Вы находитесь на странице: 1из 10

>

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. I , NO. I , MARCH 1993

ZT

High-Performance Nonlinear Feedback Control of a Permanent Magnet Stepper Motor


Marc Bodson, Member, IEEE, John N. Chiasson, Member, IEEE, Robert T. Novotnak, Student Member, IEEE, and Ronald B. Rekowski
. I

..
* .

practical significance. Issues such as rotor speed estimation from an optical encoder, the effects of high-frequency noise on control law is developed using the exact linearization methodology current measurements due to pulse-width modulated (PWM) and implemented on an industrial setup. The practical issues 'of speed estimation and voltage saturation are considered and amplifiers, and the constraints of source voltage limits are resolved through the use of a nonlinear observer and field- addressed and resolved. weakening, respectively. The results of the implementation of the The motion control industry has made extensive use of dc control algorithm for an industry-specified point-to-point move motors, due to the relative ease in achieving high-performance of a linear positioning table are presented and discussed. with precise control. However, dc motors are fading away to the benefit of ac motors as the latter are much more I. INTRODUCTION reliable (no brushes to wear out), have a 100% duty cycle HERE has been a considerable development in nonlinear (no rotor windings that need time to cool down), and have a control theory in the last 15 years with such ideas as much higher torque-to-inertia ratio (again, due to the lack of feedback linearization, inputloutput linearization, and passivity rotor windings). On the other hand, ac motors are nonlinear theory (see, e.g., [ 11-[3] and references therein). However, devices and thus more difficult to control. The standard actual applications resulting in the implementation of such method of linearizing the system dynamics about an operating control algorithms in an industrial setting have been few. point and then designing a linear controller will not suffice This situation is probably due to the computational require- because positioning systems are constantly accelerating and ments of the new algorithms and to the unfortunate lack decelerating and a gain-scheduled controller cannot provide of communication between theoreticians and practitioners of (guarantee) the precise tracking necessary to make full use of control engineering. Yet, the capabilities of computational the torque capabilities of the motor. The feedback linearizing hardware have steadily increased and, due to technological controller used in this paper guarantees global trajectory breakthroughs in digital signal processors (DSP) in the last tracking as long as the amplifier limits are not exceeded. A side few years, it is now possible to implement complex nonlinear advantage of such a control law is also that it is well-suited control algorithms with relatively inexpensive components. As to self-tuning and adaptive control since it is model-based. In f a the transfer of knowledge between university and industry, particular, the parameters of the motor were found using a it can be best improved through close cooperative efforts: this least-squares identification procedure [7]-[ 101. The experimental setup consists of a PM stepper motor paper describes the results of such a collaboration. A feedback linearization algorithm is used to control a per- with a 2000 line encoder attached to a linear positioning manent magnet (PM) stepper motor for point-to-point moves table, two PWM amplifiers, and a Motorola DSP56001 digital of a linear positioning table. The applicability of the method signal processor for implementing the algorithm. Using a to the basic stepper motor model was investigated in [4]-[6], feedback linearizing controller and a speed observer, superior where it was shown that the basic transformation that was performance is demonstrated using this nonlinear control alneeded was equivalent to the Blondel-Park or direct-quadrature gorithm. Note that some related work on the control of stepper (DQ) transformation of electric machine theory. In this work, motors and synchronous motors has been previously reported. we report on an experimental investigation implementing the Specifically, Ilic et al. [ 111 have considered the feedback control algorithm in a setup provided by the industry. The linearization technique on variable reluctance stepper motors hardware is only slightly upgraded from a product currently and an implementation of this controller was reported in [ 121. sold by Aerotech, Inc. and, therefore, the results have direct Chen and Paden [ 131 have considered low speed control of the PM stepper motor where the suppression of torque-ripple was Manuscript received June 15, 1992; revised September 25, 1992. Paper the major objective. Some nonlinear effects can be neglected recommended by Associate Editor, Hans Geering. This work was supported by Aerotech, Inc., by the Center for Motion Control Research, and by Central because of low speed. Sepe and Lang [14] considered an Research Development of the University of Pittsburgh. adaptive velocity controller for synchronous machines, but M. Bodson is with the Department of Electrical and Computer Engineering, limited the speed below those requiring field-weakening, that Camegie Mellon University, Pittsburgh, PA 152 13. J. N. Chiasson and R. T. Novotnak are with the Department of Electrical is, the direct current z d was set to zero which simplifies the Engineering, University of Pittsburgh, Pittsburgh, PA 15261. dynamics. Jahns [ 151 considered field-weakening for extended R. B. Rekowski is with Aerotech Inc., Pittsburgh, PA 15238. speed range with constant power operation. In that work, IEEE Log Number 9205959.

