IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. VOL. 2, NO. 1.

1dARCH 1994 35

[12] F. W. Ainsworth, “The effect of oil-column acoustic resonance on

hydraulic valve ‘squeal‘,” Truns. ASME, vol. 78, no. 4, pp. 773-778,
1956.
[13] A. A. Yermoshin and R. F. Nagayev, “Chatter of a hydraulic actuator,”
vibration Eng., vol. 2, no. 4,pp. 549-555, 1988.
[14] J. Watton, The stability of electrohydraulic servomotor systems with
transmission lines and nonlinear motor friction effects,” J. Fluid Control,
vol. 16, no. 2-3, pp. 118-136, 1986.

Nonlinear Differential-Geometric Techniques

\ I for Control of a Series DC Motor
500 1000 1500
John Chiasson
Frequenw bd/sl

Fig. 11. Singularities of Gz(s), k = 250.

Absfract-The problem of controlling a Series dc motor using only
current measurementsis considered. It is shown that both speed and load-
VI. CONCLUSION torque may be estimated from the c-nt measurements. Rvo nonlinear
It is concluded that hydraulic drive systems, consisting of a feedback laws are considered based on feedbrk linearization and input-
output linearization, respectively. Both of these control laws require
hydraulic transmission line with a nonlinear discharge boundary knowledge of the speed and load-torque. The speed/torque estimation
condition, may result in limit cycle conditions. When the hydraulic scheme and the control schemes are valid in the presence of magnetic
actuator is secured by a mechanical structure, additional limit cycle saturation in the field circuit and when high-speed field-weakening is
conditions exist. These additional limit cycle conditions can be shifted employed. By neglecting the armature inductance, the estimation is
accomplished using nonlinear state-space and output-space transforma-
as a function of a feedback control gain. All limit cycle conditions, tions to construct an observer with linear error-dynamics whose rate
however, remain a strong function of the hydraulic transmission line of convergence may be arbitrarily specified. (Such an observer could
parameters, which can be altered to tuned for maximum gain margin provide reliability to eldsting systems in the event pf a speed sensor
given a required control gain. The approach presented in this paper is failure.) The feedback-linearjzationcontroller involves a non-trivial state-
shown to be an effective analytical tool, as opposed to time domain space transformation allowing control of the full state trajectory. An
inpuboutput linearization controller with stable internal dynamics is also
approaches, to provide a means to manipulate the occurrence of limit explicitly constructed. Finally, simulations are given to demonstrate the
cycles in nonlinear transmission line systems, which is useful when algorithms.
designing for optimal system performance.
I. INTRODUCTION
REFERENCES A dc motor in which the field circuit is conQected in series with
the armature circuit is Gferred to as a series dc motor. Due to
[ 11 D.L. Margolis, “Signal shaping of fluid transmission lines using parallel this electrical connection, the electromagnetic torque produced by
branching,” J. Dyn. Syst., Meas. and Control, Trans. ASME, vol.. 108, this motor is proportional to the square of the current (below field
pp. 296-305, Dec. 1986.
[2] S . J. Wright et aL,“MatchedImpedance to Control Fluid Transients,” J. saturation). As a result, the series-connected dc motor produces more
Fluids Eng., Trans. ASME, vol.. 105, pp. 219-224, June 1983. torque per Ampere of current than any other dc motor. It is used in
[3] T. J. Viersma, “Suppression of pressure fluctuations in pipelines sup- applications that require high torque at low speed, e.g., subway trains
plying hydraulic servosystems,” Proc. IFAC Symp., Warsaw. Pneumatic [2]. In fact, the series motor is the most widely used dc motor for
and Hydraulic Components and Instruments in Automatic Control, pp. electric traction applications [ 161.
205-222, May 1980.
[4] M. J. Zielke et al., “Forced and self-excited oscillations in propellent The mathematical model of the series dc motor is nonlinear. The
lines,” J. Basic Eng., Trans.ASME, vol.. 91, Series D,no 4, pp. 671-677, advances made in the differential-geometric approach to nonlinear
1969. control theory (see e.g., [4]-[7]) have opened up new avenues to
[5] F. T. Brown, “The transient response of fluid lines,”J. Basic Eng., Trans. controlling such systems. For applications of this theory to motors
ASME, Series D, vol.. 84, no 4,pp. 547-553, Dec. 1962.
[6] V. L. Streeter and E. B. Wylie, Hydraulic Transients. New York: with nonlinear mathematical models, the reader is referred to [6,
McCraw-Hill, 1967. p. 2261, [13]-[15], [19]. In this paper, using the results of Keller
[7] R. E. Goodson and R. G. Leonard, “A survey of modeling techniques [8], Krener and Isidori [9], b e n e r and Respondek [lo] and Xia
for fluid line transients,” J. Basic Eng., Trans. ASME, Series D , vol.. 94, and Gao [ l l ] along with the approximation that the armature in-
June 1972. ductance L is zero, a nonlinear observer with linear error dynamics
181 A. Gelb and W. E. Vander Velde, Mvltiple Input Describing Functions
and Nonlinear System Design. New York McGraw-Hill, 1968. is constructed for the series dc motor which estimates the speed and
191 J. B. Taylor and K.L. Strobel, “Nonlinear control system design based load-torque based on measurements of the current, Further, feedback-
on quasilinear system models,” Proc. 1985 Amer. Control C o f , Boston, linearizing and input-output linearizing controllers for speed control
June 19-21, pp. 469473, 1985. are shown to exist and are explicitly constructed for the Series dc
[lo] L. J. Heyns, Describing Function-based Control Synthesis for a Nonlin-
ear Hydraulic Drive System, M.Eng Thesis, Rand Afrikaans University, J. Chiasson is with the Department of Electrical Engineering, 348 BEH,
Johannesburg, Nov 1989. University of Pittsburgh, Pittsburgh, PA 15261. Paper was recommended by
[ll] W. H. Hayt, Eng. Electromagnetics, Fourth Edition, Singapore: Associate Editor B. M.Spong.
MacGraw-Hill, 1986. IEEE Log Number 9215714

