Speed Nonlinear Control of DC Motor Drive with Field Weakening

Zuo Zong Liu and Fang Lin Luo, Senior Member IEEE Muhammad H. Rashid, Fellow IEEE
Nanyang Technological University University of West Florida
Nanyang Avenue, Singapore 639798 Pensacola, Florida 32514-5754, USA

voltage as the control input has one stable equilibrium

Abstract-In this paper, we show that it is feasible to apply point. Unfortunately, the armature current drawn at this
nonlinear MIMO feedback linearization technique to a point has been found to be much larger than the nominal
separately excited DC motor system that is operated in high- armature current thus this method has not been considered
speed field-weakening regime. Load-adaptive and sensorless as a feasible controller candidate.
control techniques to improve dynamic speed performance
are proposed and compared. Also, application results are
In this paper, several nonlinear field weakening control
presented to verify theoretical ones.
techniques for speed tracking of a SEDCM system are
investigated. We first show it is feasible to apply nonlinear
I. INTRODUCTION
MIMO feedback linearization technique to such a system
that is operated in high-speed field-weakening regimes. To
Together with the invent of low cost and reliable induction improve speed performance in the presence of load torque
motors and their power semiconductor converter based disturbance, a load-adaptive control design is then
power supply equipped with field-oriented type of proposed, which can be implemented at any desired field
controller, it was believed that DC motors would become weakening point of operation. Control scheme with speed
obsolete in industry applications. However in contrary to and load torque observer is further, in particular, presented
popular belief, DC motor drives are still widely used many for speed sensorless operation. Finally, application
industries such as rolling mills, paper machines, unwinding examples are shown to verify theoretical ones.
and rewinding machines [1]-[3]. In a separately excited DC
motor (SEDCM) drive system, linear control techniques II. DYNAMICAL SYSTEM MODELING
are easily applied to the system represented by linear
equations in the armature control region. However, system A SEDCM system is characterized by the three
nonlinearities begin to appear once the motor is operated in interconnected differential equations as follows [1]
the field weakening region due to the electromagnetic
torque being a product of field flux and armature current, di r 1
the back emf being a product of field flux and speed, and = ( u r − R r ir − E )
dt Lr
magnetic saturation. Traditional way of circumventing (1)
di s 1
such nonlinearities is by linearizing the system equations = ( u s − R s is )
dt Ls
around an operating point and designing linear controllers
dω 1
based on the linearized system equations. Recent advances = (Te − T L − B ω )
in nonlinear control systems, however, resulted in the dt J
development of more sophisticated controllers based on the
feedback linearization techniques [4]-[8]. Regardless of where Te = Ki s i r is the developed torque and E = K i s ω
being computing intensive in real time control, the the back EMF. u r and u s are armature and field voltages,
applicability of these techniques has been justified i r and i s armature and field currents, respectively. R r
particularly in motion control. This factor explains the
surge of research interest in nonlinear control applicable to and R s denote armature and field resistances, Lr and L s
motion control especially where a high dynamic armature and field inductances, K , B and J back EMF
performance is required in the closed loop control design constant (or torque constant), viscous damping coefficient
of motor drives. This paper has contributed to carrying out and inertia, respectively.
nonlinear speed control design of a SEDCM drive in the
field weakening mode. To facilitate the controller design, model (1) is rewritten in
the compact state-space form as follows
It is shown that the single-input single-output (SISO)
feedback linearization for the motor system taking the field x! = f ( x) + u r g r + u s g s + dm (2)

0-7803-7116-X/01/$10.00 (C) 2001 IEEE

1
541

Authorized licensed use limited to: UNIVERSIDAD AUTONOMA DE BUCARAMANGA (UNAB). Downloaded on February 16,2021 at 15:08:07 UTC from IEEE Xplore. Restrictions apply.
which is one to one in s ={ x ∈ R 3 : i(2)
s ≠ 0 and ω ≠ 0 }.
where x = [i r is ω ] is the state vector. f (x) is the
T
And its inverse transformation is
function of i r , i s , and ω ,
η 2 
 η ( Jη 3 + B η 2 + T L ) 
 1  i r   1 
 L (− R r i r − Ki s ω )  i  =  η1 
 r   s  (5)
Rs Kη 2 
f ( x) =  − if  ω   
 Ls   η 2 
1   
 ( Ki s i r − Bω − T LN )
 J 
In the new coordinates, the motor dynamics with normal
load torque ( d = 0 ) can be rewritten as
g r = [1 / L r 0 0]T , g s = [0 1 / L s 0]T , and
m = [0 0 − 1 / J ]T are coefficient vectors, T LN is the η!1   L f h1 ( x)  u 
 !!  =  2  + D( x) r  (6)
normal load torque, d = T L − T LN the load torque
η 2   L f h2 ( x) u s 
disturbance.

