Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
480 487, July 2010 Published online 25 April 2010 in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/asjc.210
Key Words: Magnetic levitation system, feedback linearization, integral sliding-mode, robust optimal control.
I. INTRODUCTION
Magnetic levitation systems have been widely used in various elds, such as frictionless bearings [1], maglev trains [2, 3], and educational experimentation [4]. As the open-loop electromechanical dynamics of a magnetic levitation system is nonlinear and unstable, the control design becomes very important for practical applications. In recent years, many studies on controls of magnetic levitation systems via nonlinear control techniques have been published, such as adaptive control [5], switching mode control [6], and feedback linearization methods [7, 8]. It has been shown that the dynamics of a voltagedriven magnetic levitation system (MLS) can be represented by a third-order nonlinear model [68]. Therefore, applications of the feedback linearization
Manuscript received January 3, 2008; revised November 1, 2008; accepted February 15, 2009. The authors are with the Department of Electrical Engineering, National Dong Hwa University, No. 1, Sec. 2, Daxue Rd., Shoufeng, Hualien 97401, Taiwan. Professor Hsin-Jang Shieh is the corresponding author (e-mail: hjshieh@mail.ndhu.edu.tw).
techniques to the MLS have received considerable attention [7, 8]. This is because the feedback linearization techniques can be used to transform the mathematic model of the MLS to a linear model [9], allowing the position tracking control design from the linear model to be developed conveniently. Nevertheless, feedback linearization techniques generally have a common drawback: exact information of the system parameters for a complete linearized model is always required [1]. Unfortunately, this drawback often leads to the robustness problem of control systems [10]. In order to solve the robustness problem caused by the feedback linearization strategy, an integral sliding-mode control with H method is proposed. Studies on applications of the integral sliding-mode control to the magnetic levitation systems [6, 11] have been published. These studies use the following advantages of the integral sliding-mode control: nominal performance dependent on sliding surface design and insensitivity to system uncertainties in the whole process [1214]. Unfortunately, both the conventional sliding-mode and integral sliding-mode controls are only robust to matched uncertainties or perturbations. Therefore, developments in recent years on improving the (integral) sliding-mode controls for robustness to unmatched uncertainties have
2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society
H.-J. Shieh et al.: A Robust Optimal Sliding-Mode Control Approach for Magnetic Levitation Systems
481
been presented. These methods summarily include the following: dynamic sliding-surface design [6], a modied switching algorithm [12], and robust optimal controls [13, 14]. Although these methods seem to solve the robustness problem against unmatched uncertainties, the following drawbacks still exist: (i) the proof derived in [6] of the system stability used complicated mathematical set theories; (ii) a nonlinear switching algorithm for sliding behavior was required in [12]; and (iii) a complicated combination control using the optimal linear quadratic method and linear matrix inequality scheme was developed in [13] and [14] to the sliding-mode control design. Unfortunately, these drawbacks often lead to difculty in practical implementation. Nevertheless, it is worth noticing that [13] and [14] give a good control-design concept; that is, combining the integral sliding-mode control with another robust technique can effectively guarantee the stability of the systems with the presence of unmatched uncertainties. Therefore, using the design concept, this paper proposes an integral sliding-mode control with robust optimal approach for position tracking of the MLS. It has been shown that the dynamics of the voltage-driven MLS is a third-order nonlinear model. To obtain a linear model derived from the nonlinear dynamics of the voltage-driven MLS, the feedback linearization method is adopted. Considering the parametric uncertainties and loading iron-ball, the matched and unmatched uncertainties certainly appear in the linear model. To obtain a high-performance and robust system to the unmatched uncertainty, the robust optimal sliding-mode control is developed from the linear model. In the developed control method, the integral sliding-mode control is to confront the matched uncertainty and the robust optimal controller is to guarantee the robustness to the unmatched uncertainty. To conrm the validity of the proposed control approach, an implementation of the control algorithm on the computer-controlled voltage-driven MLS is developed and the experimental results of the time responses from the MLS are illustrated.