1063-6536/94$04.00 0 1994 IEEE

36 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 2, NO. 1. MARCH 1994

motor. A preliminary (conference) version of this work is given in

[221.

11. MATHEMATICAL MODEL OF A SERIES DC MOTOR[I] [2]

As previously mentioned, a series dc motor is one in which the
field circuit is connected in series with the armature circuit. This
is illustrated in the simplified picture of a dc motor in Fig. 1 by
connecting terminal T’z to terminal TI and applying the voltage
between terminals T’1 and T z .
The armature inductance is denoted by L, and the field flux is
given by $ f ( i f ) = f ( i f ) where f(.) is the magnetization curve
as shown in Fig. 2. The magnetization curve is strictly increasing,
symmetric with respect to origin and satisfies if (i) > 0 for
i # 0. For i f small, df may be modeled as a linear function Fig. 1. Series-connecteddc motor.
of i f , i.e., q5f = L f i f where L f = d)(O). In a typical series
wound dc motor for traction applications the condition L f > L,
holds. In Fig. 2 below, R, and Rf denote the resistance of the the series dc motor is used for speed control applications as opposed
armature and field windings, respectively. The resistance Rp placed to positioning control systems. Note that the motor can only be slowed
in parallel with the field winding is for field-weakening. (R, is not down by the load-torque (neglecting the viscous-friction torque B w )
indicated in Fig. 1.) The constant K,,, denotes the torquehack-emf or by reversing the connection of field and armature circuits as
constant. described above. In the former case, this means that the physical
The dynamic equations describing a series-wound dc motor are [2, mechanism by which the controller reduces the motor speed i s by
pg. 581: lowering the current in the motor so that the load-torque may slow
L,di,/dt = V - & a Q - Rp(iQ- i f ) - Kmq5f(if)w it down. Consequently, one never unloads such a series motor nor
connects it to the load by a belt (or other mechanism that could
d 4 f / d t = -Rfif +
Rp(ia- i f )
break) as the result could be disastrous [3, page 3061. Series motors
Jdw/dt = K m 4 f ( i f ) i -
, BU - T L . (1) used for vehicular transportation are connected as in Fig. 2 while
the motor is accelerating or running at constant speed. To decelerate
These equations are valid in the field-weakening region. That is,
the motor, the field to armature connections are reversed which then
at high speeds (above the so-called base speed) the switch is closed
allows the controller to use the motor torque to slow the machine
so that the field current i f is less than the armature current i,. The
field flux is then less than it would be with full armature current, down. In summary, below the base-speedd(i.e., no field-weakening),
the system is described by (3) while above the base-speed the system
Le., p f ( i f ) < df(i,). As can be seen from the first equation of
is described by (1).
(l), the applied voltage V must be greater than the back-emf voltage
K m d f ( i f ) win order to produce armature current i,. The back-emf
being less allows the speed w to be larger with the same voltage 111. A NONLINEAR OBSERVER FOR SPEED AND LOAD TORQUE
than it would be if the field windings had full armature current. The With the currents i q , i f as the available measurements, we now
trade-off is that the torque put out by the motor is now less for the consider constructing an observer with linear error dynamics [9] to
same armature current as can be seen from the third equation of (1). estimate the speed w and load-torque 71,.With the models given by
If the first two equations of (1) are added, we have (l), (2) or (3) above, it is not possible to design such an observer.
+
d ( d f ( i f ) L,ia)/dt = - R f i f - Rai, However, by using engineering judgment, one can simplify the