This model, which is nonlinear because of involving where D(x) , the decoupling matrix, is
product of two variables, will be applied to speed tracking
design in field weakening region. Generally speaking, all  L g r h1 ( x) L g s h1 ( x) 
of the parameters change to some extent with the motor D( x) =  
L L h
 g r f 2 ( x ) L g s L f h2 ( x)

operating conditions. However, we here assume that the
variations of these parameters are negligible for the
controller design except dealing with unknown load torque and the Lie derivatives are expressed as
disturbance.
Rs K
L f h1 = − E + i s (Te − Bω − T L )
III. INPUT –OUTPUT LINEARIZATION Ls J
1
The objective here is to develop a field weakening L f h2 = (Te − Bω − T L )
controller, which can effectively stabilize and track the J
desired speed reference ω ref over the range ω ≥ ω n , and K R B
L2f h 2 = i s (− R r i r − E ) − s Te − 2 (Te − Bω − T LN )
reject load torque disturbance. To do so, the control outputs JLr JL s J
are selected to be h1 ( x) = E and h2 ( x) = ω , and define L g r h1 = 0
K
e E = E − E ref L g s h1 = ω
(3) Ls
eω = ω − ω ref K
L g r L f h2 = is
JL r
where E ref represents the set point of field weakening,
K
which is chosen to be between 0.85 to 0.95 of the rated L g s L f h2 = ir
JL s
armature voltage [1], the reference speed ω ref is required
to be twice differential. Note that D(x) is nonsingular in field weakening regimes
since det D(x) = − Ki s ω / L r L s ≠ 0 holds. Letting
By means of Lie derivative, we introduce the following
change of coordinates
u r  −1  − L f h1 ( x) + v a 
  = D ( x ) − L2 h ( x) + v  (7)
η1  h1 ( x)  

E
 u s   f 2 f 
  
   
η
 2  2= h ( x ) =
  ω (4)
1  and inserting it into (6) result in
η 3   L f h2 ( x)  (Te − Bω − T LN )
  J 

2
542

Authorized licensed use limited to: UNIVERSIDAD AUTONOMA DE BUCARAMANGA (UNAB). Downloaded on February 16,2021 at 15:08:07 UTC from IEEE Xplore. Restrictions apply.
η!1 = v a design to estimation of the unknown disturbance torque
η! 2 = η 3 (8) [10].
η! 3 = v f Define a time-varying change of coordinates with the
estimate of load torque disturbance d̂
where v = v a [ vf ]T is the input vector of the linearized
system of the well-known Brunovski canonical form (8). z1 = η1 = E
The relative degree of the system (1+2=3) is equal to the z2 = η2 = ω (13)
original (no zero dynamics).
z 3 = η 3 + dˆL m h2
To ensure desirable tracking of the reference speed ω ref we then have

and back emf E ref , v a and v f are designed as follows z!1 = L f h1 + dLm h1 + u r L gr h1 + u s Lg s h1
z! 2 = z 3 + ∆dLm h2 (14)
v a = E! ref − k a ( E − E ref )
(9) z! 3 = L2f h2 + dLm L f h2 + dˆLm h2 + u r L gr L f h2 + u s Lg s L f h2
v f = ω!! ref − k f 1 (ω! − ω! ref ) − k f 0 (ω − ω ref )
where ∆d = d − dˆ .

where the parameters k a , k f 1 and k f 2 can be suitably To linearize (14), let the control voltage inputs be as
chosen to make the linear systems follows

e! E + k a e E = 0 u r  −1
 − L f h1 − dˆL m h1 + v r 
!e!ω + k f 1 e!ω + k f 0 eω = 0 (10)   = D ( x)  2  (15)
u s  − L f h2 − dˆL m L f h2 − dˆL m h2 + v s 

