You are on page 1of 5

FM14 = 150

Gain Scheduled Control of Magnetic Suspension System*


Young Chol Kim+ Dept. of Electronics, Chugbuk National University Cheongju, 360-763, KOREA Kook Hun Kim Control Applications Lab., Korea Electrotech. Research Institute

P.O.Box 20, Chaiigwon, 641-120, KOREA


Abstract
A gain scheduling approach for tahe suspension control of a nonlinear MAGLEV System is presented. We show that. this technique is very useful for improving not only performance to the operational disturbances originating a.erodynamic force but also robustness t o the uncertainty of payload. As a scheduling variable, even though the ext,ernal disturbance need t o be estimated in real time, but the additive measurement is not required. Some simulations show t h a t the gain scheduling control system performs very well comparing with other methods using a uonlinear feedback linearization or a fixed gain linear feedback. the change of mass and the performance against the practical disturbances such as aerodynamic force or movement of payload etc. From this point of view, this paper deals with a design method of gain scheduled controller for SMLS. T h e gain scheduling is a uonlinear feedback control of special type ; i t has a linear regulator whose parameters are changed as a operating conditions. T h e model is usually arranged so that the operating condition is specified by the values of exogeneous signals or variables, so called scheduling

variable.
As a SMLS t o be controlled has known nonlinearity and is modeled in special form, a scheduling variable could have been selected easily by means of only estimating the disturbance. Some simulations show that the proposed gain scheduling controller for SMLS has much better performancc than other two methods using a nonlinear feedback linearization and a fixed-gain linear feedback.

1. Introduction

Magnetically Levitated System(MAGLEV) t o be considered is a D. C. electromagnetic suspension (EMS) system, which is highly nonlinear dynamics and open-loop unstable. Different schemes of stabilization and control of single magnet 1evita.tion system(SMLS) have been widely st,ndied. A design method of state feedback control with poleplacenient [l, 3, 51 has been mainly applied on the basis of t,he first-order linear approximate model corresponding t o only an operat,ing point. However because the operating point depends on the change of the snspended mass and ext*ernaldisturbances, the dynamics of linear approximate model should be also varied. Hence this approach is difficult of robushess. In recent, years, nonlinear control schems for MAGLEV have been researched. Sinha [3] suggested a digital implementation of model reference adaptive controller for EMS and Jin et al. [4]proposed a SMLS controller using uonlinear feedback linearization. By the way,it, is very significant problems that t,o what, extend the coiit,roller designed for EMS has the stiffness t,o *This work was partially supported by ICOSEF and partially supported by Korea MOST tI-Iis e n i d address is yckim Ocbucc . chungbuk.ac .kr

2. Gain Scheduled Control SMLS


2.1. Modeling of a SMLS For the SMLS shown in Fig.1, the vert,ical motion is described by the following nonlinear dynamic equation[l]:

,"(t)= --(-)c

i(t) 4m z ( t )

* + -r1 fd(f) n

+y

(1)

where,z(t), z ( t ) , fd(t),and ~ ( t are ) the gap distance, current, disturbance , and the applied voltage respectively. Also R a n d g are the tsotalresistance of electroinagnet and gravity respectively ,and c is a constant.
For simplicity w e write eq.(l) and state equation;
ec1.(2) a.8 a. noiiliiiear

3127

where
.rT(t)
u(t)

:=
:=

.c:1(i) z:z(t) za(t)

z(t) i ( t )i(t) ]

Then for each scheduling variable f d E F the corresponding linearized closed loop system can be expressed in tlic form

.Zl(t)

Obviously the function f is smooth except for z l = 0. where

number of turns: N

Ut'
1, t L 01
(4) and the linearized control law coefficients are given by

+i(

t)

Figl. Single ma.gnet levitation systems(SMLS)


2.2. Design of G a i n Scheduling Controller First,, the dist,nrbance f d ( ' t ) is assumed to be bounded.

F-:= { f d ( t ) I fd(t) E [fi-, f,'

At each constant disturbance,fd(t) = f d , the balance equation of dynamic model for an operating point (ZO,io) becomes

F(io,zo) =
UO

-(-) 4 20 = Rio

c io

= mg+fd

(5)
$(fd)

A family of operating point vector state space form is expressed by

rewritten in the

a ( f d )= [

2o

zO,/-jF

] ,fd

E3

(6)