problem and still obtain an accurate observer. The key approximation
- KrndJf(if)W +v is to note that L,i, is very small compared to d f ( i f ) . For example,
Jdw/dt = Km4f(if)io- B w - TL (2) below the saturation current level where $ f ( i f ) = L f i f the ratio
L,/Lf can be [l, page 901. Even in the field-weakening
This f o m will be used to design the speed-observer. region, d f ( i f ) > L,i, [2, page 591. Making this approximation
Below the base-speed, field-weakening is not present, i.e., the in (2), defining z1 = d f ( i f ) , z z = w and modeling the effect of
switch is open so that R, = 00 and (2) reduce to the (constant) load-torque as the additional state variable 2 3 = T L / J
+ +
d ( $ f ( i ) L,i)/dt = -(Rf R,)i - K m $ f ( i ) w V + results in
Jdwldt = K,df(i)i - Bw - TL (3) d z l / d t = -klzlxz - R f i f - R,i, +
V
where i = i , = i f . +
dzz/dt = - k z ~ z- 5 3 ( K m / J ) 4 f ( i f ) i ,
Note that the torque produced by the motor Kmq5f(i)iis always d ~ 3 / d= t 0
positive which is a direct result of the series connection. In more y = [ 1 0 O]z=z1. (4)
detail, when the armature current is reversed, so is the field current
(as they are the same current!) thus reversing the magnetic field in where le1 = K,, kz ;= B / J . Note fiat the measurement is now the
the air-gap and keeping the torque positive. If a negative torque is field flux z1 = 4 f ( i f ) which is available by physically measuring
required, the terminal Ti of the field circuit must be connected to the field current and using the magnetization curve to determine q5f.
terminal TZ of the armature circuit instead of TI (see Fig. 1). Even at However, (4) is nor linear in the unmeasured state variables due to
high speeds where field-weakening is employed, it is easy to see the term in the first equation as 1 2 = w is not available. As
that i, and i f have the same sign and thus the torque is again described in the literature [6&[12], the idea is to find a coordinate
always positive (or negative, depending on the connection of the transformation z* = T ( z )which will transform (4) to a system which
terminals). As a consequence of this unidirectional torque production, is linear in the unmeasured state variables. For 2 1 = d f ( i f ) > 0,
 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 2, NO. 1, MARCH 1994 31
,
i
I consider the change of coordinates IV. NONLINEAR SPEED CONTROLLERS FOR THE SERIES DC MOTOR
zr = In(x1) = In(4f) A speed controller for the series dc motor may be designed using
the concepts of feedback linearization and input-output linearization
x; = 2 2 = w (5) [6], [7].That is, by a combination of a change of coordinates and
X: = 2 3 = TL/J state feedback, the original nonlinear system (1) can be made linear
and change of output transformation from either input to state or input to output. To do so, let i f = $(pf)
where $(.) is the inverse of pf = pf(if) = f ( i f ) .Then (1) becomes
y* = h ( y ) = In(4f).
This output function comes out quite naturally from theorem
dialdt = V/La - ia(Ra + Rp)/La + (Rp/La)$(df)
4.1 of [lo] or by solving (25) in [8]. Using the output equation - (Km/La)dfw
y* = h ( x ) = ln(df), the simpler conditions in [9] may be checked + +
d 4 f l d t = - ( R f R p M ( 4 f ) Rpia (8)
to show an observer exists. The reason for the transformation becomes d w / d t = ( K m / J ) 4 f i a- T L / J
clear upon computing x ; as
where the viscous-friction coefficient has been set to zero ( B = 0).
X; = ( l / z l ) d s l / d t= ( l / X 1 ) ( - k 1 x l z 2 - Rfif - Raia + V )
+
= - k l ~ f (-Rfif - Raia +V)/df(if) A. Feedback Linearization with Field- Weakening
which is now linear in the unmeasured state variable x ; = w . Consider the nonlinear transformation
Defining the known (measurable) vector p and system matrices A ,
cas X; = T I ( x )= w - ( K m / 2 R p J ) 4 ;
+ TI(^) = ( K m / J ) ( R f / R p+ l)dJf$(4f)
~ ( i f i a, , V ) 4