asymptotically stable. The selection of these parameters is where

also limited by the motor operation conditions and the v r = −α a z1 + v aref
feasible realization using power electronics.
v s = −α f 0 z 2 − α f 1 z 3 + v fref
To enable the rejection of load torque disturbance d , an and
integrator is inserted into the linear control law as v aref = E! ref + α a E ref
v fref = ω!!ref + α f 1ω! ref + α f 0ω ref
va = E!ref − ka(E − Eref) − ka1∫ (E − Eref) dt
(11)
vf =ω
!!ref − kf 1(ω! −ω!ref) − kf 0(ω −ωref) − kf 2∫ (ω −ωref) dt The Lie derivatives related with load torque disturbance
are
IV. LOAD-ADAPTIVE CONTROL DESIGN K
L m h1 = − i s
J
The motor system with load torque disturbance T L is 1
written as Lm h2 = −
J
B
η!1 = L f h1 + dLm h1 + Lg r h1ur + Lg s h1u s Lm L f h2 = 2
(12) J
η!2 = η3 + dLm h2
η!3 = L2f h2 + dLm L f h2 + Lg r L f h2ur + Lg s L f h2u s where α a , α f 0 and α f 1 are constant gain parameters to be
designed.
It is clear that the complete feedback linearization design
of the model (2) cannot be achieved due to the presence of Using (14) and (15) results in
load torque disturbance. Inserting an integrator in the
control law shown in (11) to enable the rejection of z!1 = ∆dLm h1 − α a z1 + v aref
constant unmeasured disturbance will result in degraded
output response, which is not expected. To overcome these z! 2 = z 3 + ∆dLm h2 (16)
problems, this paper will apply a load-adaptive control z! 3 = ∆dL m L f h2 − α f 0 z 2 − α f 1 z 3 + v fref

3
543

Authorized licensed use limited to: UNIVERSIDAD AUTONOMA DE BUCARAMANGA (UNAB). Downloaded on February 16,2021 at 15:08:07 UTC from IEEE Xplore. Restrictions apply.
 L m h1 
The unknown load torque disturbance d will be estimated  
w( x) =  L m h2 
by using model reference adaptive technique. The
reference model is then selected as follows  Lm L f h2 
 

z! m = Am z m + u (17) Let V be the Lyapunov function given as

where V = e T Pe + λ∆d 2 (19)

z m = [z1m z 2m z 3m ] T

[
u = v ares 0 v fres ]T where λ is a positive number, P is a positive definite
symmetric matrix and the solution of the following
and Lyapunov equation
− α a 0 0 
  AmT P + PAm = −Q (20)
Am =  0 0 1 
 0 α f0 α f 1 
 with Q positive definite symmetric matrix. By making the

By subtracting (17) from (16), the model reference tracking time derivative of V , we have
error dynamics are obtained
V! = −e T Q e + 2 ∆d( wT P e + λ ∆d! ) (21)
e! = Am e + w( x)∆d (18) To make
where V! = −e T Qe (22)

 e1   z1 − z1m  we choose
e = e 2  =  z 2 − z 2 m 
e3   z 3 − z 3m  wT P e + λ ∆d! = 0 (23)
The following adaptation law, which describes the estimate
dynamics of the unknown constant load torque disturbance

Load-adaptive MIMO linearization

vr = −α a z1 + varef vr  − Lf h1 − dˆLmh1 + vr  ur ir , is , ω
D−1 !ˆ  SEDCM
vs = −α f0 z2 −α f1 z3 + v fref vs ˆ
− Lf h2 − dLmLf h2 − dLmh2 + vs 
2 us
 

z1 = η1
z2 = η2
z 3 = η 3 + dˆ" m h2

Eref varef zm !
αa Eref + E!ref e 1 T dˆ !
∫ ( Am zm + u )dt d̂
ω ref α f1ωref + α f2ω! ref + ω!!ref v fref λ
W Pe ∫ dˆ dt

Estimation of load torque disturbance

Fig.1 Load-adaptive nonlinear MIMO speed tracking controller

4
544

Authorized licensed use limited to: UNIVERSIDAD AUTONOMA DE BUCARAMANGA (UNAB). Downloaded on February 16,2021 at 15:08:07 UTC from IEEE Xplore. Restrictions apply.
d̂ , is then derived For i s > 0 , consider the change of coordinates

ζ 1 = ln i s
! 1 ζ2 =ω (26)
dˆ = W T Pe (24)
λ ζ 3 = TL

A Lyapunov function (19) satisfying V! (t ) ≤ 0 has been Using this change of coordinates, the system (25) becomes
found which guarantees that tracking error e and load
torque estimation error ∆d are bounded. Application of R i R K u + us
ζ!1 = − r r − s − z2 + r
Barbalat's lemma allows the further conclusion that e → 0 Ls i s L s L s Ls is
as t → ∞ , i.e. that the tracking error converges to zero Ki i
B
asymptotically. ζ! 2 = − z 2 − z 3 + r s (27)
J J
The obtained field-weakening controller for the speed ζ!3 = 0
control, which is based on nonlinear load-adaptive MIMO
linearization design, is shown in Fig. 1. In this scheme, full which is now linear in the unmeasured state variable
state, i.e. i r , i s , ω , measurements are required. ζ2 =ω .