Abstract-The permanent magnet stepper motor is considered

for use in high-performance positioning systems. A model-based

1063-6536/93$03.00 0 1993 IEEE

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 1, NO. 1, MARCH 1993

current control was employed rather than voltage control which is used in this work. The synchronous motor considered .in [I51 had four poles while the stepper motor used in this work has 50 poles which makes voltage control an appropriate choice. In the work presented here, the requirement is to achieve high-speed point-to-point tracking for a positioning system. We show how fast and precise point-to-point tracking can be obtained by using field-weakening, a speed observer, reference trajectory, and a nonlinear controller. In [ 161, the feedback linearizing algorithm was used on an experimental setup similar to the one discussed in this paper except that linear amplifiers and a 50 000 line encoder were used. In [ 171 a shortened version of this paper was presented. ,The rest of this paper is organized as follows. Section I1 presents the mathematical model of the stepper motor and develops a basic feedback linearization controller. Using the analysis of Section 11, Section I11 presents the complete state feedback controller design and develops a speed observer based on the current and position measurements. Section IV describes the experimental setup while Section V gives the experimental results of implementing the controller described in Section 111. Finally, conclusions are summarized in Section VI.
Fig. 1.

W
Cutaway view of the PM stepper motor

11. MATHEMATICAL MODELAND FEEDBACK LINEARIZATION

A. Mathematical Model

TABLE I STEPPER MOTORPARAMETERS Parameter Value


1.5 mH

The PM stepper motor consists of a slotted stator with two phases and a permanent magnet rotor as shown in Fig. 1. One side of the rotor is a north pole and the other side is a south pole. The teeth on each side of the rotor are out of alignment by a tooth-width. A nice qualitative description of the operation of the stepper motor is given in [18]. The mathematical model for the PM stepper motor is given belok and is documented in [19]-[21].
--

L R I

0.55 R
4.5x10-'

kg.mL

Ii,,,

0.19 N . d A

-\-,
B

so
8.0x10-' N.m.s/rad

di,

dt

[,U,

- Ria

+ K,wsin(N,B)]/L
+ Kmibcos(N,B)
-

dW
-

dt d6' -=w dt

= [-K&

sin(Nr6')

Bw]/J

where U,, 2)b and ,z ab are the voltages and currents in phases A and B, respectively. Further, w is the rotor (angular) speed, 6' is the rotor (angular) position, B is the viscous friction coefficient, J is the inertia of the motor and translation table, K,, is the motor torque constant, R is the resistance of the phase winding, L is the inductance of the phase winding, and N , is the number of rotor teeth. The model was validated and its parameters were determined using a least-squares identification procedure. Table I gives the results obtained with our experimental setup (cf.

modeled), and the variation in inductance due to magnetic saturation. The identification work done by [ 7 ] [ 101 suggests that the model (1) is sufficient for control design. Note that the model (1) is nonlinear, due to the sinusoidal functions and their multiplication with state variables. The sinusoidal terms vary at the electrical frequency of the motor which, for a 50 pole stepper motor, is 50 times the mechanical frequency, i.e., w, = N,w,,. As discussed later, this large difference between the electrical and mechanical bandwidth has a significant impact on the control design.

B. Feedback Linearization

[71-[101).
The model neglects the slight magnetic coupling between the phases, the small change in inductance as a function of the rotor position, the detent torque (see [13] where this is

The point of feedback linearization control is to find a (nonlinear) state-space transformation such that, in the new coordinates, the nonlinearities may be canceled out by state feedback. For the PM stepper motor, the appropriate nonlinear coordinate transformation is known as the direct-quadrature (DQ) transformation [4]-[6], [22]. The DQ transformation for

-1

>

BODSON er.&.: HIGH-PERFORMANCE NONLINEAR FEEDBACK CONTROL

the phase voltages and currents is defined as follows:

C. Design of the Current References


(2) While the linearized system is decoupled, constraints on the magnitude of the voltages introduce couplings. To achieve a desired (or reference) speed w, and acceleration a,, the required quadrature current is found from (8) to be

[tt] A [ [ [
id

COS(N,B) - sin(Nr8) cos(N,8)


- sin(Nr8)

sin(Nr8) cos(Nr8)] sin(Nr8) cos(Nr8)]

[]:

iq] =

[: I

(3)

The direct current i d corresponds to the component of the stator magnetic field along the axis of the rotor magnetic field, while the quadrature current i, corresponds to the orthogonal component. The application of the DQ transformation to the original system (1) yields the following system of equations:

d8 =w dt

(4)

where Vd is the direct voltage, vq is the quadrature voltage, i d is the direct current, i, is the quadrature current, w is the angular velocity, and 8 is the angular position. Although the resulting system (4) is still nonlinear, the nonlinear terms can now be canceled by state feedback. Another advantage of this coordinate transformation is that the transformed currents i d , i, now vary approximately at the mechanical frequency of the motor. This enables the controller to be operated at a much lower frequency, since the mechanical variables typically have bandwidths in the range of 0-100 Hz compared to 0-5 kHz bandwidth for U,, V b , i,, and ib. The form of system (4) suggests choosing V d and U, to be of The following form:

The quadrature component i, of the current produces torque while the direct component i d does not produce any torque. However, in order to attain higher speeds, it is necessary to apply a negative direct current to cancel the effect of the backemf of the motor. Specifically, note from the second equation of (4),the back-emf term in the DQ coordinates is K,w. With K, taken to be 0.2, the back-emf voltage at a speed of 2000 rpm is 41 V and a decoupling control law as in (5) requires that U, cancel this back-emf term. For our setup, the source voltage is 40 V, so that cancellation of the back-emf would lead to saturation at the input. However, if i d were forced to be negative (field-weakening) by the correct choice of Ud, the term N,Lwid would help to cancel the back-emf term K , w [see (4) or (5)]. The design of an appropriate reference for i d is thus essential to avoid saturation of the phase voltages at high rotor speeds. The desired direct current i d r is found by maximizing the torque (i.e., i,) at constant speed subject to the constraint vi = U : vt 5 V 2 where V is fixed [ 5 ] . The actual physical constraints for U,, V b are of the form I5 vmazl lVbl 5 U , , , rather than U, 21; 5 V 2 . However, the latter constraint leads to a simple analytical expression for id,. Using the standard Lagrange multiplier technique, the maximum is found to correspond to

Iv,

-(U)

vd

== --

N,wL

U9

(1 1)

idr(w) = - R 2

NrLKmw2 (N,WL)~

(12)

The system then becomes

(7)

The relationship (1 1) is often referred to as the optimal leadangle [ 191-[21]. The inverse tangent of ( I 1) yields the angle advance that the phase voltages U,, Vb need to be commanded relative to the angle NrO (corresponding to the position of the rotor). The equivalent relationship (12) is not usually derived in the literature, but is better suited to our purposes. Although (1 1) and (12) were derived assuming constant speed, it has been found by Brown et al. [23] that the time optimal control (point-to-point, final speed unconstrained) of a stepper motor was only 1% better than the optimal lead-angle control (1 1) and (12).
111. FEEDBACK CONTROLLER DEVELOPMENT

Note that the original fourth-order system has been transformed into a first-order linear system (6), decoupled from a third-order linear system (7)-(9). Therefore, linear control techniques can be used for the system in the new variables (6), (7). This is discussed in Section 111. Note that, while the controller has been presented in a simple and intuitive manner, it can be shown to be a special example of feedback linearization theory [4].

The feedback linearization theory leaves open the design of the feedback controller in the new state and control variables. The actual design of the feedback controller is, of course, heavily influenced by implementation issues. Typically, pulse width modulated (PWM) amplifiers are used for actuation because of their high efficiency. Due to the switching nature

>

2 . 8

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. I . NO. 1, MARCH 1993

of the PWM, there is a high frequency signal (E 20 kHz in our setup) contained in the measurements of the phase currents i, and ib. To remove this noise, analog filters (e.g., Butterworth filters with a 5 kHz cutoff frequency) may be added to filter the currents before sampling. However, these filters add significant phase shifts to the current measurements which then affects the calculated currents i d and i, by introducing a cross-coupling between the variables [24]. In fact, simulations in [24] have sHown that these phase shifts distort the DQ transformation to the extent that the exact linearization controller will not work. A solution to the current measurement noise is to apply the DQ transformation on the noisy currents and then filter them on the digital signal processor (DSP). At any constant speed w, the DQ transformation (2) transforms the signal components of the currents i, and i b at frequency N,w to the dc components of i d and i,. The large harmonics at 20 kHz on i, and i b (due to the PWM amplifier) are transformed by (2) to signals of frequencies 20 k H z h N T w / ( 2 7 r ) . If the motor is run below 4000 rpm, the maximum of N,w/(27r) is 3.3 kHz. Even if the motor is not going at constant speed, the required bandwidth for i d and i, is below 200 Hz. Consequently, the bandwidth required of a filter to remove the high-frequency signals and leave the desired lowTfrequency signal is greatly reduced by filtering in the DQ system. The transformed currents i d and i 4 were computed and filtered on the DSP at SO kHz using low pass second-order Butterworth filters. The filtered currents were then down sampled to 10 kHz. This approach was followed but, in the end, it was found that the anti-aliasing filters could be eliminated, and the high-frequency component on the current measurements has a negligible effect on the commanded (control) voltages.

found so as to satisfy (4), that is

-=

di,,

dt

( I ) ,, Riq, - ~li,w,L,i,j,. - K,,w,)/L

(14)

-=

d0,. dt

w,.

The desired quadrature current was found by solving ( 1 5 ) for i,, [see (lo)] and its derivative di,,/dt is then found by differentiating this expression. The desired direct current id,. was defined by (12) and its derivative is found by simply differentiating this expression. Doing these calculations gives

,. corresponding to 0,. The reference trajectories for i,!, and i and w, as given above are shown as dashed lines in Figs. 8 and 9 cited in Section V. Solving (1 3)-( 16) for 'iid, and ~ i results in

'?Iyr

= L-

(l&
dl

+ Ri,,. + N,.w,.Li,l,. + K,,,w,.

