Sie sind auf Seite 1von 8

618

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 4, JULY 2002

Exact Linearization of a Voltage-Controlled 3-Pole Active Magnetic Bearing System


Chan-Tang Hsu and Shyh-Leh Chen

AbstractMost magnetic radial bearings adopt the configuration of eight magnetic poles. Such a configuration usually requires four power amplifiers and has high power loss, resulting in high cost. In this study, a 3-pole active magnetic bearing (AMB) is investigated, which requires only two power amplifiers. In addition, it has lower power loss. Thus, the overall cost can be reduced. However, the 3-pole AMB system is strongly nonlinear due to magnetic flux coupling. If the actuator dynamics are also considered, the overall system will be much more complicated. Nonlinear control is thus necessary. By taking magnetic fluxes as part of the system states, the state equations of the voltage-controlled 3-pole AMB system admit a quadratic nonlinear form. It is shown that this quadratic nonlinear system is feedback linearizable. An admissible domain for the initial states such that the resulting system responses will not hit the bearing boundary (or back-up bearing, if exists) is also obtained via a Lyapunov analysis. Numerical simulations are carried out to verify the theoretical results. Index Terms3-pole, active magnetic bearing, exact linearization, magnetic flux coupling, voltage control.

, , ,

NOMENCLATURE , Pole face area. Closed-loop system matrix for nondimensional linear system. Decoupled linear system matrix. Flux density. Constant . Set . Domain . Domain . Set and . Magnetic force at each pole. Number of coil turns. Positive definite matrices satisfying Lyaand . punov function with Reluctance of the air gap between the rotor and each pole. Admissible set. Coordinate transformation. Decoupled linear system input matrix. Magnetic force in the and direction.

Gravitational acceleration. Coil current generated by input voltage , . Linear state feedback gains. Nominal air gap between the rotor and poles. Air gap between the rotor and each pole. Rotor mass. Resistances of electrical circuit. Time. System inputs. Input voltages. Upper bound on rotor velocity. System inputs for the feedback linearized system. System state. The position of rotor center. System state for the feedback linearized system. Bias flux. Magnetic flux related quantities. Decoupling matrix. System nonlinearities. Magnetic flux passing through each pole. Minimum eigenvalue of and . Minimum of . Magnetic permeability of the air. Nondimensional time. Rotors nominal rotating speed. I. INTRODUCTION

Manuscript received September 11, 2000; revised April 12, 2001. Manuscript received in final form March 15, 2002. Recommended by Associate Editor D. Dawson. This work was supported in part by the National Science Council of Taiwan, R.O.C., under Grant NSC 89-2212-E-194-011. C.-T. Hsu was with the Department of Mechanical Engineering, National Chung Cheng University, Chia-Yi 621, Taiwan, R.O.C. He is now with the Information and Communication Research Division, Chung-Shan Institute of Science and Technology, Tsoying, Kaohsiung 813, Taiwan, R.O.C. (e-mail: hctcsist@ms58.hinet.net). S.-L. Chen is with the Department of Mechanical Engineering, National Chung Cheng University, Chia-Yi 621, Taiwan. Publisher Item Identifier S 1063-6536(02)05352-6.

IGH speed and high accuracy are two of the major trends for many systems in the 21st century, such as machine tools and semiconductor equipments. For the purpose of high speed, it is well known that active magnetic bearing (AMB) is an inevitable substitute for conventional bearings. The noncontact nature of AMB brings up many advantages over the conventional bearing, including energy efficiency, low wearing, long life span, and absence of lubrication. Hence, AMBs are especially essential for high-speed rotating devices. To the aim of high accuracy, controllers are essential to the inherently unstable AMB systems. For easy controller design, it is desired that the AMB system possess simple mathematical model, which depends on its configuration. The most common configuration for magnetic radial bearings is the 8-pole type, i.e., four pole pairs, as shown in Fig. 1. It is reasonable to assume that each pole pair forms an independent magnetic circuit and hence produces independent magnetic force. With differential driving and large bias currents, the 8-pole AMB can be

1063-6536/02$17.00 2002 IEEE

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 4, JULY 2002

619

Fig. 1. Typical magnetic flux in the 8-pole AMB.