As the operating condition is a function of external disturbance xl,we select f d ( t ) as a scheduling variable. Here y(fd) t h a t for each constant f d othe scheduling variable f < l ( t ) represents , the desired output is called output t r i m condition. Then the output trim condition of SMLS is y(fd) = ZO. Now the objective is t o design a control law of the form

Summing up the design problem for gain scheduling, it, is to obtain a nonlinear cont,rol 1a.w (7) such that. sat.isfy eq.(8) and (13) subject t o (9). Here in order t,o achieve pole-assignment to the closed-loop 1inea.riza.tion, the eq.(9) is assumed t o be controlla.ble for a u y E F. Using Rugh's procedure [7, 81, we get, a st,ruct,nreof t,hc gain scheduling control law as follows ;
u ( t ) = h'T(fd(t))[Z(t)

- Z(fd(t))] + Q(fd(t))

(1.1)

where, I< is a smooth function such t h a t a t each f d the clo5ed- loop system should have a constant ~ ( f d ) aud the linearized closed-loop system about each constant fd should be asymptotically stable. In other words, the existance of such a closed-loop constant E ( & ) implies that for 5 s E F the smooth function z ( f d ) and .(fa) must satisfy the followings
f ( z ( f d ) j .(fd),

It is easy t o see t h a t the eq.(14) is derived from (9) mid (10) when setting fd = f&). In pole placement, scheine K l ( f d ) is obtained such that A ( f d ) + B l < (f<l) ~ should have desired characteristic roots for each f < i . T h e local st,at>ilil.y problem including uncertainty of rimss will be ha.ntlling later in section 2.3. We now consider two problems ; on(^ is the iincertainty of suspended mass ,and the othei the acquisition of scheduling variable fd(t) on real time. The payload of MAGLEV is changed according t o the nurnbei of passengers. Hence the mass can be expressed in the form m E [m- ,m+ ] where m- and mt indicate tlic minimum and the maximum value of m. Let m o denote the uoininal value of mass. At the design stage, only mo is given.

fd)

=0
fd)

a(fd)
Y(fl)

= li(@d), = -?o

= Rzo

-d

(8)

Next an estimator is introduced t o obtain t,he scheduling variable f d ( t ) .

is based on a linear approximate model corresponding to only one operating point.


2.2.

fd(t) = m0a(t) -(-)' 4 z(t)

+c

Z(t)

- mog

..

(measured data)

Since t,he state vector [ z i elT in the control law (14) can be constructed with measurements [Z z ZIT, the eq.( 15) is available without measuring another variables in addition. I t is now possible t o carry out the poleplacement scheme for computing KI . T h e characteristic equation of the closed-loop linearization system is given by

reference

regulator

Fig2.Block diagram of SMLS with scheduling control where


2.3. Local stability to both inass chaiige and disturbance

62

-(R-

2 zo

I & )
(18)

:= mg

+fd
(s
s3

and the desired characteristic equation is defined as

A*(.)

+a ) ( . ' + ~CW,,S +

U:)

+ d2sZ + d1s + do

(19)

T h e gain scheduling approach is local in nature. T h u s the overall performance and the stability must he checked by extensive simulations. However, under the condition t h a t the scheduling variable is slowly varying in a time average sense, i t has been proved in[7, 91 t h a t the gain scheduled control system provides stability. In this section, the local stability problems t o both the niicertaiiity of suspended mass and the constaiit, disturbance over : F will be discussed. Substituting eq.(20) into eq.( 16), the actual characteristic equation of closed-loop linearization system becomes
A,(S)

If f d ( t ) and m in eq.(18) are replaced by [&) and mo respectively, then the linear gain I S * i ( f d ( t ) ) = [Eill(t) K l z ( f ) Ii~a(t)] is given from (16)-(19)

:= s3

+ aas2 + +

((0

(24)

where

K13(t)

R- 2 ZO
(31)

cd2

From (1),(15),(18) and (21), we get

& ( t ) := mog + & t )

6 ( t )= W(t)- ( m - mo)Sa(t) Thus while the disturbance is slowly varying

(26)

At last, a gain scheduled controller for SMLS is obtained as follows, modifying (14) slighty,

u(t)= I C ( ~ ( ~ ) ) A q& W t ) ))
where

(22)

A i ( t ) :=
Z(fd(t))

z(t)