A 4 0 -kz
[
(-Rfif - Raia V)/4f(if)
(Km/J)$f(if)ia

[I -;-;I; 0 I; X;

-TL/J=dxf/dt
x: = T3(2)= ( K m / J ) ( R f / R p+ l ) d ( 4 & ( 4 f ) ) / d t
= a($fh) + 4f4'(4f))
+ +
(-(Rf Rp)$(4f) R&) = d x f / d t
(9)

c A [ l 0 01 +
where (Y 4% ( K m / J ) ( R f / R p l ) , $ ' = d $ , / d $ f .
In this new coordinate system, the system equations are then
the transformed system becomes
+
2* = Ax* p(if,ia,V ) (6)
d x I / d t = xf
y* = cx* = ln(4f) dx;/dt = xl (10)

An observer is defined by
d x g / d t = a(ia,b f , W ) + b(4f)V
3i'* = A i * + p(if,i,, V ) + I(y* - c*) I = [I1 12 IalT +
where b ( ~ f )= 4 $ ( 4 f ) 4r$'(9f) )(Rp/La)*The quantity
U(&, $f, w ) is a messy expression but straightforward to compute.
(7) Setting V = (-u(ia, (bf, w)+u)/b(q5f) results in a third-order linear
e*= CP'. system from the new input U to the new state space coordinates x*.
Clearly, the condition b ( 4 f ) # 0 must hold for this controller to
Defining e = x* - 2* and subtracting (7) from (6) results in the
linear error dynamics
be valid. It is easy to see that $(4f) df$'(df) > 0 as long as +
i f > 0. That is, the saturation (magnetization) curve @f = f ( i f )and
E = ( A - IC)& its inverse if = $($p) are both strictly increasing so that dJf > 0
for i f > 0 and 1c, > 0 for 4f > 0. Further, $'(4f) > 0 for all
As the pair ( c , A ) is observable, E may be chosen so that A - IC bf. As this controller is designed for the field-weakening region, the
is stable. current i f is bounded above zero and the controller is thus valid.
Remark 1: In the development of the observer, the assumption
The value of rL/J would be obtained from the observer described
that 4 f ( i f )> 0 was made. (Conversely, the condition r j f ( i f )< 0 in the previous section. Note that the load-torque has been canceled
may be used with x ; = lnl4f I to construct an observer.) That is, the out using feedforward of its estimate.
field current must be kept positive and boupded away from zero to use This feedback controller is dependent on the value of La which,
this observer. However, as the torque produced by the motor d f ( i f ) i a as noted previously, is quite small. We now consider a controller
is always positive, irrespective of the direction of the current, there i s
designed with La taken to be zero which is consistent with the
no reason to have the current change sign. At start-up of the motor, approximation made in the observer design. To this end, set La = 0
the observer would not be initiated until the measured field current in (1) (again with B = 0 far simplicity) and solve for i, to get
(and thus field flux) is above a certain level. In fact, a controller
must be designed to keep the current bounded away from zero for
controlled operation of the motor. This is easily understood by noting
+
i, = (V Rpif - Kmdf(if)w)/(Ra Rp) (11) +
that if tho system (3) is linearized about if = 0, the resulting linear This is then substituted into the second and third equations of (1)
system is not controllable. to obtain
Remark 2: Note that the structure of the observer does not change
as the motor goes into (or out of) field-weakening, although the +
d 4 f / d t = - ( ~ f ~ p ) l ~ l ( 4 f ) (Rp/(Ra ~ p ) + + )
dynamics of the motor do change! The structural change in dynamics,
that is, from (1) to (3) is accounted for in (2) by the redundancy
+
x (V RP$(dJf)- K n d f W ) (12)
of the current measurements at speeds below field-weakening since d w / d t = (Km/J)4f(V -t Rp$($f) Km&fw)/(Ra Rp) .- +
then i f = i,. (4J) -
 '
38 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. VOL. 2, NO. 1, MARCH 1994

where t o is the time that field-weakening was initiated and i f 0 is the

$2f
ffeld cut-rent at time t o . From this last expression, it is clear that
is bounded if the input U iS bounded (i.e., bounded acceleration, see
(15)), Thus, the second-order system (12) with the feedback control
strategy (14) is BIB0 stable. Note this controller gives a simple,
straightfonv~dmethod to control the speed.

C. Controller Below Field- Weakening

The above controller designs were done assuming field-weakening
was in effect. We now consider the case where Rp = 00 so that
the system equations are given by (3). In this case, consider the
coordinatq transformation
z; = Tl(S)= w (18)
I; = T~(I)= ( K m / J ) 4 f ( i )-i T L / J= d x ; / d t
where x ; , xz are the speed and acceleratiotl, respectively. In these
new coordinates, the system dynamics becomes
d x ; / d t = x; (19)

B. Input-Output Linearizdtion with Field- Weakning

The feedback lineariqation controllers are complicated and incon-
venient due to the fact that none of *e transfprmed state variables is
d x l / d t = (K,/J)
0 $>(i)i ++f(i)di/dt

the speed. 4 s a consequence, any reference trajectory for the speed

w would have to be transformed to the I* coordinates and then the
feedback controller could be designed to track this reference. It would
be simpler aqd more like standard engineering practice not to require
this. This-can be accomplished using Input-output linearization rather
than feedback linearization. To do so, he input V in the second
equation of (12) is set equal to
V = -Rp$(dfI+ K m 4 f w + u(Ra + R n ) / ( ( K m / J ) d f ) (14)
to obtain the system i s then linear from the new input U to the new state
coordinates x* , As4f (i) is strictly increasing,$; ( i ) > 0 for all i and
dw/dt = U - TL/J (15) +
consequently, d > ( i ) i $ f ( i ) > 0 for i > 0. That is, the controller
mq output (speed) is now linearly related to the new input and U
is valid for d > 0. In this case (i.e., without field weakening), the
hence, input-output linearization has been achieved. An appropriate feedback and input-output linearization controllers are identical. The
choice for U is new input U may be chosen as