Electromagne et
i Drive Amplifier u
are taken: (i) the ux linkage in the space between the electromagnet and ball is uniform; (ii) the crosssections of both the electromagnet and ball are identical; (iii) the ux leakage and eld fringing of the electromagnet is ignored; and (iv) the magnetic eld is always in the linear region of the B-H curve. With these assumptions, the differential equations of the MLS can be expressed below: d2 SN2 i2 m 2 h = mg 0 (1) 4 h2 dt S N 2 di S N 2 dh Vin = Ri + 0 0 2 i (2) 2h dt 2h dt where 0 denotes permeability, g denotes gravity, m denotes mass of the loading ball, S denotes the crosssection of the electromagnet, N denotes the number of turns in the coil, h denotes the air-gap distance between the electromagnet and ball, i denotes the coil current, R denotes the equivalent resistance of the coil, and Vin denotes the applied voltage to the system. Allowing x T = [x1 x2 x3 ] = [h h i], (1) and (2) can be rewritten as follows: x2 0 2 2 0 u x = g a1 x3 x1 + (3) 1 a3 x 1 x1 x2 x3 a2 x1 x3 where u = Vin , a1 = ( 0 S N 2 )/4m, a2 = 2R/( 0 S N 2 ), and a3 = 2/( 0 S N 2 ). Considering the system uncertainties due to deviations of the resistance and loading mass, (3) becomes: x2 0 g a1 x 2 x 2 3 1 0 u x = + a3 x 1 x 1 x x a x x
1 2 3 2 1 3
0 + d1 d2
(4)
2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society
482
Asian Journal of Control, Vol. 12, No. 4, pp. 480 487, July 2010
2 2 where d1 = a1 x3 x1 and d2 = a2 x1 x3 , in which a1 and a2 denote the deviations of a1 and a2 , respectively. From (4) it can be found that the dynamics of motion of the magnetic levitation system is obviously nonlinear. Therefore, for control design convenience, the following feedback linearization technique is adopted.
transformation, the new state vector similar to (8) can be derived in the following:
2 2 z T = [z 1 z 2 z 3 ] = [x1 x2 g a1 x3 x1 ]
(11)
2.2 Feedback linearization model Consider a single input single output (SISO) nonlinear system described by: x = f (x)+ g(x)u and x(0) = x0 (5)
Furthermore, the new control input associated with (9) can be designated as: x1 a3 us (12) u = x3 a2 2a1 a3 x3 where u s is referred to as a new control input. Consequently, the linearized model based on the new state variables of (11) can be expressed by: z1 z1 0 1 0 0 z 2 = 0 0 1 z 2 + 0 u s z3 0 0 0 0 + d1 d2 Here, the proposed control approach from (13) will be given in the next section. z3 1
where x R n is state vector, f (x) and g(x) are real smooth vector elds on R n , and u is a control input. The nonlinear system given in (1) is input-state linearizable if and only if there exists a region R n such that the following conditions hold. (i) The vector elds of {g ad f g adn1 g} are f linearly independent in . j (ii) [adif g, ad f g] span{g ad f g adn2 g} for f i, j = 0, 1, . . . , n 2, where [] indicates LeeBracket [9], and adif g is dened as follows: ad0 g(x) = g(x) and f adif g(x) = [ f, adi1 g(x)] f (6)
(13)
If the conditions of (i) and (ii) are satised, we can nd a scalar valued function
(x) such that: *
(x) adi g(x) = 0, i = 1, . . . , n 2 f *x T (7) *
(x) n1 ad f g(x) = 0 *x T The state transformation is dened as z=T (x)=[L 0
(x) L 1
(x) L n1
(x)]T f f f and the system input is given below: u= L n
(x) f L g L n1
(x) f + uN L g L n1
(x) f 1 (9) (8)
where u N denotes a new control input for system given in (5). With the use of (8) and (9), a linear system from (5) can be obtained as follows: z = [z 2 z n 0]T +[0 0 1]T u N (10)
Now, considering the dynamic system of (3), the state transformation which satises the conditions of (i), (ii), and (6) can be selected as
(x) = x 1 . Using this state q
2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society
H.-J. Shieh et al.: A Robust Optimal Sliding-Mode Control Approach for Magnetic Levitation Systems
483