Fig. 2.

The voltage-controlled 3-pole AMB system.

well approximated by a linear model with uncoupled magnetic forces. Therefore, many linear controllers have been proposed [1][9]. The major disadvantage of the 8-pole AMB is the expensive cost and high power loss. The 3-pole AMBs were found to possess several advantages over the 8-pole system recently [10]. For example, only two power amplifiers are needed, as opposed to the four power amplifiers required by the 8-pole AMB. Also, the remagnetization frequency for the 3-pole AMB is lower than that for the 8-pole AMB, resulting in lower iron loss. Moreover, with the same bearing size supporting the same rotor weight, the 3-pole AMB possesses lower steady-state copper loss. All of these will contribute to the low cost of the 3-pole AMB. For detailed discussions and analysis of 3-pole AMBs, please refer to [10]. While the 3-pole AMB has been proposed for many years [11], it received little attention due to the magnetic coupling between poles, as shown in Fig. 2. Consequently, the 3-pole AMB system is strongly nonlinear. Nonlinear controllers are thus necessary for high accuracy. When computers were slow and control theory was not well developed, it was difficult to realize nonlinear controllers necessary for the 3-pole configuration. With todays PC/DSP-based technology and nonlinear control theory, it is believed that the 3-pole system is now feasible. The objective of this study is to show that the voltage-controlled 3-pole AMB system is feedback linearizable. As a result, nonlinear controllers can be obtained easily. Feedback linearization has been extensively used in the controller design of 8-pole AMB systems [12][14] and magnetic

suspension systems [15][18].1 It requires that the systems state equation be expressible in the affine form, i.e., the AMB system is linear in the control inputs. Unfortunately, the AMB system is nonlinear with respect to its control inputs in the current-controlled mode. However, several studies [13], [15], [18], [19] have shown that the 8-pole system is still feedback linearizable provided that some conditions are satisfied. First, the magnetic forces must be uncoupled. In other words, the rotors displacement in -direction is controlled by and only by the magnetic poles in -direction. Second, each direction is controlled by two independent coils. The control current is switched between them. Take the horizontal direction as an example. When the right coil is provided with control current and the left coil is off, the rotor will be pulled to the right and vice versa. While the 8-pole AMB system is linearizable under such assumptions, the required drivers will be doubled and the total cost will increase considerably. Although the current-controlled 3-pole AMB does not satisfy the two conditions, it is found to be also feedback linearizable due to its special structure in the magnetic force model [20]. For the voltage-controlled mode, the AMB system is in the affine form. It can be shown that the voltage-controlled 8-pole AMB system is feedback linearizable if the uncoupled force model is assumed [12], [17], [19]. In this paper, it will be shown that the voltage-controlled 3-pole AMB system is also feedback linearizable even with strong magnetic coupling effect. To be of practical use, feedback linearization approach must be accompanied with some robust control method (e.g., sliding mode control [21], [22]) since uncertainties always exist in reality. Note that with voltage control, the uncertainties in general do not satisfy the matching condition [21] and one has to resort to the backstepping technique [19]. As the first step in the backstepping procedure, coil currents are taken as the virtual control inputs, leading to the current-controlled mode. Therefore, a complete robust controller for the AMB system should integrate the current-controlled and voltage-controlled modes. Such a robust controller can be obtained based on the results in this study. We shall not pursue the robustness issue here, but leave it as a future work. This paper is organized as follows. The 3-pole AMB system is described in Section II. Section III is devoted to designing a nonlinear controller for the 3-pole AMB system by feedback linearization approach. Simulations and discussions are presented in Section IV. Finally, conclusions are given in Section V. II. SYSTEM DESCRIPTION AND MATHEMATICAL MODEL Fig. 2 shows the voltage-controlled 3-pole AMB system considered in this study. It is a single AMB supporting a disk-like rotor mass. The three poles are arranged in a radially symmetric Y structure to produce a uniform force distribution in the two-dimensional configuration space. For easy reference, each magnetic pole is labeled with a number. The differential winding scheme is employed to poles #1 and #2. In other words, they share the same coil current but with different directions. The third pole is wound with another independent coil. The two independent coil currents, and , are generated by two input
1Under the uncoupled force model, the 8-pole AMB can be regarded as the composition of two suspension systems.

