEEE
-3 Power, Communication and Information Technology Conference (PC-
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (sec) ITC). IEEE, 2015, pp. 583–588.
[10] Y. Zhang, B. Xian, and S. Ma, “Continuous robust tracking control for
magnetic levitation system with unidirectional input constraint,” IEEE
Fig. 10. Air gap variation with adaptive PID (APID) for large disturbance. Transactions on Industrial Electronics, vol. 62, no. 9, pp. 5971–5980,
2015.

Authorized licensed use limited to: Universidad de Talca. Downloaded on April 29,2021 at 00:36:06 UTC from IEEE Xplore. Restrictions apply.
2017 9th IEEE-GCC Conference and Exhibition (GCCCE)

[11] Z.-J. Yang, Y. Fukushima, S. Kanae, and K. Wada, “Robust non-linear


output-feedback control of a magnetic levitation system by k-filter
approach,” IET Control Theory & Applications, vol. 3, no. 7, pp. 852–
864, 2009.
[12] I. Ahmad, M. Shahzad, and P. Palensky, “Optimal PID control of mag-
netic levitation system using genetic algorithm,” in IEEE International
Energy Conference (ENERGYCON). IEEE, 2014, pp. 1429–1433.
[13] A. El Hajjaji and M. Ouladsine, “Modeling and nonlinear control of
magnetic levitation systems,” IEEE Transactions on Industrial Electron-
ics, vol. 48, no. 4, pp. 831–838, 2001.
[14] H.-W. Lee, K.-C. Kim, and J. Lee, “Review of maglev train technolo-
gies,” IEEE Transactions on Magnetics, vol. 42, no. 7, pp. 1917–1925,
2006.
[15] T.-H. Meen, S. Prior, and A. D. K.-T. Lam, Innovation, Communication
and Engineering. CRC Press, 2013.
[16] Z. Liu, X. Li, and Z. Long, Maglev Trains. Springer, 2015.
[17] D. L. Trumper, S. M. Olson, and P. K. Subrahmanyan, “Linearizing
control of magnetic suspension systems,” IEEE Transactions on Control
Systems Technology, vol. 5, no. 4, pp. 427–438, 1997.
[18] G. C. Goodwin, S. F. Graebe, and M. E. Salgado, Control system design.
Prentice Hall New Jersey, 2001, vol. 240.
[19] L. Reznik, O. Ghanayem, and A. Bourmistrov, “PID plus fuzzy con-
troller structures as a design base for industrial applications,” Engineer-
ing Applications of Artificial Intelligence, vol. 13, no. 4, pp. 419–430,
2000.
[20] K. H. Ang, G. Chong, and Y. Li, “PID control system analysis, design,
and technology,” IEEE Transactions on Control Systems Technology,
vol. 13, no. 4, pp. 559–576, 2005.
[21] P. Baldi, “Gradient descent learning algorithm overview: a general
dynamical systems perspective,” IEEE Transactions on Neural Networks,
vol. 6, no. 1, pp. 182–195, 1995.
[22] S. Boyd and L. Vandenberghe, Convex optimization. Cambridge
university press, 2004.

Authorized licensed use limited to: Universidad de Talca. Downloaded on April 29,2021 at 00:36:06 UTC from IEEE Xplore. Restrictions apply.

You might also like