U = K1
I'
(Wref - 2 ) 4-K Z ( W r e f - b ) + + +L/J
aref

where W r e f , aref= dw,,f/dt are the reference speed and acceleration,

(16) U = K1

+
I' - 2 ) + K2(wref - 2 )
(Wref

KS(aref - h ) + jref (21)

respectively, h is the estimated speed and +L is the estimated load- where wref,a,,f = dwref/dt,jref = d a r e f / d tare the reference speed,
torque. acceleration and jerk, respectively, 2 is the estimated speed and
With the control law (141, the first equation of (12) becomes ,& = ( K m / J ) d f ( i )-
i r i / J is the acceleration found psing the
+ +
d 4 f / d t -(Rf Rp)ll(df) u J R p / ( K m d f ) (17)
estimated load-torque,
If the field flux i s not in saturation so that $ f ( i ) = L f i , (18)
To check the stability of (171, multiply both sides by 24f to obtain and (19) reduce to the results given io [20], [21]. Finally, note that
by taking Rp = w, the first two equations of (9) reduce to the
+ +
d$2f/dt = - 2 ( ~ f R p ) $ f + ( $ f ) u ( ~ J R ~ / K ~ ) transfoyation (18) as expected.
The saturation curve is such that I$f(if)l5 ( L f i f l where L f =
Q;(O). It then follows that (recall fiat if = $($f)) D. Summary of the Control Laws
In the field weakening region, the input-output controller has stable
internal dynamics as well as being much simpler to implement than
either of the feedback linearization controllers. Consequently, there
where is no advantage to either of the feedback linearizing controllers in the
field weakening region. Below base speed, the controller presented
above is both an input-output and a feedback linearizing controller,
that is, they we identical in this case.
In summary, it is proposed to use the control law given by (20)
and (21) below field weakening and the control law (14) and (16) in
the field weakening region. The use of this control combined with
 ~

EEE TRANSACTIONS ON CONTROL SYSIEMS TECHNOLOGY,VOL. 2, NO. 1, MARCH 1994 39

Magnet lzatlon/ m
Saturat I on curve -1
0
T
2
Fig. 2. Schematic circuit for a series-connected de mot&.

the speednoad-torque observer is explored with a simulation study

in the next section.

V. SIMULATIONS
As explained in the previous section, the input-output controller
combined with the speednoad-torque observer is proposed. A Block
diagram illustrating this configuration is given in Fig. 3.
To illustrate the proposed observtWcontroller algorithm, simuld-
tions were performed using the simulation package SIMNON [23).
The parameter values used are La = 0.0014 H, R f = 0.01485
R, Rp = 0.01696 R, R, = 0.00989 R, B = .1 N . m / r d / s ,
Km = .04329 Nt . m/Wb - A, J = 3 Kg m 2 (the resistance
values are given at 110'C). The parameters La and B are taken
to be zero in the controller and observer algorithms. The field flux
saturation curve q5(if) and its derivative q5'(if) are given in the left
side and right side, respectively, of Fig. 4. The flux curve q5(if) was
calculated as a piecewise linear fit to given data points (ik, q5(ik)) for
k = 0, ... , l o . As a consequence, c$'(if)is a (dereasing) staircase Oyer 'per ar*rtdt 4r-r *
function.
For later reference, the maximum torque at constant speed id the
field weakening region is computed as follows. The current lirilit
. .
is aamax = 1000 Amperes and the corresponding maximum field
+
current is then ifmax = i a m a x R p / ( R j &) = -53iamax. Thus
the maximum torque of the motor in the field weakening region is
rmax Km$(ifmax)iamax = (.04329)(31.6)(1000) Nt-m = 1368
Nt-m.
The load-torque disturbance T L ( ~ was) chosen as wrar 'par =
Fig. 3. Block diagram for the series de motor contyoller.
7-L = 0 N t - m OSt5.5
to be changing at the time of field-weakening to demonstrate the
TL = 1250(t- 5 ) / 5 N t . m 5 5 t 5 10
effectiveness of the controllei. in dealing with such a disturbance.
TL = 1250 Nt-m 10<t
For comparison purposes, a standard controller given in Fig. 9.10
Note that the load-torque disturbance itlcreases up to 91% of the of Leonhard [2,p. 1371 was also simulated. f i e only modification
maximum torque capability of the motor. was the addition of the load torque and speed estimator (7).
The base speed Wbase was set at 200 rdsh The (input) voltage V In bath the dontrollet and observer algorithms, there is a singularity
into the motor is restricted to 0 5 V 5 1000 V. The observer gains when the field current is zero y d thus, at the start of the simulation.
To get m u n d this difficulty, the field flux in the controller and
were chosen to put all three poles of the observer e m r system at -40.
observer algorithms (not in the simulation of the motor) was bounded
The gains of the controller below field-weakening were chosen to put
the closed-loop poles at -10, that is, K1 = 30, K Z = 300, K3 = below by 1. That is to say, if 4 < 1, then 4 was set equal to 1 in
the conhpller and observer al8orithms. The observer was jn@$~ed
1000 in (21). In the field-weakening region, the gains of the input-
output controller were chosen tb put the closed-loop poles at -15, with the load-torque and speqd zero and it yas hot started until the
that is, K I = 30, K Z = 225 in (16). The reference speed trajectory, field flux (calculated using the measured field currefit) was one or
which was chosen to have the motor start from 0 and go up to 520 greater (e0.Olsecond).
rdds in 20 seconds, is shown in Fig. 5. For this motor, there are 6.5
rdddmph so that 520 rdds corresponds to 80 miles p& hour. The A. Sensitivity to Parameter hlues
reference speed trajectory reaches the base speed of 200 rdds (30.8 There axe sensitivity issues to consider. The flux curve 4 ( i f ) is
mph) at t = 8.45 k o n d s , The load-torque was purposely chosen known accurately by design since it determines the torque capability
40 IEEE TRANS;ACTIONS ON CONTROL SYSTEMS TECHNOLOGY. VOL. 2, NO. 1, MARCH 1994