(22)

A . Reference Trajectov

An application for a positioning system was suggested by the Aerotech Corporation of Pittsburgh. The requirement was to have the motor move (point-to-point) a small linear positioning table 1.8 mm under 30 ms while minimizing the vibrations excited in the table. It is easy to show that a permanent magnet dc brush motor [25] with a torque constant of KT = 0.2 (K,,, = 0.1'3 for the stepper motor used here), inertia of .I = 3.6 x lo-' (the inertia is much larger than the stepper motor due to windings on the rotor), voltage ceiling limits of 1 8 0 V, and a peak current limit of 22 A is unable to make such a move under 30 ms. This is due to the fact that field weakening is not possible. However, simulations showed that a stepper motor modeled by ( I ) with the parameters given in Table I, voltage limits of f40 V, and current limits of 6 A can indeed make such a move. The specified position Of., speed w,. = dH,./tlt, and acceleration my,. = dw,/dt are chosen to make the table move 1.8 mm ( 0 . 9 ~ radian turn of the motor, the motor being attached to the linear stage through a ball screw); 0,. and w, are shown as dashed lines in Figs. 6 and 7, respectively. After specifying adesired reference trajectory [H,. w,. CY,. 1' the corresponding state trajectory [id,.. i,,.. w,. 8, 1' and reference input voltages lid,, TI,, were chosen. They were

where the expressions in (17)-(20) are substituted into their respective places on the right-hand sides of (21) and (22). Simulation has shown that these reference input voltages ('((, and 7 i y r do not saturate the amplifier.
B. State Feedback Controller (SFC)

We assume that the current, speed, and position measurements are available and use them for feedback by letting
I,,/
/I(/

-N,.wLi,

= N,.dLi,/

+ +

I,(/,.

U,,, -

+ N,.w,.L%,,. + LU,/ lV,W,Lid, + Lw,

(23)
(24)

where U,!. i i y are yet to be defined. Substituting (23) and (24) for !(I,/ and 'oy in (4) results in the following:
-=

di (j

(it

( L / / (/ Xi,/

+
-

(ti, = ( L / L, Xi, dt

+ AV,wrLi,,,.)/L (25) K,,,w + i i y r , - N r ~ , L i d , . ) / L (26)


o(/~.

Bw)/.J

Equations (23) and (24) contain the nonlinear cancellation terms, the desired reference voltages, and the new inputs 'U,!

BODSON et al.: HIGH-PERFORMANCE NONLINEAR FEEDBACK CONTROL

z
and uqwhich are used to stabilize the error system as is now shown. The tracking error is defined as
-5

A =

[id

- i d r , zq - Zqrr w -

6 - 6'rl

[I'

(e( t ) -Or ( t ) ) d t . The integrator is added to the where [ controller to eliminate any steady-state error due to constant . disturbances. Subtracting the reference trajectory given by (1 3)-( 16) from the system equations in (25)-(28) results in the following linear error dynamics:

Ji

E=

-R/L
0 0 0

Km/J
0 0

-B/J
0 -0 0 0,

or, more compactly


i = AE

+ Bu

(29)

with the obvious definitions for A and B. Through the use of a nonlinear state transformation, input transformation, and nonlinear feedback, a linear time-invariant error system has been obtained. The feedback is defined by
U

This observer computation, that is, the integration of (30), (31) (as well as the control algorithm) is done every 0.1 ms (10 kHz). As shown in Section V, the observer performed well and was essential in getting the tracking controller to work properly. In fact, we were unable to get a speed estimator based on w ( k T ) = [6'(kT)- O((k - l)T)]/T to result in closedloop tracking. As the velocity is found by differentiating the position, the resolution of the velocity depends on the sample rate as well as the resolution of the encoder. To understand this, note that a 2000 line encoder gives a minimum velocity resolution of (27~/2000)/T which, for T = 1/10 kHz, gives 31.4 rad& or 300 rpm! Since the desired speed trajectory requires the motor to go from zero to 141 rad& (1350 rpm) in 10 ms, a sample rate of 10 kHz results in 100 iterations through the control loop. On the other hand, a sample rate of 1 kHz would give a speed resolution of 3.14 rads/s, but only 10 control loop iterations! The simple difference algorithm will work if a 50 000 line encoder is used (see [ 161). However, this is not a practical choice for an industrial setup due to the cost and the fact that such an encoder is limited to motor speeds under 1200 rpm to work properly.

Iv. EXPERIMENTAL SETUP


The hardware used to implement the feedback controller (designed in Section 111) consists of the following: Motorola's Advanced Development System (ADS) including a Motorola DSP56001 digital signal processor, a DSP extension board, two Aerotech 20 kHz PWM amplifiers, and a PM stepper motor with a 2000 line optical encoder. The Motorola DSP56001 digital signal processor was used to implement the controller due to its computational power: it can execute up to 10.24 million instructions per second. There are typically two clock cycles per instruction; however, moves to extended memory, peripheral readdwrites, and other instructions require more clock cycleshnstruction. Included with the processor board is a development system that allows the user to check and debug software. The processor also contains two external interrupts which are used with an external cbunter to set the sample rate. In order to measure the phase voltages, phase currents, and position, a data acquisition board was built by Aerotech, Inc. This board contains four 8-bit A/D converters for sampling the voltages and currents, two 12-bit D/A converters to output the control voltages, and a timer chip that is used to set the sample rate and to count the optical encoder pulses. The stepper motor used for the experiments has 50 rotor teeth, two stator phases, and the following manufacturer supplied (Eastern Air Devices) specifications: maximum current = 6.0 A, maximum torque = 1.42 N.m, phase resistance = 0.55 R, phase inductance = 1.2 mH, and rotor inertia = 2 . 7 lop5 ~ kg.m2 (does not include translation table). These should be compared to the values found in Table I which were determined using a least-squares identification procedure. In order to obtain position measurements, an optical encoder was attached to the shaft of the motor. The optical encoder outputs two square waves which are out of phase by 90". The square waves are used as inputs to a programmable logic