620

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 4, JULY 2002

Fig. 3. The magnetic circuit for the 3-pole AMB in Fig. 2.

voltages, and . It is assumed that the gravitational field exists in the negative -direction. At the steady state, a nonzero coil current for is required to support the rotor weight, as shown in Fig. 2. In other words, there exists a bias current for , whereas no bias current is needed for . The proposed pole arrangement, winding scheme and bias currents will result in minimum steady-state copper loss. Furthermore, only two power amplifiers are required under such a configuration. Please refer to [10] for more details. To obtain the mathematical model for the 3-pole AMB system, the magnetic forces produced by three magnetic poles are first derived by a magnetic circuit analysis. The force models are then incorporated into the rotor and actuator dynamics to yield the mathematical model for the overall system. Finally, a proper choice of system states will lead to a set of quadratic nonlinear state equations. A. Magnetic Circuit Analysis With the configuration of Fig. 2, the corresponding magnetic circuit is given in Fig. 3, assuming that the reluctances exist only on the air gaps. By simple circuit analysis, the magnetic flux passing through each pole can be obtained as (1) (2) (3) , are the where is the number of coil turns and , reluctances of the air gaps between the rotor and the three poles. The reluctances can be expressed as (4) where is the magnetic permeability of the air, is the pole face area and are the air gaps and can be expressed as (5) (6) (7) where is nominal air gap and ( , ) is the position of the rotor center. Fig. 4 is the allowable region that the rotor center can reach, which clearly demonstrates the geometric relationship between and ( , ). Note that air gap is the distance

Fig. 4. Geometric relationship between l and (x , y ).

between two surfaces, those of rotor and pole face. The relations in (5)(7) can be easily obtained by mapping the position vector of the center ( , ) onto the three pole directions. The magnetic circuit analysis reveals that (8) It is in fact a consequence of the conservation of magnetic fluxes across a closed surface, which is the rotors surface here. This constraint holds no matter what the winding scheme is. As a result, only two independent magnetic fluxes can be generated even if there may be three independent coil currents. Hence two independent coil currents are sufficient to generate the desired magnetic fluxes. B. Rotor and Actuator Dynamics Based on the magnetic circuit analysis, the magnetic force of the 3-pole AMB can be easily obtained. By Amperes law and principle of virtual work [11], the attractive magnetic force is proportional to the square of magnetic flux on each pole passing through that pole, i.e.,

where is magnetic flux density on that pole. Note here that a linear magnetic characteristics (i.e., linear B-H relationship) is assumed and the effects of fringing and flux leakage are neglected. Expressed in Cartesian coordinates, the magnetic forces are

Hence, the rotor dynamics can be described by (9) (10) where is the rotor mass and is the gravitational acceleration. On the other hand, the dynamics of the actuators driving the three coils can be described by (11) (12)

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 4, JULY 2002

621

where and are the resistances of the two electrical circuits in Fig. 2. The second terms on the right-hand side of (11) and (12) are induced voltage. C. State Equations In considering the actuator dynamics, a common practice is to take control currents as the state variables. It is obvious from (1)(7) and (9)(12) that this will lead to complicated state equations. To obtain a simpler mathematical model, rather than the control currents, but the magnetic flux related quantities (13) (14) are taken as the state variables for the actuator dynamics. Note and have the unit of magnetic field intensity. Inthat serting (1)(7) into (13) and (14) gives (15) is always positive due to physical where . Furthermore, the determinant of the constraint , which is always nonzero. Therefore, matrix in (15) is there is a one-to-one correspondence between , and , . The coil currents and can thus be expressed in terms of and by (16) and , the open-loop dynamics of the proIn terms of posed 3-pole AMB system described by (9)(12) possesses two equilibrium points