- ?i(jf)
0

= =

20

m 1"

(23)

ii(fd(t))

Rzod4-

a s much <is it does not make the velocity of lifting body chaiige, it is clear from (24) and (26) t h a t the closed-loop roots are independent on such a disturbance, aid are affected by only the change of mass. Now we are going to find the constraints for design parameters of d o . d l ancl t l ~ in ( 1 0 ) based on the sufficient conditions of local stability. Let the constant value of disturbance fa(t) rspress 1, E [ f : , f,'] , j=d E T ,similary t o the mass uncertainty. Assumed t h a t the maximum acceleration / deceleration is given by

T h e block diagram of SMLS with gain scheduling control is shown in Fig.2.

aCc:= mpx I&(t)l

, t 2 0.

(37)

Remark : If t,he disturbance force is disregarded, that


is
fd(t)

= 0, t

2 0,

then the eq.(22) is the same which

Of course, the allowable acceration which should be limited for ride quality depends highly on design factors

3129

(CY,C,W~~ in (19)), external disturbances, etc.


From (26), we have

6 ( t )5 Iw(t)

+ ( m - mo)accl

(28)

If 6 in (25) is replaced by the right side of (28), the eq.(24) represents a family of polynomial whose coefficeints are real and nonlinearly dependent on uncertain parameters m and f d . Therefore the above problem is concerned with t,lie invariance of Hnrwitz property for uncertain polynomial family as follows;
&(s,

500 x103, dl = 15 x l o 3 and d2=200). T h e bounds of mass uncertainty are regarded as m-=O.8m0 and mS =1.3mo. Only two cases of m = m- and m = m t a r e examined. In particular, we have considered several types of disturbance such as parabola, ramp, step, sinusoid, various general functions etc., whose magnitudes are all equal t o or less than 40% of nominal weight(m.og). T h e foliowing t,wo cases out of them will be shown in Fig.3 - Fig.5. Type I(parabo1a) :
fd(t)

= 17.Gsin(xt), 0 5

<1

q ) = s3

+ u2(q)s2 + ai(q)s + ao(q), 4 E Q


f d E [f;, fill

Type %general function) : (29)

where,
qT := [m f d ] , Q := {qlm E [m-, m'l,

Propositionl: T h e gain scheduled control law (22) retains the local stability of SMLS if the design parameters d,'s(i = 0 , 1 , 2 ) satisfy

(i) di > 0

( i i ) L > b:
(rii)+ot d2

d dz

(30)

+ b2+ < dlq-

Suppose t h a t ucc = 0 . l g = 0.98m/s2. Then '7 E [0.7568 , 1.28251, b: = 130.12, and bz E [-165,1501 are calculated. T h u s the above dt's values satisfies the ineqality of (30) for local stability. Now, three kinds of controller are compared with together ; the first one is a gain scheduled controller (GSC), the second is a nonlinear feedback linearizing controller(NFC) proposed by Jin et a1.[4] and the third is a conventional linear approximation feedback controller(LFC) [& 51. NFC :v(t) = Fl(z)/F2(2), where

F I ( ~ )= dozo - doz(t) - d2(--(-)' 4mo z ( t )


- + ~ ~ - ( W + M ) a , , ) ] mo zo

i(t)

+y) (33)

F2(z) (31) max { b z } , i = 1 , 2


qEQ qEQ

1 i(t) -- m o z(t)

(34)

max { v } , 17- := min (17)


qEQ

mg+fd, M:=m-mo

(32)

fd,

Proof:Using some algebra and Hermite-Bieler Theorem [lo], it is not difficult t o prove it. T h u s it is omitted here.

In above procedure, it has been assumed t h a t mg > q E Q. Three conditions of (30) above can be used as a design guide when we choose d,'s(i=0,1,2) within stable region.