= -KE

where K is defined to have the form


0 0
k22
k23

0
k24
k:5]'

As the system is controllable, Ackerman's [26] formula was used to find the gain matrix K that places the poles of the closed-loop error system.
C. 'speed Observer

The above controller design assumed full state measurements. However, the hardware system used for the experimental work provides only the current measurements and the position measurements. In order to estimate the speed from the position (optical encoder) and current measurements, a reduced-order observer [27] was implemented as follows:
11(6 dt dl3 - = (Km/J)Zq - (B/J)h dt
de = l3

e)

+ /2(6' - S)

(31)

where 11, 12 are the observer gains and i, is a known input to the observer since ,z , zb, and 6' are measured. Subtracting a this from the last two equations of (4) with el = 6' - 8, A e2 = w - l3 results in (32) (33) The observer gains were chosen as 11 = 5272 and 12 = 7 . 0 lo6, ~ resulting in closed-loop poles at s = -2646, -2646.

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. I , NO. 1, MARCH 1993

F l y -w
lxQm

Fig 2. System block diagram

Dssirrd Dirrsf Voltage. 1.8 mm Move. 1350 RPM 20 I 5 5

Desirsd Qumdrerurr C u m n t . 1 . 8 mm Move. 1350 RPM

J
0
0.005

0.01

0.015

0.02
Time (s)

0.025

0.03

0.035

0.04

nmc

Fig. 3.

t.<ir

for the 1.8 mm move.

Fig. 4.

i4,

for the 1.8 mm move

array (PLA) to determine which direction the motor is moving. The PLA contains two outputs that represent clockwise and counterclockwise movement. V. RESULTS A block diagram of the feedback controller (SFC) (designed in Section 111) structure is shown in Fig. 2. The control algorithm was tested on two different trajectories: the first one, as previously mentioned, consists of having the table move 1.8 mm under 30 ms. This move represents a very stringent performance requirement. Specifically, the reference input u d l (Fig. 3) reaches below -25 V, leaving less than 15 V for the feedback controller. Furthermore, the current iqr (Fig. 4) is over 4 A (limit of 6 A) in order to achieve the required torque K m l i qThe . second move forces the motor up to a speed of 3000 rpm and brings it back down into position as shown in Fig. 12. This move was chosen to study high-speed control of the stepper motor (well into the field-weakening region) and it also required ' u d r to get close to saturation (Fig. 5). Simulations were performed using SIMNON [ 2 8 ] . Although not given here, results of the simulations demonstrate a close approximation of the real motor by the model used for control design.

'

- 1 0-15-20-25-30-

-35Time
(8)

i
0.14 0.16

Fig. 5.

t'dr

for the 3000 rpm run

Section 111-B. The last part of the position response is shown in more detail in Fig. 10. The desired and estimated (by the observer) speed are shown in Fig. 7. Note the very close tracking of the reference position and velocity. Figs. 8 and 9 show the tracking of the currents in the direct and quadrature axis. Again, the tracking is excellent despite large fluctuations in the measured currents. It was found through simulations that these fluctuations were due to the limited resolution of the position measurement (2000 line encoder), and not to the PWM noise. This was concluded by performing simulations which showed the oscillatory behavior to be present when a 2000 line encoder was modeled, but not when a 50 000 line A. Control o f U 1.8 mm Move encoder was modeled. The oscillatory behavior of Lj at the Fig. 6 gives the actual and desired position response for the end of the run in Fig. 7 is a direct result of integrating the 1.8 mm move of the positioning table (corresponding to a 0 . 9 ~ quadrature current z , and thus is due to the oscillatory behavior radian rotation of the motor) using the feedback controller of of i,, i.e., the noise due to quantization of the position 0.

>

BODSON et al.: HIGH-PERFORMANCE NONLINEAR FEEDBACK CONTROL

1.8 _n More. 1350RPM

1.8

Move. 1350 RPM

2.5

901.5

'%b3

0.031

0.032

0.033

0.034

0.035

0.036

0.037

0.038

0.039

0.A

Time (s)

Time ( . )

Fig. 6. Desired and actual position for the 1.8 mm move.

Fig. 10. Position response in counts after 30 ms. Desired position is 900 counts (= 900/2000(2a) = 0 . 9 ~ ) .

1.8 mm Move. 1350 RPM


160

-20 I 0

0.005

0.01 '

0.015

0.02
Time (a)

0.W

I
0.03
0.035

0.04

Fig. 7. Desired and estimated speed for the 1.8 mm move.


Time ( s )

Fig. 11.
1.8 _D Move. 1310 RPM

Desired and actual position for the 3000 rpm run.