, are taken as system inputs so that the origin will correspond to the steady state. The resulting state equations in matrix form are (17) where , and

where note that all nonlinear terms are quadratic. If some other quantities, e.g., the coil currents, instead of the magnetic fluxes, were taken as system states, the resulting state equations would be much more complicated. The other desired steady state with can be studied in the same way. We shall not repeat it here. III. FEEDBACK LINEARIZATION APPROACH CONTROLLER DESIGN
TO

where and . Unfortunately, they in practice. In fact, are physically nonexistent since it is desirable that the rotor be kept at the bearing center, i.e., at the steady state. Moreover, (9) and (10) imply that to maintain the rotor at the bearing center, and must be . Thus the desired steady state should read

The system given by (17) is already in the affine form. This affine nonlinear system is in fact input-state linearizable. Theorem 1: The system (17) is input-state linearizable on the . domain has a constant rank 2. By [23, Proof: Note that Lemma 5.2.2] and [23, Lemma 5.2.3], it suffices to prove that , , such that the overall there exist two outputs system possesses a total relative degree equal to the number of and be the system outputs. Then, states. Let taking successive time-derivatives of each output yields

where

and

To this aim, the steady-state input voltages are required to be

by (11) and (12). In the above, the superscript denotes the steady-state quantities. Although there are two desired steady states, they are physically equivalent. The only difference in . By (16), they have the same magniboth steady states is tude but opposite direction in coil current and . Physically they represent nothing more than opposite winding schemes. To consider the stabilization of the steady state with , , , , , and are taken as state variables and

and the decoupling matrix

is given by

whose determinant is

622

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 4, JULY 2002

which is nonzero on . Therefore, the system described by . (17) has a well-defined vector relative degree {3, 3} on Since the total relative degree equals the number of states, i.e., , the system (17) is exactly linearizable. The above theorem indicates that the system (17) can be recast into a linear time-invariant system by a coordinate transformation and a state feedback. The required coordinate transformation for linearization can be defined by
Fig. 5. The transformation between z and x.

(18)

, demonstrating that under is the image of

is one-to-one. Furthermore,

Furthermore, the state feedback (19) are new inputs, can be used to cancel where the nonlinearities. Upon applying (18) and (19), the system (17) becomes two decoupled linear systems (20) (21) where , and

where and . Please refer to Fig. 5. That is a mapping from onto . Moreover, condition 2) is, is given is also satisfied since the Jacobian of the mapping by

Note that is controllable. It remains to verify that the coordinate transformation defined by (18) is a diffeomorphism . A diffeomorphism is a continuously differentiable mapon . Such ping with a continuously differentiable inverse on a transformation will preserve the qualitative behavior of the system. Lemma: The coordinate transformation defined by (18) is a . diffeomorphism on Proof: A diffeomorphism satisfies two conditions [21, p. 479] 1) it is one-to-one and onto; 2) its Jacobian matrix is nonsingular. and , where To see 1), let and Suppose that . It follows immediately that

which is nonzero and this completes the proof. With the exactly linearized system (20) and (21), a stabilizing linear state feedback controller will be designed. Supposedly, the closed-loop system is globally asymptotically stable. That is, for every initial condition where the coordinate transformation is defined, the system states will approach the origin at the steady state. Physically, the rotor can be brought to the bearing center from anywhere. However, the practical constraint that the rotor can not hit the stator (or backup bearing, if exists) at any time will limit the admissible initial conditions. In other words, remain inside the set it is desired that the system states

and (22) (23) Multiplying (23) by ( ) and imposing (22) gives

where recall that is the nominal air gap. Thus, the with transient response out of are not initial states in allowed. The problem of the controller design can now be restated as and and to determine an admissible follows. It is to design set for initial conditions such that , , i.e., the origin is an asymp1) totically stable equilibrium point and is contained in the domain of attraction of the origin; , , i.e., the rotor will be confined in 2) the AMBs configuration space; is a diffeomorphism on , i.e., the coordinate trans3) is valid on . formation To these goals (in particular, to determine ), the linearized systems (20) and (21) are first nondimensionalized. Let

which implies that are positive on . Then,

since both and by (22). Thus, we have

where is a positive constant representing an upper bound on the rotor velocity. The objective of asymptotic stability can be

