.

Electromagnetic Levitation System : An Experimental Approach

H. D. Taghirad, M. Abrishamchian, R. Ghabcheloo K. N. Toosi University of Technology Department of Electrical Engineering Tehran, Iran
Abstract | In this paper, we propose a simple method for controlling an electromagnetic levitation (Maglev) system. The model of the Maglev system under consideration in this paper is third order, inherently nonlinear and unstable. For designing a controller to obtain a good disturbance rejection and being insensitive to parameter variations, we use an innerloop and an outer-loop con guration. The conguration is the key to solve this problem. By using a nonlinear state transformation in the feedback path, a cascaded PI controller and a rede ned input in the inner loop, we obtain a linear second order system. Furthermore, to achieve a stable system with a good disturbance rejection, we design a PID controller cascaded with the inner loop system in the negative unity feedback con guration in the outer loop. We show that the closed-loop system is robustly stabilized against the mass variation. Simulation results show that despite actuator saturation limits, the performance of our simple structure controller is comparable to that reported in the literature. In order to assess these results in practice, an experimental setup is designed and being constructed. The simulation analysis is based on the parameters of this experimental setup. Keywords | Magnetic levitation system, Suspension system, Feedback linearization, PID, Nonlinear system, Robustness.
I.

is being used in many other applications such as magnetic suspension high speed trains 2]. The electromagnetic suspension systems (EMS systems) can be divided into two types see 3] Repulsion type In this type of the electromagnetic suspension system, suspended part(such as train) repulses the x part(such as rail). This type is realizable with materials of permeability ( r ) less than unity (diamagnetic material) and with superconductors with ( r = 0). A disadvantage of this type of suspension systems is that an auxiliary wheeling suspension is needed for operation below the critical speed (typically 80Km=h ) specially at zero speed, i.e., when the suspended object is stationary. These systems are naturally stable with small damping ratio. Attraction type In this type suspended part(such as train) attracts the x part(such as rail). Realization of this type is possible by ferromagnetic and/or permanent magnet. This type of EMS system has one signi cant advantage in that it provides attraction force at zero speed, but such a system is inherently unstable. The increasing use of the Maglev technology in its various forms has resulted in active research in this area. The electromagnetic suspension technology has already been applied to magnetically levitated carriers 4], magnetic bearings 5], and vibration isolators 6]. Hence, some advanced theories are utilized to get better performance, such as, H1 control 7], -synthesis 8] and feedback linearizing control law 9]. In this paper we will summarize the mathematical model of the system that is taken from 3], which is elaborated in Section II. In Section III, we design a controller that achieve our objectives.

One of the ways that non-contact surfaces are maintained is via magnetic suspension. This system is commonly referred to as Magnetically Levitated system (Maglev) which has been used in the vehicle suspension system and magnetic bearing system by B. A. Holes of university of Virginia in 1937 for the rst time in 1954, this system was utilized by Laurencean and Tournier at ORENA in France for the purpose of aerodynamic testing in wind tunnels 1]. Furthermore, the system

Introduction

In this controller, we consider a state transformation followed by a PI controller in the inner loop that satis es the necessary performance characteristics for the actuator. Next we design a PID controller that stabilizes the whole system and discuss the robustness of the closed-loop system against mass variations. The foundation of the implementation including some details of mathematical design and data acquisition system is the subject of Section IV. In Section V, some issues that required to be addressed for proper simulation of the model are discussed and also simulation results are presented. In Section VI, we provide a comparison between the performance of the proposed controller and a feedback linearizing con1. Schematic of an electromagnetic suspension troller for stabilizing and commanding the EMS Fig. system system which has been proposed in 9]. Finally conclusions are summarized. Thus if the total resistance of the circuit is denoted by R, then the instaneous voltage v(t) II. Mathematical Model winding) controlling the exTo synthesize a feedback control system, a rel- (across the magnet is given by atively precise model of the plant is required. It citation current i(t) 2 2 is known that a mathematical model cannot fully 0 N Ai(t) dz (t) 0 N A di(t) express the behavior of the real physical plant. An v(t) = Ri(t) + 2z (t) dt ; 2z (t)2 dt : ideal mathematical model has various uncertain(3) ties such as parameter identi cation errors, un- Note a varying inductance with respect to z (t) modeled dynamics and neglected nonlinearities. in the second term, and that the third term deThe controller is required to have robust stability notes a voltage which varies with changes in the and performance in presence of such uncertainties air gap z (t) and its rate of change similar to back in the model and also due to bounded disturbance. EMF voltage introduced in a DC motor. This exCrude simulation studies do not reveal all the pression displays the actuator dynamic (electrical practical pitfalls of a real system. Therefore, more plant). realistic mathematical model which includes all We now give the vertical dynamic of the system di erent aspects of the system behavior is needed by the following equation, to give us a better knowledge in constructing a mz (t) = ;F (i z) + mg + fd (t) (4) Maglev system. We use the model that is given by 3]. In the model as shown in Figure 1, the instan- where fd (t) is disturbance force input and m is the taneous ux linkage between the two magnetized suspended object mass. This equation describes bodies through the air gap z (t) is m . If the leak- the dynamics of the plant (mechanical plant). The foregoing EMS system can be rewritten in the folage ux is ignored, that is t = m , then lowing equations: 2A 0N L(z) = 2z (t) (1) x1 = x2 _ (5) 2 1 (6) x2 = ; 2k x3 + g + m fd _ where 0 is vacuum permeability, N is the number m x1 1 of turns of the magnet winding, and A is pole 2 x3 = ; R x1 x3 + xxx3 + k x1 u(t) (7) _ face area. Furthermore, note that the inductance k 1 L varies with changes in the air gap. Also the y = x1 (8) attraction force at any instant of time is given by, where y is output, x1 = z (vertical air gap) m], 2 A i(t) 2 _ velocity) i F (i z) = 0 N z (t) (2) x2 = z (vertical relative = 9:8 m=m/s], x3 k== (magnet current) A], g s2 ] and 4