Fig. 4. @ ( i f ) in webers (left), @‘(if)

in W A (tight) versus current in Fig. 8. w and wref as well as w - wref in rdds versus time in seconds.
amperes.

values at 110” as given above. Although the controller is insensitive

to this variation, it does affect the speed estimator (7). However, for
p ( i f ,i,, V) in (7) note that V - R f i f - R,i, z V as R f ,R, a
x so

aoK
small. As shown in Fig. 8 below, a 50% variation in these resistance
values leads to only a 1 mph error in the speed estimate. However,
as Fig. 9 shows, the standard controller with the speed observer (7) is
highly sensitive to this parameter variation and is unable to achieve
200
acceptable tracking of the reference trajectory.

0
B. Simulation Results
Fig. 5 below is a plot of the actual speed w and the reference speed

i/-
0 5 10 15 20 0 5 10 15 20
Fig. 5. Plot of w and w,,f as well as w--ur& in rdds versus time in seconds. w,f along with a plot of the speed tracking error w - w,,f. Note that
the largest value of the error w - wref is about 5 rds/s.
A plot of the speed estimation error ij - w is given in the left side
of Fig. 6 below while the right side is +L/J. At the time of field
weakening (E 8.45 s.), the estimate + L / J is off because the both
the load torque and friction torque are ramping up. In addition, the
current i , rises steeply at field weakening (see Fig. 7) resulting in
100 L , d i , / d t having some effect. These two facts account for the small
transient behavior in the estimates of ij and ? L / J initiated at the
onset of field weakening.
The field and m a t u r e currents during the run are shown on the
left side of Fig. 7. Of course, until field-weakening is initiated, these
0 5 10 15 20 two currents are the same. The voltage (input) required for the run
Fig. 6. Plot of LJ- w in rds/s and ?L/ J versus time in seconds. is shown on the right side of Fig. 7. Note that at the onset of
field-weakening, the rate of increase of the voltage is significantly
reduced.
Variations in R, and R f are now considered. The resistance values
R, and R f were reduced by 50% in the motor simulation, but kept
at their original values in the controller and observer algorithms.
The left side of Fig. 8 below is a plot of the speed w and its
reference wref while the right side is a plot of the speed tracking
error w - wref. Note that the steady-state error for w - Wref is
about 6.5 rdds (1 mph). Finally, Fig. 9 is plot of w and wref under
the same conditions as Fig. 8, but using the standard controller.
Note that the tracking is unacceptable when the motor goes into
o 5 io i5 20
field weakening. The author was unable to improve the tracking
Fig. 7. Field and annature currents in amperes as well as input voltage in by “tweaking” the gains. Note that the standard controller does not
volts versus time in seconds. account for the change in dynamics as the motor goes into (or our
of) field weakening.
of the motor. The system is insensitive to L, since it is so small.
This is demonstrated in the simulations below where L , is included
in the simulation of the motor dynamics, but is taken to be zem in VI. CONCLUSION
the controller and observer algorithms. The controller and observer It has been shown that nonlinear differential-geometric techniques
algorithms are also insensitive to the viscous friction coefficient B ; can be successfully applied to the series dc motor. Specifically,
the load torque observer is just estimating the unknown quantity this work indicates that using the proposed nonlinear observer and
+
T L / J B w to maintain an accurate estimate of speed. controller scheme results in accurate speed tracking while being
The values of R f ,Rp and R, do change due to ohmic heating. insensitive to variations in resistance and viscous friction parameter
In starting a cold motor, these resistances would be less than their values as well as varying load torques. An immediate consequence
IEEI3 TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 2, NO. 1, MARCH 1994 41

400

200

0 5 10 15 20
Fig. 9. w and wref in rdsls versus time in seconds-standard controller.

the actual voltage into the motor. Specifically, U is multiplied by

(J/Km)4>(i)/(4>(i)i + $ f ( i ) ) which can be viewed as a current

v
7

L+
W C )