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 4, JULY 2002

623

easily achieved by a linear state feedback using the method of eigenvalue assignment. In other words, one can design (24) so that the closed-loop systems are (25)

(26) , , where and both and are Hurwitz. Equations (25) and (26) represent, respectively, the closed-loop horizontal and vertical dynamics of the rotor. The desired performance of the closed-loop dynamics can be . obtained by designing proper feedback gains To determine the admissible set , note that the requirement . Thus, a positively invariant 2) and 3) imply that will be eligible for an admissible set. set contained in , , and be 3 3 positive definite To this aim, let , matrices satisfying (27) (28) be the minimum eigenvalue of Let and . Theorem 2: Let and be that of

Fig. 6. State trajectories projecting on (a) x x plane and on (b) x x plane, where solid: IC , dash-dot: IC , dotted: the boundary of admissible set S.

require more control effort. Moreover, the closed-loop perfor, which are also mance and depend on the feedback gains related to the control effort. IV. SIMULATION RESULTS Simulations are carried out to verify the stability and performance of the proposed 3-pole AMB system under feedback linearizing control. The system used for simulation is (17) with control inputs given by (19) and (24). The system parameters kg, m s , m, are chosen as: H/m, m , , and . These parameters are taken from the experimental system of a current project. Assume that the rotors r/min. Then a reasonable nominal rotating speed is , resulting in upper bound on the rotor velocity is m/s. For simplicity, the control parameters are , and so that taken as: . By pole placement, the eigenvalues of both and are placed at [ 0.5 0.75 1], leading to , and . With the above parameters, and can be determined by (27) and (28) with , where is the 3 3 positive identity matrix. and are thus identical and possess a minimum eigenvalue of 0.2995. Hence an admissible set can be obtained from . In order to verify the system stability and performance, four sets of initial states are taken for simulations. The first two are on the boundary of and the other two are near the stator boundary, as shown in the equation at the bottom of the next page. Fig. 6 shows the state trajectories with the fist two initial states. In the figure, the boundary of the admissible set is marked by dots. Fig. 6(a) is the trajectories projecting on the

and . Then, the set is an admissible set satisfying the requirements

1), 2) and 3). Proof: In coordinates, the closed-loop system is linear is in the domain of attraction of time-invariant and hence . Also, if , then the transformation is diffeomorphic on and hence is in the domain of attraction . Indeed, the RayleighRitz inequality implies that of

and one obtains (29) One implication of (29) is that and hence that , i.e., , implying . Another implication is that

since and . Thus, . Finally, since is a Lyapunov function for the closed-loop system, Lyapunov is positively theory asserts that the set in the form of invariant. Therefore, is positively invariant in the -domain and is positively invariant in the -domain. We conclude that is a positive invariant set contained in the domain . of attraction of Remark: The inequality (29) also implies that is indeed an upper bound on the rotor velocity ( and , or and ). can give a larger admissible set , but it will Setting larger

624

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 4, JULY 2002

Fig. 7. Control efforts with IC and IC : (a) v ; and (b) v .

plane, i.e., AMBs configuration space. One can see that all trajectories will not hit the bearing boundary at any time. Fig. 6(b) plane. In this figure, the trais the state trajectories on the jectories are clustered in the neighborhood of the origin. From , i.e., the figure, it is clear that all of the states remain in A/m . Also, Fig. 6 clearly indicates that the origin is asymptotically stable. The requirements 1), 2), and 3) given in Section III, are satisfied under control law (19) and (24). Fig. 7 displays the corresponding control efand are ranging from 2 V to forts. The input voltages 4 V at the transient period. The steady-state voltage is about V , which is to support the rotor weight. Fig. 8 depicts the state trajectories with the last two initial states which are close to the stator boundary and are outside the admissible set . In the figure, the dashed circle denotes the stator boundary. Although both trajectories approach the origin will cross over the stator at the steady state, the one with boundary during the transient period. In practice, it will impact with the stator, which is not allowed. This shows that the origin is globally asymptotically stable mathematically, but some initial states may lead to undesirable transient response due to the physical constraint of the stator boundary. In other words, if the initial state out of the admissible set is taken, the requirement that the rotor be kept inside the stator boundary is not guaranteed. Robustness is always an issue for the feedback linearizing control since there is no perfect model in the real world. Simple