V. SPEED SENSORLESS CONTROL Such an observer is then designed as

Measuring speed usually needs a mechanical speed sensor, ! K u +u Ri R

which adds to the costs and the number of electrical ζˆ1 = − ζˆ2 + r s − r r − s + l1 (ζ 1 − ζˆ1 )
Ls Lsis Lsis Ls
connections to the motor drive system. Therefore, it would
!ˆ B ˆ K
be desirable to eliminate the need for such a sensor. The ζ 2 = − ζ 2 − ζˆ3 + ir is + l2 (ζ 1 − ζˆ1 ) (28)
nonlinear speed and load torque observer with linear error J J
dynamics is then proposed. !ˆ ˆ
ζ = l (ζ − ζ )
3 3 1 1

To design such an observer via model (2), the key

The observer error system is found by subtracting (28)
approximation made is that L r i r is very small compared to
from (27), i.e.
L s i s . Based on this approximation, an accurate observer
can still be obtained because the ratio L r / L s can be K
e!1 = − e 2 − l1 e1
10−4 [2]. Ls
B
e! 2 = − e 2 − e3 − l 2 e1 (29)
Modeling the effect of the constant load torque TL as the J
additional state variable results in e!3 = −l 3 e1
where
di s 1
= (− R r i r − R s i s − Ki s ω + u r + u s )
dt Ls e1 = ζ 1 − ζˆ1 , e 2 = ζ 2 − ζˆ 2 and e3 = ζ 3 − ζˆ3 .
dω 1
= ( Ki r i s − Bω − T L ) (25)
dt J Choosing the gains l1 , l 2 and l3 as
dT L
=0
dt B
l1 = p1 + p 2 + p 3 −
J
Note that (25) is not linear in the measured state variables Ls B
due to the term K i s ω in the first equation as ω is not l2 = − ( p1 p 2 + p1 p 3 + p 2 p 3 − l1 ) (30)
K J
available. Based on the idea described in [7], we can find a Ls
coordinate transformation ζ = T (x) , which will transform l3 = p1 p 2 p 3
K
(25) to a system which is linear in the unmeasured state
variables. results in the characteristic equation of (29) being given by

5
545

Authorized licensed use limited to: UNIVERSIDAD AUTONOMA DE BUCARAMANGA (UNAB). Downloaded on February 16,2021 at 15:08:07 UTC from IEEE Xplore. Restrictions apply.
B B K K
s3 + ( + l1 ) s 2 + ( l1 − l 2 )s + l3 2400

J J Ls Ls (31) 2300

= ( s + p1 )( s + p 2 )( s + p 3 ) 2200

Speed (rpm)
2100
That is, these gains put the poles of the observer at
− p1 , − p 2 , − p3 , where p1 , p 2 , p 3 are positive. 2000

1900

Such an observer is combined with the nonlinear MIMO 1800

field weakening controller designed in section III to obtain
a sensorless control scheme, where only the current 1700
0 2 4 6 8 10
Time (s)
measurements ( i r , i s ), are needed and the speed feedback (a)
( ω ) and load torque signal ( TL ) are replaced by the 4.2

estimated values ( ω̂ , T̂L ) respectively. 4

3.8
VI. APPLICATIONS

Field current (A)

3.6

To evaluate the performance of the several nonlinear speed 3.4

control schemes proposed above, extensive studies have 3.2

been carried out for a 3.7 kW motor operating in field
weakening regime. The motor parameters are given in 3

Table I. 2.8
0 2 4 6 8 10
Time (s)

TABLE I (b)
PARAMETERS OF SEDCM
60
Rated power 3.7kW Rated voltage 240V
55
Rated speed 1750rpm Rated torque 18Nm
50
Field voltage 240V Armature inductance 0.01H
Developed torque (Nm)

Armature 1.2 Ω Field inductance 60H 45

resistance 40
Field resistance 60Ω Motor constant 0.3Nm/A2 35
Inertia 0.208kgm2 Damping coefficient 0.011kgm2s-1 30

25
The system performance of the nonlinear MIMO control 20
algorithm is shown in Fig. 2. The gain parameters used are 15

k a = 20 , k f 1 = 400 and k f 2 = 40 to place all the closed 10

0 2 4 6 8 10
Time (s)
loop system’s poles at –20. The adaptive gain λ = 3.8 , and (c)
normal load torque T LN = 18 Nm. 60

55
Initially field weakening point is set at E = 220 V, the 50
motor is operated at base speed of 1750 rpm. Then three
Armature current (A)

45
step changes of 250 rpm in a reference speed is applied at t 40
= 2, 4 and 6 second. When the speed ω increases (from 35
1750 rpm to 2350 rpm) the field current i s decreases (from 30