(force factor). Equations (1), (2) and (3) are valid when the magnetized core is not working in saturation region and also when the air gap clearance, z (t), is not too small. Therefore, state-space Equations (5), (6), (7) and (8) for an EMS system are valid in these conditions. We have considered them in our experimental setup design. To concern the limited permability and leakage, we should add a bias to x1 (in constructed set up is about 0:2mm).
0 2

N A
2

the Equation (12), we have

x + ax = Kp e + Ki edt _
0

Zt

(14)

After some manipulations, for a step input xd , we achieve the following dynamics for error,

e + (Kp + a)e + (Ki + a)e = xd a: _ _ _

(15)

and for actuator dynamics, we have x = ; R x x + 1 u(t) _

By considering the following Lyapunov function and its derivative, we nd conditions for stability: III. Control System Design 1 1 _ Electromagnetic suspension system is essen- V (x t) = 2 (e + a(t)e)2 + 2 (Ki + a(t)Kp )2 tially an unstable system due to inverse force_ V = ;Kpe2 ; (Ki a ; 1 a)e2 _ (16) distance characteristic, see Equation (2). From 2_ Equations (3) and (4), it is obvious that the system dynamics consists of two parts, actua- The dynamics (15) is robustly stable and the ertor dynamic (electrical part), and plant dynam- ror goes to_ zero if following two conditions are ics (mechanical part). These two subsystem are satis ed: V a<) 0 or equivalently Ki , Kp > 0, _ (t K _ strongly coupled because distance of suspended K > maxt 2a(t) and limt!1 a = 0. Indeed by choosing Ki high enough, we satisfy rst condition object from the magnet z (t) in uences the magand the second condition is satis ed if the whole net inductance, see Equation (1). We can reduce the e ects of two parts by a simple closed-loop is stabilized. By choosing high gains state transformation of x = x3 =x1 . Then the new Ki and Kp such that the actuator has large bandwidth, the actuator dynamics can be neglected. state equation for the plant dynamics become Hence x can be considered as the input for plant x1 = x 2 _ (9) dynamics. Considering x2 as the input to the dynamics (9) 1 x2 = ; 2k x2 + g + m fd _ (10) plantwith u = x2and rewriting Equations to and (10) , the model is simpli ed m
i p

We now consider the control synthesis for the actuator system. Let's view the actuator dynamic di erently. Rewrite Equation (11) for input u and output x and consider the term ; R x1 as an unk certainty. Here, x + a(t)x = 1 u _ (12)

(11)

where a(t) = R x1 represents the uncertainty. k Since x1 is a mechanical variable, it varies much slower than an electrical variable x, provided the closed-loop actuator has high-bandwidth. We propose that by choosing a high gain PI controller, we can achieve a high bandwidth actuator inspite of the variation of a(t). Consider, IV. Experimental Setup Zt (13) A. Experimental Con guration u = koKp e + koKi edt 0 In order to examine the proposed control law where e = xd ; x, xd is desired value of the x, and in practice, and to obtain a more realistic estiko is a given value of k. By substitution of u in mate of the practical challenges of EMS system,

We now consider designing a controller such that the closed-loop system achieve the following objectives: 1. Robust stability of the closed-loop system against mass variation. 2. Good disturbance attenuation performance. 3. Non-negative control action, u. We can achieve all these objectives with a PID controller. The integral term makes the system a type-1 system to assure zero steady-state error. The proportional and derivative gains are selected such that to guarantee the robust stability of the system. The whole closed-loop system is shown in Figure 5 in Section V.