-
-
QL/J
dependent gain. It can be shown by simulation that this gain varies
from about 4 at t = .01 s.to .04 at the time of field weakening
(t = 8.45 seconds). Thus, the effect of the gains K1, Kz, K3 are
--
Lies Look-up
M: Table 1n(6!Jr)) Load-Torque
&
Motor . Observer greatly reduced in (21) for the input-output controller compared to
the standard controller (22). This is consistent with the fact that the
I I error terms in the standard controller must generate enough voltage to
+
overcome the voltage drop - ( R f R, ) i - K , df( i ) w as well as do
the corrections for trajectory tracking. On the other hand, the input-
output controller takes these terms specifically into account with the
input U only accounting for corrections in the trajectory tracking. (In
either controller, the bulk of the voltage into the motor is used to
overcome the back-emf term K m 4 f ( i ) w . )
In the field weakening region, the standard control law (22) is also
similar to U in (16). Note that U in (16) is substituted into (14) to give
the voltage into the motor. In this case, the current dependent gain
t wt.r f
+
is (R, R p ) / ( ( K m / J ) 4 fwhich,
) from simulation, turns out to be
about .06in the field weakening region. Again, the effect of the gains
Fig. 10. Block diagram for a conventional series dc motor controller with
a speed and load torque observer.
K1, Kz, K3 on the actual input voltage are greatly reduced in (16)
for the input-output controller compared to the standard controller
(22). This is again consistent with the fact that the feedback error
of this result is the possibility of using these motors without a speed terms in the standard controller (22) must account for the voltage
sensor. Even in existing motors, the scheme proposed here would -
drop -R,q!~($f) K m $ f w while the input-output controller takes
provide reliability to the system in that, if the speed sensor were to this directly into account.
fail, the control scheme proposed here would provide the capability In the simulations, the gains were chosen as K p = 147, KO= 21
for continued controlled operation of the motor. and K I = 334.

ACKNOWLEDGMENT
APPENDIX A
The author is grateful to AEG Transportations Systems of Pitts-
A. Standard Approach to Control of Series DC Motor burgh, PA, for their cooperation in this work. In particular, he would
A block diagram of a standard controller for a series dc motor like to thank Ed Harbist of AEG for valuable discussions on this
is given Fig. 10 below. This approach is described in Leonhard [2, work and for providing the parameters used in the simulations. He
p. 1371. The speed error is put through a PI controller which is then is also grateful to Keith Clawson of AEG for his time and expertise
used as the torque reference. The computed torque K m 4 f ( i f ) i a-.it in going over the workings of such motors. Finally, he would also
is then subtracted from this torque reference and used as the voltage like to thank Bob Novotnak, the Associate Editor, and the reviewers
commanded to the motor. The control structure in Fig. 9.10 of [2] has for their valuable comments.
been modified in Fig. 10 by adding the load torqudspeed observer
and the acceleration reference. REFERENCES
The input V into the motor is given by
[l] P. Krause, Analysis of Electric Machines. McGraw-Hill, 1986.
[2] W. Leonhard, Control of EZecfricalDrives. Springer-Verlag, 1985.
v = KOKI I’ (Wref’Oj)+l(gKp(Wref-~)+KO((Yref-(Y) (22) [3] S. J. Chapman, Electric Machinery Fundamentals. McGraw-Hill,
1985.
[4] B. Jakubczyk and W.Respondek, “On linearization of control systems,”
(Y = (Km/J)4f(if)d,- +L/J. With KI = KOKI,ICZ =
where Bull. Acad. Polon. Sci., Ser. Sci. Math Astr. Phys. 28, pp. 517-522, 1980.
K D K P , K3 = KO, this control law is the same as that given [SI R. Su er al., “Design for multi-input nonlinear systems,” Differenfial
Geometric Control Theory, edited by R. W. Brockett et al., Boston, pp.
by U in (21) except that the jerk reference jrefis not present. 268-298, 1983.
However, note that U in (21) is then substituted into (20) to produce [6] A. Isidori, Nonlinear Control Systems, Second Edition, 1989.
 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 2, NO. 1, MARCH 1994