I,

'0

0 . 0 0 1

0.01

0.015

0.02

0.025

0.03

0.035

0.04

not indicate that the controller is not maintaining the correct final position. The oscillations in i d (Fig. 8) are also due to the encoder. In the implementation, the gains for the 1.8 mm move were set at k l l = 1.06 x lo4, k22 = 1.78 x lo4, k23 = 3.40 x lo2, = 3.80 x lo6, and kZ5 = 1.07 x lo8 to place the closedloop poles at s = -18178 for the first-order system and at s = -11047, -28.3, -54.86 & j1190.7 for the fourth-order system.

Time ( . )

Fig. 8. Desired and actual direct current for the 1.8 mm move.

B. Control o f a 3000 rpm Run


I

1.8 n_ Move. 1350 RPM


6

-.6

1' 'I
0 . k

1
0.03
0.035
0 . i

0.WS

0.01

0.01s

0.02

Tim- ( . )

Fig. 9. Desired and actual quadrature current for the 1.8 mm move.

Fig. 10 (note the units are in counts not radians) shows that the motor stays within one encoder count of the final position after 30 ms which is the highest accuracy the controller can achieve with any given encoder. Consequently, the estimate of speed 3 is responding to the noisy oscillations in i, and does

Fig. 11 shows the actual and desired (reference) position using the feedback controller to bring the motor up to 3000 rpm. The actual and estimated speed is shown in Fig. 12. Again, the excellent tracking is observed despite oscillatory behavior in i d and i, (see Figs. 13 and 14). These oscillations are again attributed to the lack of resolution in the 2000 line encoder. Note also that the reference input Wdr (Fig. 5 ) reaches -30 V leaving less than 10 V for the feedback controller. In the implementation, the gains for the 3000 rpm run were set at k l l = 2.5871 x lo4, k22 = 2.4465 x lo4, k23 = 6.0314 x lo2, k24 = 2.9352 x lo6, and kZ5 = 1.0042 x lo7 to place the closed loop poles at s = -26314 first-order system and at s = -24727, -3.4, -101 f j893 for the fourth-order system. In the course of our investigations, we also considered a feedback controller based only on the optical encoder measurements, i.e., without using current measurements [ 17],[24],[29],[30]. This controller represents what we felt to be the best possible implementation of the standard approach

I?

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. I . NO. 1, MARCH 1993

3000 RPM

I
0
002 0.04 0.06

0.08
Time (s)

0.1

0.12

0.14

0.16

Fig. 12. Desired and estimated speed for the 3000 rpm run.

3000 RPM

I
6O 0.02 0 . 0 4 0.06

0.08
Tim= (s)

II 0 . 1

,
0.12 0.14

0.16

Fig. 13. Desired and actual direct current for the 3ooO rpm run.

3000 RPM 5

-3 0

002 0.04

0.06

0.08
Time (a)

0.1

0.12

0.14

I 0.16

Fig. 14. Desired and actual quadrature current for the 3000 rpm run.

to position feedback control of a stepper motor as outlined in [20] and [21]. It required using the simple difference algorithm w ( k T ) = [ Q ( k T) O(k(T - l ) ) ] / Tfor speed estimation since the speed observer requires current measurements. The controller was unable to make the 1.8 mm move in 30 ms. Using the current measurements for the speed observer only, it was possible to make this move, but the tracking performance was significantly deteriorated compared to that of the original controller [16], [17], [29]. Of course, it makes little sense to measure the currents and then use them only for the speed estimate. However, it shows the current measurements are required for feedback as well as for the speed observer in order to achieve high performance.

VI. CONCLUSIONS
This paper has illustrated how modern theories of nonlinear control can be applied to problems of current industrial interest. We considered the tracking of trajectories that pushed the

system to the limits of performance, in particular to high-speed regimes where nonlinear terms could not be neglected. As the experimental results showed, very good tracking performance was obtained. The results having been obtained with an experimental setup that is only slightly modified from an existing motion control product, they bear direct relevance to current industrial practice. The feedback linearization method provided a framework for the design of a high performance controller for the PM stepper motor. However, practical issues in implementation needed to be addressed and should not be underestimated. The issues encountered in this application were generic to some extent and some observations might be useful. The first problem that was discussed was that of input saturation. Specifically, the design of the trajectory for the state variable i d required taking into account the saturation of the phase voltages. Input saturation is, of course, a problem that is also present in linear controller designs. However, for nonlinear systems the transformation of input variables that is required for linearization will generally mix up the constraints. Specifically, while the linearized system was a decoupled multivariable system, the inputs remained coupled through the true input constraints. Fortunately, we could resolve this problem rather simply, adapting some standard concepts of field-weakening in motor control. Note, however, that one might expect input saturation to be an important problem to address in general, as inputs are required to cancel nonlinear terms which may typically be large. Another problem that we encountered was that of sensor noise. In this application, noise was present due to the pulsewidth-modulated signals, but, to our surprise, an even greater source of noise was due to the quantization of the position measurement. Again, noise must be taken into consideration in linear control design as well, but it takes new dimensions in nonlinear control, because of the transformations that must be performed to obtain a linear system. For example, quantization noise in the position measurement introduced large errors on the transformed currents in the DQ coordinates, since the transformation uses the position measurement. Also, phase detays in the current measurements due to the analog anti-aliasing filters created cross-coupling in the DQ current variables. In other words, small additive noise in some variables could become large errors in the transformed variables and, of course, need not be additive. In summary, modem theories for nonlinear control can be successfully applied in an industrial environment. Practical issues should not be underestimated and a good design must incorporate the usual constraints of input saturation and output noise. As systems are pushed to their limits of performance, it can be expected that nonlinear effects will often require that the controllers become themselves nonlinear. Recent nonlinear control theory constitutes a source of inspiration to develop such control algorithms, and these may turn out to be quite different from the algorithms used previously. While we discussed the application to the control of permanent magnet stepper motors, several other problems, such as brushless dc motor control [31], [32] and induction motor control [33], 1341, have been discussed in the literature. In some cases,