Fig. 8. State trajectories projecting on (a) x x plane and on (b) x plane, where solid: IC , dash-dot: IC , dashed: stator boundary.

0x

Fig. 9. Effects of rotor mass variation (a) and (b) on rotor displacement; and (c) and (d) on control efforts, where the initial condition is IC and solid: nominal rotor mass dash-dot: 10 variation of rotor mass.

0 %

m m m m

m s m s m m m

m s

A m

A m

m s

A m A m

A m A m A m

m s m s

m s m s

A m

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 10, NO. 4, JULY 2002

625

in reality. For matched uncertainties, the robustness is easy to achieve by using sliding mode control instead of the linear control (24). However, if unmatched uncertainties exist, a backstepping approach is necessary in the controller design. How the rotor dynamics (e.g., the gyroscopic effect) affect the feedback linearization is another work to be explored in the future. REFERENCES
[1] J. Boehm, R. Gerber, J. R. Hartley, and S. Whitley, Development of active bearings for high speed rotors, IEEE Trans. Magn., vol. 26, pp. 25442546, 1990. [2] T. Ishimatsu, T. Shimomachi, and N. Taguchi, Active vibration control of flexible rotor using electromagnetic damper, in Proc. IEEE Int. Conf. Ind. Electron., Contr. Instrumentation 91, 1991, pp. 437442. [3] J. S. Kim and C. W. Lee, Flexible rotor system identification and vibration control by using a magnetic bearing, in Proc. 3rd Int. Symp. Magn. Bearings, 1992, pp. 123132. [4] T. Higuchi, M. Otsuka, and T. Mizuno, Identification of rotor unbalance and reduction of housing vibration by periodic learning control in magnetic bearings, in Proc. 3rd Int. Symp. Magn. Bearings, 1992, pp. 571579. [5] G. Curtelin, H. Saari, and B. Caron, Repetitive control of continuous systems: Comparative study and application, in Proc. IEEE Int. Conf. Syst., Man, Cybern. 93, 1993, pp. 229234. [6] A. M. Mohamed and B. V. Ilene, Imbalance compensation and automation balancing in magnetic bearing systems using the Q-parameterization theory, IEEE Trans. Contr. Syst. Technol., vol. 3, pp. 202211, June 1995. [7] K. Nonami and T. Ito, -synthesis of flexible rotor-magnetic bearing systems, IEEE Trans. Contr. Syst. Technol., vol. 4, pp. 503512, Sept. 1996. [8] K. Y. Lum, V. T. Coppola, and D. S. Bernstein, Adaptive autocentering control for an active magnetic bearing supporting a rotor with unknown mass imbalance, IEEE Trans. Contr. Syst. Technol., vol. 4, pp. 587597, Sept. 1996. [9] T.-N. Shiau, G.-J. Sheu, and C.-D. Yang, Vibrations and control of a flexible rotor in magnetic bearings using hybrid method and H control theory, ASME Trans., J. Eng. Gas Turbine Power, vol. 119, pp. 179195, 1997. [10] S.-L. Chen and C.-T. Hsu, Optimal design of a 3-Pole active magnetic bearing system, IEEE Trans. Magn., vol. 36, 2000. [11] G. Schweitzer, H. Bleuler, and A. Traxler, Active Magnetic Bearing. Zurich, Switzerland: Vdf Hochschulverlag AG, 1994. [12] R. D. Smith and W. F. Weldon, Nonlinear control of a rigid rotor magnetic bearing system: Modeling and simulation with full state feedback, IEEE Trans. Magn., vol. 31, pp. 973980, 1995. [13] J. Lvine, J. Lottin, and J.-C. Ponsart, A nonlinear approach to the control of magnetic bearings, IEEE Trans. Contr. Syst. Technol., vol. 4, pp. 524544, Sept. 1996. [14] L.-C. Lin and T.-B. Gau, Feedback linearization and fuzzy control for conical magnetic bearing, IEEE Trans. Contr. Syst. Technol., vol. 5, pp. 417426, July 1997. [15] L. Li and J. Mao, Feedback linearization of magnetic bearing actuators for a uniform upper bound of force slew rate, Proc. Electric Power Applicat., vol. 146, no. 4, pp. 378382, 1999. [16] L. Li, Linearizing magnetic bearing actuators by constant current sum, constant voltage sum and constant flux sum, IEEE Trans. Magn., vol. 35, pp. 528535, 1999. [17] S. Joo and J. H. Seo, Design and analysis of the nonlinear feedback linearizing control for a electromagnetic suspension systems, IEEE Trans. Contr. Syst. Technol., vol. 5, pp. 135144, Jan. 1997. [18] D. L. Trumper, S. M. Olson, and K. Subrahmanyan, Linearizing control of magnetic suspension systems, IEEE Trans. Contr. Syst. Technol., vol. 5, pp. 427438, July 1997. [19] M. D. de Queiroz and D. M. Dawson, Nonlinear control of active magnetic bearing: a backstepping approach, IEEE Trans. Contr. Syst. Technol., vol. 4, pp. 545552, Sept. 1996. [20] C.-T. Hsu and S.-L. Chen, Nonlinear control of a 3-Pole active magnetic bearing system, Automatica, 2000. [21] H. K. Khalil, Nonlinear Systems, 2nd ed. Upper Saddle River, NJ: Prentice-Hall, 1996. [22] A. E. Rundell, S. V. Drakunov, and R. A. DeCarlo, A sliding mode observer and controller for stabilization of rotational motion of a vertical shaft magnetic bearing, IEEE Trans. Contr. Syst. Technol., vol. 4, pp. 598608, Sept. 1996. [23] A. Isidori, Nonlinear Control Systems, 3rd ed. New York: SpringerVerlag, 1995.