4.0 to 2.98 A), and ω ∝ 1 / i s . The steady state values of 25

transient armature current i r and developed torque Te 20

increase slightly. 15
0 2 4 6 8 10
Time (s)

(d)
In the case of load-adaptive control design, the responses
of the motor speed are reported in Fig. 3. After reaching Fig. 2 System response with normal load toque
speed

6
546

Authorized licensed use limited to: UNIVERSIDAD AUTONOMA DE BUCARAMANGA (UNAB). Downloaded on February 16,2021 at 15:08:07 UTC from IEEE Xplore. Restrictions apply.
2050
to improve the dynamic control performance. The scheme
with the nonlinear speed and load torque observer is
2000
proposed for speed tracking control. The main advantage
1950 of the scheme is that it uses only current measurements,
eliminating costly speed sensors. Because of the
Speed (rpm)

1900
decoupling and linearization, control implementation of the
1850
motor speed ω and back emf E is achieved
1800 independently. As a result, these designs can assure speed
1750
tracking at any desired field point.
1700
0 2 4 6 8 10 Applications to a given specific SEDCM system show the
Time (s)
proposed nonlinear speed control schemes have high
(a)
dynamic tracking performance even when the system with
unknown load torque is operated in wide dynamic regimes
2
of field weakening.
0
REFERENCES
-2
Load disturbance (Nm)

-4
[1] W. Leonhard, “Control of Electrical Drives,” 2nd,
Springer-Verlag, 1985.
-6 [2] M. H. Rashid, “Power Electronics: Circuits,
Devices, and Applications,” Prentice-Hall,
-8
Englewood Cliffs, NJ, 1988.
-10 [3] F. L. Luo, “Conputerized DC Motor Field
0 2 4 6 8 10
Time (s) Weakening Technique and Constant Power
(b) Control,” Proc. of IPEC’97, Singapore, pp752-
757.
Fig. 3 speed response with load torque disturbance
[4] M. Bodson, J. Chiasson and R. Novotnak, “High-
reference ω ref = 1950 rpm, a sudden change of load torque Performance Inductance Motor Control via Input-
Output Linearization,” IEEE Control System
( d = 9 Nm) is applied at t = 6 second. It can be observed Magazine, Vol. 14, No. 4, pp25-33, 1994.
that the proposed load-adaptive controller demonstrates [5] T. Raumer, J. M. Dion, L. Dugard and J. L.
satisfactory performance with a short time transient of Thomas, “Applied Nonlinear control of an
speed and no steady state error (see Fig. 3(a)). The actual Induction Motor Using Digital Singnal
and estimated load torque disturbances are given in Fig. Processing”, IEEE Trans. Control Systems
3(b) (dotted line represents actual, and solid line describes Technology, Vol. 2, No. 4, pp327-335, 1994.
estimated dynamics). [6] S. K. Panda and P. K. Dash, “Application of
Nonlinear Control to Switched Reluctance
It has also been shown that the proposed nonlinear MIMO Motors: a Feedback Linearization Approach,” IEE
speed sensorless control scheme can be implemented with Proc-Electrical Power Applications, Vol. 143, No.
high control performance, even in the unknown load torque 5, pp371-379, 1996.
disturbance. [7] J. Chiasson, “Nonlinear Differential-Geometric
Techniques for Control of a Series DC Motor,”
VII. CONCLUSION IEEE Trans. Control Systems Technology, Vol. 2,
No. 1, pp. 35-42, 1994.
Base on nonlinear MIMO feedback linearization, several [8] J. Chiasson and M. Bodson, “Nonlinear Control
nonlinear control techniques have been investigated for of a Shunt DC Motor,” IEEE Trans. Automatic
speed tracking of a SEDCM system modeled in Control, Vol. 38, NO. 11, 1993, pp. 1662-1666.
interconnected differential equations. Under unknown load [9] A. Isodori, “Nonlinear Control Systems,”
torque disturbance, the design with the linear control law Spinger-Verlag, Second edition, 1989.
obtained yields the steady-state errors as well as the [10] R. Marino, S. Peresada and P. Valigi, “Adaptive
degraded system responses as a result of the incomplete Input-Output Linearizing Control of Induction
linearization. In addition, introducing an integrator cannot Motors,” IEEE Trans. AC, Vol. 38, pp. 208-221,
effectively deal with this problem. To overcome this 1993
limitation, load-adaptive control design has been developed

7
547

Authorized licensed use limited to: UNIVERSIDAD AUTONOMA DE BUCARAMANGA (UNAB). Downloaded on February 16,2021 at 15:08:07 UTC from IEEE Xplore. Restrictions apply.