BODSON e i a l . : HIGH-PERFORMANCE NONLINEAR FEEDBACK CONTROL

13

the controllers are radically different from existing controllers (cf. [34]) and the success encountered in this paper should be an encouragement for experimental investigation of such problems as well.
ACKNOWLEDGMENT

The authors would like to thank Aerotech, Inc. for supplying the hardware used to carry out this research. In particular, a very special thanks to M. Aiello, Industrial Controls Manager of Aerotech, for his help with building and debugging the hardware. The support of the Center for Motion Control Research, under the direction of Prof. E. W. Kamen, is also gratefully acknowledged. The authors would also like to take this opportunity to extend their sincere appreciation to S. Botos, President of Aerotech, Inc., for his encouragement of Universityflndustry cooperation. Finally, the Motorola Corporation is gratefully acknowledged for donating the DSP56001 Development System used in this research.

REFERENCES
A. Isidori, Nonlinear Control Systems. New York: Springer-Verlag. 2nd ed., 1989. H. Nijmeijer and A. J. van der Schaft, Nonlinear Dynamical Control Systems. New York: Springer-Verlag, 1990. C. I. Bymes, A. Isidori, and J. C. Willems, Passivity, feedback equivalence, and the global stabilization of minimum phase nonlinear systems, IEEE Trans. Automat. Contr., vol. 36, no. 1 I , pp. 1228-1240, Nov. 1991. M. Zribi and J. Chiasson, Position control of a pm stepper motor by exact linearization, IEEE Trans. Automat. Contr., vol. 36, no. 5 , pp. 620-625, May 1991. M. Bodson and J. Chiasson, Application of nonlinear control methods to the positioning of a permanent magnet stepper motor, in Proc. 28th IEEE Conf: Decision Conrr., Tampa, FL, Dec. 1989, pp. 531-532. X. Z. Liu, G. C. Verghese, J. H. Lang, and M. K. Onder, Generalizing the Blondel-Park transformation of electrical machines: Necessary and sufficient conditions, IEEE Trans. Circuits Syst., vol. 36, no. 8, pp. 1058-1067, Aug. 1989. G. Campbell, On line parameter identification of a pm stepper motor, M.S. thesis, Univ. Pittsburgh, 1991. D. Schuerer, Parameter identification of a permanent magnet stepping motor, M.S. thesis, Univ. Pittsburgh, 1990. A. Blauch, High-speed parameter estimation of step motors, M.S. thesis, Dep. Elect. Comput. Eng., Carnegie Mellon Univ., May 1992. A. Blauch, M. Bodson, and J. Chiasson, High-speed parameter estimation of step motors, submitted for publication. M. Ilic-Spong, R. Marino, S. M. Persada, and D. G. Taylor, Feedback linearizing control of switched reluctance motors, IEEE Trans. Automat. Contr., vol. AC-32, no. 5 , pp. 371-379, May 1987. D. G. Taylor, M. J. Woolley, and M. Ilic, Design and implementation of a linearizing and decoupling feedback transformation for switched reluctance motors. in Proc. 17th Symp. Incremental Motion Contr. Syst. Devices, 1988. D. Chen and B. Paden, Nonlinear adaptive torque-ripple cancellation for step motors, in Proc. 29th IEEE Cont Decision Contr., Honolulu, HI, Dec. 1990, pp. 3319-3324. R. B. Sepe and J. H. Lang, Real-time adaptive control of the permanentmagnet synchronous motor, IEEE Trans. Industry Appl., vol. 27, no. 4, JulylAug. 1991. T. M. Jahns, Flux-weakening regime operation of an interior permanent-magnet synchronous motor drive, IEEE Trans. Industry Appl., vol. IA-23, no. 4, July/Aug. 1987. M. Bodson, J. Chiasson, R. Novotnak, and R. Rekowski, A state feedback tracking controller for a permanent magnet stepper motor, in Proc. Zlst Annual Symp. Incremental Motion Contr. Syst. Devices, San Jose, CA, June 1992. High M. Bodson, J. Chiasson, R. Novotnak, and R. Rekowski, performance nonlinear feedback control of a permanent magnet stepping Control Appl., Dayton, OH, Sept. 1992. motor, in Proc. 1st IEEE Conf:

V. Deltoro, Electric Machines and Power Systems. Englewood Cliffs, NJ: Prentice-Hall, 1985. B. C. Kuo and J. Tal, Incremental Motion Step Motors and Control Systems, Volume 11. SRL Publishing, 1978. P. P. Acarnley, Stepping Motors: A Guide to Modem Theory and Practice. New York: Peregrinus, 1982. T. Kenjo, Stepping Motors and Their Microprocessor Control. New York: Clarendon, 1984. P. C. Krause and 0. Wasynczuk, Elecrromechanical Motion Devices. New York: McGraw-Hill, 1989. R. H. Brown, Y. Zhu, and X. Feng, A new relaxation algorithm for the time optimal control problems of hybrid step motors, in Proc. 28th IEEE Con5 Decision Contr., Tampa, FL, Dec. 1989, pp. 907-908. R. Rekowski, Implementation of an exact linearization controller for a permanent magnet stepping motor using only optical encoder feedback, Masters thesis, Univ. Pittsburgh, 1990. Motion Control Product Guide. Aerotech, Inc., Pittsburgh, PA, p. 98. G. F. Franklin, J. D. Powell, and M. L. Workman, Digital Control of Dynamic Systems. Reading, MA: Addison-Wesley, 2nd ed., 1990. J. Chiasson and R. Novotnak, Nonlinear speed observer for the pm stepper motor, IEEE Trans. Auromat. Contr., to be published. E. Hilding, K. J. Astrom, T. Schonthal, and B. Wittenmark, Simnon Users Guide for MS-DOS Computers. SSPA Systems, 3rd. ed., 1986. R. Novotnak, A state feedback tracking controller for a pm stepper motor, M.S. thesis, Univ. Pittsburgh, 1991. M. Aiello, R. Rekowski, M. Bodson, J. Chiasson, and D. Schuerer, Experimental results of parameter identification and nonlinear control of a PM stepper motor, in Microprocessors in Robotic and Manufacturing Systems. Spyros G. Tzafestas, Ed. New York: Kluwer. N. Hemati and M. C. Leu, Nonlinear tracking control of brushless dc motors for high-performance applications, in Proc. 28th Conf: Decision Contr., Tampa, FL, Dec. 1989, pp. 527-530. N. Hemati, J. S. Thorp, and M. C. Leu, Robust nonlinear control of brushless dc motors for direct-drive robotic applications, IEEE Trans. Industrial Electron., vol. 37, no. 6, Dec. 1990. Z. Krzeminski, Nonlinear control of the induction motor, in 10th IFAC World Congress, Munich, 1987, pp. 349-354. R. Marino, S. Peresada, and P. Valigi, Adaptive partial feedback linearization of induction motors, in Proc. 29th Conf: Decision Contr., Honolulu, HI, Dec. 1990.

Marc Bodson (S82-M86) received the Ph.D. degree in electrical engineering and computer science from the University of Califomia, Berkeley, in 1986, the M.S. degree in electrical engineering and computer science and the M.S. degree in aeronautics and astronautics from the Massachusetts Institute of Technology, Cambridge, in 1982, and the Electrical and Mechanical Engineer degree from the Universite Libre de Bruxelles, Belgium, in I 9 8 8 Currently he is an Assistant Professor in the Department of Electrical and Computer Engineering, Camegie-Melon University, Pittsburgh, PA. He is coauthor with S. Sastry of the book Adaptive Control: Srability, Convergence, and Robustness (Englewood Cliffs, NJ: Prentice-Hall, 1989). Dr. Bodson was a Belgian American Educational Foundation Fellow in 1980, a Visiting Lecturer at the University of Califomia, Berkeley, in 1987, and a Lady Davis Fellow at the Technion -Israel Insitute of Technology, Haifa, Israel, in 1990.

John N. Chiasson (S82-M84) received the B.S. degree in mathematics from the University of Arizona, Tucson, the M.S. degree in electrical engineering from Washington State University, Pullman, and the Ph.D. degree in control sciences from the University of Minnesota. Since 1988 he has been an Assistant Professor in the Department of Electrical Engineering, University of Pittsburgh, Pittsburgh, PA. His current area of research interest I S in the applications of geometric nonlinear control theory to electromechanical systems

14

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. I , NO. 1, MARCH 1993

Robert T. Novotnak (SM92) received the B.S. and M.S. degrees in electrical engineering from the University of Pittsburgh, Pittsburgh, PA, in 1989 and 1991, respectively. He is currently enrolled in the Ph.D. program at the University of Pittsburgh. His interests include the practical issues involved in implementing nonlinear control algorithms for electrical drives.

Ronald B. Rekowski received the undergraduate and graduate degrees in electrical engineering from the University of Pittsburgh, Pittsburgh, PA, in 1988 and 1990, respectively. He is currently Chief Engineer of the Systems Engineering Department at Aerotech Inc., Pittsburgh, PA, where he oversees the application of modern control theory to high performance motion control systems.

Вам также может понравиться