H. Nijmeijer and A. J. van der Schaft, Nonlinear Dynamical Control

Systems. Springer-Verlag, 1990.
H. Keller, “Nonlinear observer design by transformation into a gen-
eralized observer canonical form,” In’l J. Control, vol. 46, no. 6, pp. J
1915-1930, 1987. I I
I
A. J. Krener and A. Isidori, “Linearization by output injection and Fig. 1. Cascade control set-up.
nonlinear observers,” Systems and Control L.erters, 3, pp. 47-52, 1983.
A. J. Krener and W. Respondek, “Nonlinear observers with linearizable
error dynamics,” Siam J. Control and Optimization, vol. 23, pp 197-216, in regulating the output in the face of load disturbances provided
1985. that suitable secondarfr variables are available for measurement. The
X. Xia and W. Gao, “Nonlinear observer design by observer lineariza- manual tuning of cascade controllers, however, is generally a tedious
tion,” Siam J. Control and Optimization vol. 27, no. 1, pp. 199-216, knd time consuming task. In view of its wide spread application,
Jan., 1989. it would be very useful to realize the automation of the tuning
D.Bestle and M. Zeitz, “Canonical form observer designs for nonlinear
time-variable systems,” Ifit’l J. Control, 38, pp. 419-431, 1983. procedure.
R. Marino, “An example of a nonlinear regulator,” IEEE Trans. on There exist many methods for auto-tuning or self-tuning of the
Automat. Control, Irol. AC-29, no. 3, Mar. 1989. single-loop PID controller [3]-[6]. However very little has been
M. Zribi and I. Chiasson, “Position control of a PM stepper motor by reported in the literature on the development of self-tuning or auto-
exact linearization,” IEEE Trans. un Automat. Conml, vol. 36, no. 5,
May 1991. tuning techniques for cascade control. An exception is a recent paper
M. Bodson et al., “High performance nonlinear feedback control of a by Li et al [7] which makes use of fuzzy logic for self-tuning of
permanent magnet stepper motor,” IEEE Trans. on Control Syst. Tech., cascade controllers. In this paper, a simpler means for auto-tuning
vol. 1, no. 1, Mar. 1993. cascade controllers will be developed by extending the relay feedback
R. D. Begamudre, Electro-Mechanical Energy Conversion wifhDynam- auto-tuning technique to tune the cascade controllers. Fig. 2 gives an
ics of Machines, Wiley, 1988.
T. Kailath, Lihear Syktems. Prentice-Hall, 1980. illustration of the scheme. The relay is used to tune the secotldary
S. Sastry and M. Bodson, Adaptive Control4tability. Convergence and and primary loop controllers one at a time. Information about the
Robustness. Prentice-Hall, 1989. relative speeds of the loops in cascade is also derived from the relay
J. Chiasson and M. Bodson, “Nonlinear and adaptive control of a shunt feedback oscillations.
dc motor,” to be published in IEEE Trans. on Automat. Control. vol. After a short overview of the relay auto-tuning method for single-
38, no. 11, pp. 1662-1666, Nov. 1943.
Oliver, Philip D., “Feedback linearization of dc motors:’ IEEE Trans. loop systems, relay auto-tuning of cascade loops is analysed in section
on Industrial Electronics, vol. 38, no. 6, Dec. 1991. II. Section III describes additional information that may be derived
J. Chiasson, “Nonlinear Control Systems-EE 3648 Class Notes,” May from relay feedback of cascade loops. The development has been
1989. substantiated by simulation. Conclusions are presented in section IV.
I. Chiasson, “Nonlinear differential-geometric techniques for control of
a series dc motor,” Proc. I993 Amer. Control Con$. June 1993, San
Francisco, CA. 11. AUTO-TUNING
OF CASCADE LOOPS
H. Elmqvist et al., SIMNON-User’s Guide for MS-DOS Computers,
SSPA Systems, 1990. The relay feedback auto-tuning of PYPID controllers [5]-[6] is well
developed and is commercially available in single-loop controllers.
The relay is used to obtain the ultimate gain (kL) and period (tu)
of the process. These quantities are needed to apply established
tuning rules [8]-[9] for the PYPID parameters of the controller. The
method involves replacing the PYPD controller by an on-off or relay
Relay Feedback Auto-Thing of Cascade Controllers controller. Here for simplicity an ideal relay is shown without any loss
of generality, though in practice a relay with a hysterisis is used [6].
C.C. Hang, Senior Member, IEEE, A. P. Loh, Member, Controlled limit cycle oscillations are observed when this is done.
IEEE, and Y. U. Vasnani, Student Member, IEEE The period of the oscillation is the ultimate period. To calculate
the ultimate gain, a describing function approximation is used to
Abshact-It is shown that the relay feedback auto-tuning technique calculate an equivalent gain for the relay. Let a be the amplitude
can be extended to auto-tune cascade controllers. The use of the relay of the oscillation obtained at the input of the relay and d be the
allows the tuning of the secondary loop to be done without necessarily amplitude of the relay. Then the ultimate gain is given by [6]:
placing the primary controller in manual mode. It plso turns out that the 4d
relative speeds of the loops in cascade can be derived from the limit cycle ktl = ;
oscillations obtained b m relay feedback auto-tuning.
With this ultimate gain and period, Ziegler-Nichols (ZN) [8] rules
(Table I) can be applied to these quantities to set the parameters
I. INTRODUCTION
for PIPID control. Once the loop is closed with a PUPID controller
Cascade control as shown in Fig. 1 is a multi-loop control scheme tuned with ZN rules, a subsequent setpoint change or load response
commonly used in chemical process control [1]-[2]. They are ex- allows the steady-state gain of the process to be determined. Once the
tremely effective when the single-loop PID controller finds difficulty steady-state. gain of the process is known, the more recent Refined
Manuscript received April 12, 1993; revised April 10, 1993. Paper was Ziegler-Nichols (FUN) tuning rules by Hang et al [9] can be applied
recommended by Associate Editor D. Clarke. This paper was financially to improve setpoint and load responses. The RZN tuning rules for
supported by the National University of Singapore and the Singapore Tech- PI control are given in Table 11. The rules for PID control and more
nologies Industrial Coxporation.
The authors are with the Department of Electrical Engineering, National detailed discussions are given in Hang et al [9].
University of Singapore, Singapore. In order to understand how relay feedback aids in the tuning
EEE Log Number 9215713. of controllers in cascade loops, it is necessary first to analyse the

1063-6536/94$04.000 1994 IEEE