Fig. 10. Effects of air gap variation (a) and (b) on rotor displacement; and (c) and (d) on control efforts, where the initial condition is IC and solid: nominal rotor mass dash-dot: 10 variation of air gap.

0 %

analysis can be done based on the present results using well-developed tools (see, e.g., [21]) and is omitted here. However, we shall examine the effects of parameter variation through simulations. Fig. 9 demonstrates the effects of variations in the rotor mass and Fig. 10 shows those in the nominal air gap. One can see that the system is more sensitive to the rotor mass variation which induces a large steady-state error. This is because the variation in the rotor mass will yield a wrong bias flux , resulting in a wrong steady state. On the other hand, the magnetic flux is a system state which is assumed to be directly measured and fedback. Although the variation in the nominal air gap may affect the magnetic flux, its influence on the system response is little. V. CONCLUSION A 3-pole AMB has been proposed in this study. It has the advantage of lower power loss and lower cost compared with the conventional 8-pole system. However, the magnetic flux coupling effect makes the 3-pole AMB strongly nonlinear. With the addition of the actuator dynamics, the overall system is more complicated. Fortunately, the voltage-controlled 3-pole AMB system can be modeled by quadratic nonlinear state equations provided that the magnetic fluxes are taken as part of the system states. This quadratic nonlinear system is shown to be feedback linearizable. A feedback linearizing controller is thus designed to stabilize the desired steady state. By a Lyapunov analysis, an admissible set in the state space is obtained. For initial conditions in the admissible set, it is guaranteed that the rotor will not hit the stator boundary at any time. Numerical simulations verify the theoretical analysis. One of the future works includes the experimental study of a 3-pole magnetic rotor-bearing system, which is the subject of a current project. Therefore, the robustness needs to be taken into account because of the imperfect cancellation that may arise

Das könnte Ihnen auch gefallen