3. Simulation and Discussion Simulations are performed on the well-known laboratory modelk] whose d a t a are i o = 2[A] , zo = 1.5[mm] , mog = ~ [ I i g , ] c = 9.9 x T h e design specifications for above SMLS are considered such that satisfies (i) damping ratio 2 0.707 (ii) suspension natural frequency 2! 10 Hz. In (19), we choose a set of design parameters which are given by a=100, C =0.707, wn=50/0.707 (i.e.,do =

LFC : T h e GSC becomes a LFC only if we set f < j ( t ) =0 in equations (20),(22), and (23). I t is shown from Fig. 3 t o Fig. 4 that. the gain schednling control have very nice performances in all t,he ca.ses. Despite the large magnitude of disturbance and mass 1111certainty are applied to, the excellent, tra.nsieiit, response and very little steady state errors are shown. Note t h a t no integral action is introduced in t,liis gain scheduling controller. In some cases(when the mass increases 30% of nominal), a linear approximate feedba.ck controller (LFC) have the system t o be unst,a.ble. An integrator added in LFC loop make t h e transient behaviour and steady state error not t o improve against general disturbances as shown in Fig.5. A nonlinear feedback linearization controller [4]has good robust stability t o almost all the initial condit,ions. However, simulations show that its control performance is very sensitive t o both mass change and disturbances.

4. Conclusion

One of the most important problems in EMS systems is how much robust the designed controller is against pra.ct,ical disturbances and uncertainty. With a view t,o solve it,, a gain scheduling controller for SMLS has been proposed.

3160

As the operating condition is a function of disturbance, we choose the estimated disturbance as a scheduling variable. A simple estimator can b e constructed without additive measurements. We showed that the gain scheduling is robust if the disturbance is slowly varying as much as it does not make the velocity of lifting body change (i.e., so that acceleration/deceleration is nearly zero). Also some constraints for the design parameters are suggested. They can be used when we find the design parameters within stable region. T h e gain scheduling is inherently local in nature. So the overall performance and stability of the control system have been demonstrated by different simulations. In most cases, the gain scheduling control designed for a SMLS model showed very good results. Coniparing with the gain scheduling, we performed the same simulations for two other control methods ; (a) nonlinear feedback linearizing controller, (b)a fixed gain linear approximate controller. Both method does not show the satisfactory performances t o the disturbance and the mass change.

x 10~

........: ............................

...............

: :

GSC
0

0.2

0.4 0.6 time(sec)

0.8

fd

Fig.3 System response for GSC and NFC wit,li W I = 1 . 3 7 1 1 ~ . =type 1, and zo = 1.5.

4 1
References [l] B. V. Jayawant, et al. 3, Control and Dynamics of magnetically Suspended Vecles, T h e 7th IFAC World Congress ,pp. 1325-1332, 1978. [a] P. I<. Sinha, Electromagnetic Suspension : Dynamic and Control, Peter Peregrinus, 1987. [3] P. K. Sinha, Real-time control of a nonlinear MAGLEV system using concurrent processing techniques, Proc. Int. Conf. MAGLEV89, pp. 417-422, 1989. [4] J. H. Jin, J. 1%.Seo and I ( .3. Kim, Controller development for a single-magnet suspension system using nonlinear feedback linearization, Trans. KIEE, Vol. 41, NO. 3, pp. 292-299, 1992. [5] Y. C. Kim et al 5, A design of robust controller for multimagnet levitation system : part - I1 decentralized multilevel control, J. Ind. Sci and Tech. Institute, Vol. 5, NO. 2, pp. 169-185, 1991. I<. J. Astrom and B. Wittenmark, Adaptive [GI Control,Addison-Wesley, 1989. [7] W. J. Rugh, Analytical framework for gain scheduling, IEEE Control System, Vol. 11, No. 1, pp. 79-84,1991. [8] W. T. Baumann, W. J. Rugh, Feedback control of nonlinear systems by extended linearization, IEEE Trans. 011 Auto. Con. , Vol. AC-31, No. 1, pp. 40-46, 1986.

io9

.......... . :. . . . . . . . . .

.:. . . . . . . . . . .:. . . . . . . . . . . .

_--GSC
0.5
1 time(sec)

1.5

Fig.4 System response for GSC and NFC with n, = 1 . 3 r n 0 , fd =type 2 , and 20 = 1.5.

[g] D. A. Lawrence, W. I. Rugli, On a stability theorem for nonlinear systems with slowly varying inputs, IEEE Trans. Auto. Cont. , Vol. AC-35, No. 7, pp. 860864, 1990. [lo] F. K. Gantmacher, Theory of matrices, Chelsea,1959.

-0

0.5

1 time(sec)

1.5

Fig.5 System response for LFC with Integratoi. m = 1.3m0, fd =type 2, and 20 = 1.5mnt.

3131

You might also like