1 x1 = ; 2k u + g + m fd: m

(17)

80 60 40 20

Imag Axis

0 20 40 60 80 80 60 40 20 0 Real Axis 20 40 60 80

Fig. 4. Root locus of the mechanical plant Fig. 2. A scanned photo of the constructed setup

an experimental setup of this system is designed and being constructed. Figure 3 illustrates the block diagram of the constructed system. In this gure it is shown that to provide more exibility and simplicity to the hardware, all the blocks in the dashed-line box are implemented through software. In order to do that, the setup is equipped with an IBM-PC (Pentium 133 MHz) and a National Instruments data acquisition and conditioning card (NIDAQ: AT-MIO-16X-50). Figure 2 illustrates a photo of the experimental setup. The mechanical setup consists of a U-shape magnet and a rail-shape track to simulate the inherent application of magnetic suspension trains. We choose a laminated and U-shape magnet to reduce the eddy current and hysteresis e ects 3]. The suspended object motion is concentrated by a long exible beam to be free only in one direction 10]. By some corrections in the parameters, we can assume that the track travels approximately only in the vertical direction. The total accessible range of the travel is approximately between 0:5 to 5mm. In order to measure the air gap clearance, we tried sensors which are commercially available. However their linearity and range of application

dose not match our requirements. Therefore, we have designed and have constructed a custom made capacitive transducer for this purpose. The sensor consists of a rectangular plate surrounded by a thin border as a guard to reduce the electromagnetic noise e ect. The capacitance is demodulated to a voltage directly proportional to air gap that is approximately linear in the accessible range. The current is measured using a small resistor cascaded with the electromagnetic coil. The power source is designed using a PWM type voltage chopper with DC link voltage of 24 volts, which are considered in the simulations as to be described in Section V. All inputs and output are decoupled from NIDAQ card by optocouplers. The setup components are individually tested and calibrated and open-loop experiments are conducted to verify the integrability of the whole setup. Closed-loop experiments are underway.
B. Derivation of Model and Controller After the construction of the setup, the nominal parameters of the experimental setup are estimated and given in Table 1 (All parameters in the table and in this section are in SI units.) For these parameters and nominal air gap, zo = 2:5mm, we can nd ao from Equation (12) as ao = 75. To achieve a high bandwidth closed-loop system with !n = 500 and = 1, we choose PI controller as Kp = 1000, Ki = 250000. And also for the PID controller considering mass uncertainty we choose controller gains as follows to assure robust stability, m 0 Kp = 12 103 2k , m Ki0 = 100 103 2k and m 0 Kd = 0:21 103 2k . where mo and ko denote nominal values of m and k, respectively. By this choice the closed-loop plant dynamics has its roots at ;10, ;100 and ;100 in the nominal condition. Let's choose the m root locus gain as km = 2k 2k , it means in m
o o o o o o o o

Fig. 3. Block diagram of the experimental setup

the nominal condition we have km = 1. The root locus of the closed-loop plant dynamics is shown with respect to variation of the factor km in Figure 4. The whole system is stable for large values of km and also for small values down to km = 3:46 10;2 . Therefore, closed-loop system is very robust against mass and force factor variation. This fact is veri ed in the following section via simulation.
o

4. PID controller can only be implemented approximately in practice. The e ect of this approximation considered by using approximate PID controller box of simulink. The e ects of variations in mass and also disturbance attenuation of closed-loop system are shown in corresponding gures as explained in the following section.
VI.

V.

For simulation, we use simulink toolbox of MATLAB as shown in Figure 5. Since this simulation is done to check the feasibility of the controller in practice, we consider the following points in the simulation: 1. We use the square root of the absolute value of the output to avoid the failure of simulation because of negative values of the controller output. But, we note that the state x in the actuator dynamics would not be physically meaningful to allow negative values. 2. We model hard stops in mechanical setup as very sti springs to avoid the discontinuity in the equation of motion of the plant and the numerical di culties of integrating them. In this method when the track is passing through the limits as