denition of (14), the dynamics of the tracking errors can be derived as follows: e1 e1 0 0 1 0 0 e 2 = 0 0 1 e2 + 0 u + d 1 0 0 0 1 e3 d2 e3 = Ae + B u +d (15) where d denotes the system uncertainty vector and ... u = u s +r . Furthermore, the following control law is given by: u(x) = u 0 (x)+ u 1 (x) (16) where u 0 (x) is a nominal control responsible for the nominal system performance and u 1 (x) is a discontin uous control that rejects the matched uncertainty when the control system is in the sliding phase. The stable sliding manifold is designated as follows: s =G e
0 t
Then, by setting s = 0, the following equivalent control can be obtained u 1eq = B + d (G B)1 Gdu (21)
By substituting u 1eq for u 1 into (15), the dynamics of the system in which the states remain at the sliding manifold can be derived in the following: eeq = Aeeq + B[u 0 B + d (G B)1 Gdu ]+d = Aeeq + B u 0 +[I B(G B)1 G]du = Aeeq + B u 0 + du (22)
(Ae + B u 0 ) d
(17)
where G denotes a constant matrix. Before analyzing the effect of the unmatched uncertainty, the following proposition is introduced here [15]. Proposition 1. Considering the error dynamics of (15), the following identity matrix I3 R 33 can be derived: I3 = B B + + B B + (18) where B + denotes the left inverse of B, dened as B + = (B T B)1 B T , and the columns of B R 32 span the null space of B + ; i.e. B + B = 0. According to Proposition 1, the following matrices associated with (15) can be derived. 1 0 B + = [0 0 1] and B = 0 1 0 0 Therefore, the vector d can be expressed by: d = dm +du (19) where dm = B B + d = [0 0 d2 ]T and du = B B + d = [0 d1 0]T denote the matched and unmatched uncertainties, respectively. To determine the motion equations at the sliding manifold, the equivalent control concept is adopted here. Taking the time derivative of s yields: s = G[e Ae B u 0 ] = G(Ae + B u +d) G(Ae + B u 0 ) = G B(u 1 + B + d)+ Gdu q (20)
where = [I B(G B)1 G] and eeq denotes the equivalent error state-vector in the sliding phase. From (20)(22), it can be found that the dynamics in the sliding phase does not contain the matched uncertainty. On the other hand, in the sliding phase, the control of u 0 can be used for two issues: to stabilize (22) and to confront the unmatched uncertainty. To complete these two issues, the sliding-mode with H control is developed. Here, we purposely set G = B + and give the following switching control: u1 = s s (23)
where is a positive constant satisfying B + d . To prove the stability, the Lyapunov function below V= 1 s 2
2
(24)
is selected. Since B + du = 0, the time derivative of (24) can be deduced as follows: V = s T s = s T [B + B(u 1 + B + d)+ B + du ] = sT = 0 s + B + d + B + du s
s +s T B + d (25)
s ( B +d )
Consequently, it shows that the system described by (15) can be stabilized in the sense of Lyapuov stability by the sliding-mode control of (23) subject to the integral sliding manifold of (17). 3.2 Robust optimal control From the development in the previous subsection, it can be found that the dynamics of the system on the
2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society
484
Asian Journal of Control, Vol. 12, No. 4, pp. 480 487, July 2010
sliding manifold can be equivalently expressed by: e1 e1 0 1 0 0 0 e 2 = 0 0 1 e2 + 0 u 0 + 1 d 1 0 0 0 1 0 e3 e3 = Ae + B u 0 + Bu d1 (26) Moreover, an output vector related to (26) is purposely given as follows: 20 0 0 0 e1 0 5 0 0 z= e2 + 0 0 20 0 e3 0 0 0 1 u 0 = Ce + D u 0 (27) Then, to minimize the H norm of the transfer matrix Tzd that goes from d1 to z, the following theorem is adopted [16]. Theorem 1. There exists a controller satisfying Tzd
<
Remark 1. Using the controls of (21) and (23), the position tracking control system of the MLS has an improved tracking performance and is robust to the uncertainty of the electric parameter deviations. Moreover, the control of (31) can further guarantee the robustness to the suspended load uncertainty that cannot directly be completed by the controls of (21) and (23). Remark 2. From the review of the previous studies [6, 1214], the following consequences can be given: (i) The proof of the system stability shown in (25) is simple; (ii) A simple switching control of (23) has been adopted; and (iii) For the robust control algorithm, a linear Riccati equation of (29) with a robust optimal state-feedback control of (31) has been considered.