Simulation Results

A nonlinear feedback linearizing controller is reported in 9]. Behaviors of our controller and the feedback linearizing controller are compared in Figures 6 and 7. In these gures, dash-dot curves correspond to the feedback linearizing control and our proposed control is shown by solid curves. Below each gure, the experiment conditions including transient behavior and disturbance rejection of the nominal and disturbed plants are mentioned. Figure 6 shows the transient response of the system, (a) with nominal mass m = 3:3, and (b) with m = 16:5. One can see that the response of the feedback linearizing controller is a bit better than ours in case of nominal mass (Figure 6-a), but has steady-state error in perturbed conditions (Figure 6-b), while the response of the proposed controller approaches the desired value. Their behaviors in response to the disturbance force is deif x > xsat picted in Figures 7. Our proposed controller has Fsat = k(x ; xsat) larger overshoot but the output reaches the de.. .. sired value. We note that the feedback lineariz. . 2 ing control reported in 9] needs an extra sensor 1 x2 = ; 2k x3 + g + m (fd + Fsat ) to measure the rate of the air gap variations, but _ m x1 performance is not much better than our proposed 3. One of important practical limits of the con- controller. troller implementation is the bounded control e ort we can provide in practice. To visuVII. Conclusions alize this limit, a simulation block is added This paper presents a new method for controlwhich illustrates the required voltages and ling the electromagnetic suspension system as a 1their limit of 0 ; 24 volts. DOF system. By a state transformation the complex nonlinearities and the coupling between the plant and the actuator are reduced, each subdynamics are controlled separately as reported for DC motor control 11]. With this di erence that in the motor control, the inner loop path feeds back just current, while because of dynamic of the EMS system, we choose current divided by the air gap as the feedback. PI and PID controllers are designed to satisfy our objectives, and their performance is compared to some methods reported Fig. 5. Closed-loop system in simulink in the literature. It is concluded that, in spite of
Stats ref + e PID f(u) + sum1 PID sqrt root sum2 Mux Maglev 1e3s+250e3 s PI v 1 s Saturation Integrator Mux plant x Fd u[3]/u[1] state transformation

Comparison with Feedback Linearizing Controller

The nom inal parameters of experimental setup

TABLE I

Description Notation Value Track mass m 3.3 Number of turns N 565 Pole face area A 78:6 10;5 Total resistance R 1.5 Force factor k 5 10;5 Nominal gap zo 2:5 10;3 simple shape of controller, good performances are achievable.
References

1] E. E. Covert, \Magnetic Suspension and Balance Systems," IEEE AES Magazine, pp. 14{22, May 1988. 2] A. R. Eastham and W. F. Hayes, \Maglev Systems Development Status," IEEE AES Magazine, pp. 21{ 30, January 1988. 3] P. K. Sinha, Electromagnetic Suspension, Dynamics and Control. IEE control engineering series, 1987. 4] M. Morishita et al, \A New Maglev System for Magnetically Levitated Carrier System," IEEE Transactions on Vehicular Technology, vol. 38, pp. 230{236, November 1989. 5] A. M. Mohamed and I. Busch-Vishniac, \Imbalance Compensation and Automation Balancing in Magnetic Bearing Systems Using the Q-parameterization Theory," IEEE Transactions on Control System Technology, vol. 3, pp. 202{211, June 1995. 6] K. Watanabe et al, \Combination of H1 and PI Control for an Electromagnetically Levitated Vibration Isolation System," in Proceedings of the 35st Conference on Decision and Control, pp. 1223{1228, 1996. 7] F. Zhang and K. Suyama, \H1 Control of Magnetic Suspension System," in Proceedings of the 33st Conference on Decision and Control, pp. 605{610, 1994. 8] M. Fujita et al, \ -Synthesis of an Electromagnetic Suspension System," IEEE Transactions on Automatic Control, vol. 40, pp. 530{536, March 1995. 9] S. Joo and J. M. Seo, \Design and Analysis of the Nonlinear Feedback Linearizing Control for an Electromagnetic Suspension System," IEEE Transactions on Control System Technology, vol. 5, pp. 135{144, January 1997. 10] D. L. Trumper, S. M. Olson, and P. K. Subrahmanyan, \Linearizing Control of Magnetic Suspension Systems," IEEE Transactions on Control System Technology, vol. 5, pp. 427{438, July 1997. 11] P. R. Be'langer, Control Engineering : A Modern Approch. Saunders College, 1995.

6 5 Output(mm)
Output(mm) 3.5

4 3 2 1 0 0

2.5

50 time(msec)

100

150

450

500

550

600

650 time(msec)

700

750

800

850

25 Control action(V)
Control action(V)

16 14 12 10 8 6 4 450 500 550 600 650 time(msec) 700 750 800 850

20 15 10 5 0 0 50 time(msec) 100

150

(a)

(a)

5.5 5 Output(mm) 4.5 4 3.5 3 2.5 0 50 100 150 200 250 300 time(msec) 350 400 450 500
Output(mm)

4.5 4 3.5 3 2.5 400

500

600

700 time(msec)

800

900

1000

25 Control action(V) 20 15 10 5 0
Control action(V)

20 18 16 14 12 10 8 400 500 600 700 time(msec) 800 900 1000

50

100

150

200

250 300 time(msec)

350

400

450

500

Fig. 6. Transient response in traveling from 5mm to Fig. 7. Response to disturbance force of fd = 50 N of (a) the nominal system (m = 3:3 Kg) and 2.5mm of (a) the nominal system (m = 3:3 Kg) (b) perturbed system (m = 16:5 Kg). Solid: proand (b) perturbed system (m = 16:5 Kg). Solid: posed controller, dash-dot: feedback linearizing proposed controller, dash-dot: feedback linearizing controller controller

(b)

(b)