(28)
if there exists a real, symmetric, positive semidenite matrix X satisfying the Riccati equation in the following X A+A T X X (B B T =0
2 T Bu Bu )X +C T C
(29)
Considering (26)(28), the following solution to (29) can be obtained by the optimal control tool of the MATLAB computer software. 5883 17386 855 X = 2182 57947 2854 and 855 2854 161 = 2.35 (30)
Then, the robust optimal control with state feedback can be given by: u 0 = B T X e = [855 2854 161]e (31)
Consequently, the proposed control approach is shown in (16) subject to the sliding manifold below: s = B+ e
0 t
(Ae + B u 0 ) d
(32)
From equation (16) it can be observed that using the proposed control approach to the position tracking of the MLS can give the advantages of the optimization and robustness to both the matched and unmatched uncertainties. q
2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society
H.-J. Shieh et al.: A Robust Optimal Sliding-Mode Control Approach for Magnetic Levitation Systems
485
Integral Sliding-Mode Controller Command + _ h + Robust Optimal Controller + Magnetic Levitation System Position
h Current i
Fig. 2. The overall control architecture for the magnetic levitation system. Table I. Specications of the MLS. Parameters R m io h0 a1 a2 a3 Values and Units 2.4 2102 kg 0.46 A 0.023 m 8.9845103 6.12165104 2782.57 Fig. 3. Square trajectory response with a steel ball suspended using the proposed control approach.
transient tracking performance is mainly governed by the designated sliding manifold and the convergence of the tracking errors can be ensured when the states of the system remain at the manifold. Moreover, to show the robustness to the uncertain load mass of the suspended balls, experiments simultaneously using two suspended steel balls were performed. In this case, Figs 5 and 6 show the experimental results of the square and sinusoidal trajectory response, respectively. From a comparison of Figs 36, it can be found that the transient performance in trajectory tracking response with one-ball load mass is almost the same as that with two-ball load mass; that is, the robustness to an uncertain load mass is substantially obtained. In addition, to show the improved performance by illustration comparison, experimental time response under the conventional integral sliding-mode control is illustrated in Fig. 7. By comparing Fig. 4 with Fig. 7, one can nd that signicant improvement in tracking performance is achieved. In summary, the experimental results have demonstrated that the effectiveness, robustness, and performance improvement of the robust optimal sliding-mode control on the MLS are ensured.
Fig. 4. Sinusoidal trajectory response with a steel ball suspended using the proposed control approach.
V. CONCLUSION
An optimal integral sliding-mode control design for position tracking of a magnetic levitation system has been proposed in this paper. To develop easily q
the proposed control approach, a linear model which approximately represents the nonlinear dynamics of motion of the magnetic levitation system was established by the feedback linearization method. From the linear model established, the integral sliding-mode controller with optimal control technique was developed. In the proposed control approach, the integral sliding-mode control was designed to guarantee the sliding mode behavior on the stable sliding manifold and the robust optimal control accompanied by the integral sliding manifold also was developed. In the robust optimal control method, the state feedback gain was obtained by solving the Riccati equation. Using the proposed control method to position tracking of the magnetic levitation system can give the following advantages: (i) robustness to matched and unmatched
2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society
486
Asian Journal of Control, Vol. 12, No. 4, pp. 480 487, July 2010
Fig. 5. Square trajectory response with two steel balls suspended using the proposed control approach.
Fig. 7. Sinusoidal trajectory response using the conventional integral sliding-mode control approach.
Fig. 6. Sinusoidal trajectory response with two steel balls suspended using the proposed control approach.
uncertainties, (ii) stability in tracking control, and (iii) improved transient performance. Finally, the experimental results of the time responses from the MLS were illustrated to conrm the validity of the proposed control approach for practical applications.
REFERENCES 1. Chen, M. Y., M. J. Wang, L. C. Fu, A novel dual-axis repulsive maglev guiding system with permanent magnet: modeling and control design, IEEE/ASME Trans. Mechatron., Vol. 8, No. 1, pp. 7786 (2003). 2. Ono, M., S. Koga, H. Ohtsuki, Japans superconducting maglev train, IEEE Instrum. Meas., Vol. 5, No. 1, pp. 915 (2002). q
3. Holmer, P., Faster than a speeding bullet train, IEEE Spectr., Vol. 40, No. 8, pp. 3034 (2003). 4. Hurley, W. G. and W. H. Wole, Electromagnetic design of a magnetic suspension system, IEEE Trans. Educ., Vol. 40, No. 2, pp. 124130 (1997). 5. Yang, Z. J. and M. Tateishi, Adaptive robust nonlinear control of a magnetic levitation system, Automatica, Vol. 37, pp. 11251131 (2001). 6. Yang, Z. J., K. Miyazaki, S. Kanae, K. Wada, Robust position control of a magnetic levitation system via dynamic surface control technique, IEEE Trans. Ind. Electron., Vol. 51, No. 1, pp. 2634 (2004). 7. Hajjaji, A. E. and M. Ouladsine, Modeling and nonlinear control of magnetic levitation systems, IEEE Trans. Ind. Electron., Vol. 48, No. 4, pp. 831838 (2001). 8. Joo, S. J. and J. H. Seo, Design and analysis of the nonlinear feedback linearizing control for an electromagnetic suspension system, IEEE Trans. Control Syst. Technol., Vol. 5, No. 1, pp. 135144 (1997). 9. Khalil, H. K., Nonlinear Systems, Prentice-Hall, New Jersey, U.S.A. (2002). 10. Sastry, S., Nonlinear Systems: Analysis, Stability, and Control, Springer, New York, U.S.A. (1999). 11. Lee, J. H., P. E. Allaire, G. Tao, X. Zhang, Integral sliding mode control of a magnetically suspended balance beam: analysis, simulation, and experiment, IEEE Trans. Ind. Electron., Vol. 6, No. 3, pp. 338346 (2001). 12. Cao, W. J. and J. X. Xu, Nonlinear integral-type sliding surface for both matched and unmatched
2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society
H.-J. Shieh et al.: A Robust Optimal Sliding-Mode Control Approach for Magnetic Levitation Systems
487
uncertain systems, IEEE Trans. Autom. Control, Vol. 49, No. 8, pp. 13551360 (2004). 13. Castanos, F. and L. Fridman, Design of integral sliding manifolds for multi-model uncertain systems via LMI. Proc. Int. Workshop Variable Struct. Syst., Italy, pp. 6367 (2006). 14. Bejarano, F. J., L. Fridman, A. Poznyak, Output integral sliding mode with application to the LQoptimal control, Proc. Int. Workshop Variable Struct. Syst., Italy, pp. 6873 (2006). 15. Castanos, F. and L. Fridman, Analysis and design of integral sliding manifolds for systems with unmatched perturbations, IEEE Trans. Autom. Control, Vol. 51, No. 5, pp. 853858 (2006). 16. Doyle, J. C., K. Glover, P. P. Khargonekar, B. A. Francis, State-space solutions to standard H2 and H control problems, IEEE Trans. Autom. Control, Vol. 34, No. 8, pp. 831847 (1989).
Hsinchu, Taiwan, as a Researcher. Since Aug. 2002, he has been with the Department of Electrical Engineering, National Dong Hwa University, Hualien, Taiwan, where he is currently an Associate Professor. His research interests include piezoelectric mechanisms, power converters, photovoltaic systems, and control theory applications. Dr. Shieh was the recipient of the 2007 Taiwain Power Electronics Conference Best Paper Award. He is a member of the IEEE, Taiwan Power Electronics Association, and Chinese Automatic Control Society of Taiwan. Jheng-Hong Siao was born in Tainan, Taiwan, in 1985. He received the B.Eng. degree in Electrical Engineering from National Dong Hwa University (NDHU), Hualien, Taiwan, in 2007. He is currently working toward the Ph.D. degree at NDHU. His research interests include drive and control of magnetic levitation systems and electric machines control for hybrid electric vehicles. Yu-Chen Liu was born in Hualien, Taiwan, in 1983. He received the M.S. degree in electrical engineering from National Dong Hwa University, Hualien, Taiwan, in 2007. His reserach interests include drive and control of magnetic levitation systems.
Hsin-Jang Shieh received the B.Eng. and Ph.D. degrees in Electrical Engineering from the National Central University, Chung-Li, Taiwan, in 1992 and 1997, respectively. From 1997 to 2002, Dr. Shieh was with the Mechanical Industry Research Laboratories, Industrial Technology Research Institute,
2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society