Study of The Synchronization of Line-Start Permanent Magnet Synchronous

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Study of the Synchronization of Line-Start Permanent Magnet Synchronous

Motors
Juliette Soulard, Hans-Peter Nee
Electrical Machines and Power Electronics
Royal Institute of Technology
100 4 4 Stockholm SWEDEN
tel: +46 8 79077 36 -fax : + 4 6 8 205268
juliette@ekc.kth.se,hansi@ekc.kth.se

Abstract-The synchronization process of three-phase Line Start Per- rise during the start process would lead to a too complicated
manent Magnet synchronous motors (LSPM) with buried magnets and model for our purpose. An other approach was then consid-
a squirrel-cage is studied in this paper. The goal is to define a pull-in
criterion by using a Lyapunov function. A model is first derived and a
ered.
Lyapunov function is defined using the Lagrange-Charpit method. Ex- In order to calculate the critical load torque in a rapid and
periments and simulations are then compared to check the validity of the simple way, the chosen method considers a Lyapunov func-
model. A criterion to define the capability of synchronization of LSPM tion of the system. A similar method was successfully used
motors is then presented. This criterion is used to study the influence
of the electrical parameters on the synchronization capability of two four by Senjyu and Uezato to predict the synchronization capabil-
pole LSPM motors. The first experiments proved that the prototypes have ity of reluctance motors [4]. No consideration on the shape of
greater synchronization capability than the criterion predicts. the trajectory (slip-load angle relationship) during the synchro-
nization has to be made to obtain the critical load torque. No
I. INTRODUCTION simulation of the starting transients is required either. As this
method is intended to be fast, a parametric study can then be
The decreasing price of permanent magnets and their im- conducted to define the worst case of influence for each influ-
proved performances make PM motors even more interesting ent parameter.
in an industrial point of view as there is an increasing demand A model is first derived so the Lyapunov function can be de-
for high efficiency motors. For applications where the motor is fined by the Lagrange-Charpit method [ 6 ] . Experiments and
to be connected directly to the mains, a squirrel-cage has to be simulations are then compared to check the validity of the
added to obtain a line-start permanent magnet (LSPM) motor. model. The way to obtain an estimation of the critical load
A specific design has to be made as the permanent magnets torque of LSPM motors is introduced and simplified expres-
create a braking torque during the asynchronous start as well sions are derived. This criterion is then used to evaluate the
as they help synchronizing the rotor with the mains. To find synchronization capability and to study the influence of the
a good compromise, both the asynchronous early start and the electrical parameters of a four pole LSPM motor.
synchronizationprocess have to be studied. The determination
of simple criterion to quantify the capability of LSPM motors 11. MODELOF THE LSPM MOTOR
in the early start and synchronization process can then allow to To be able to find a similar Lyapunov function as in [4] for
consider an optimization of the design. LSPM motors, the form of the dynamical equation has to be :
The synchronization process of three-phase LSPM motors
8 6
with buried magnets and a squirrel-cage has been studied in
[ 1,2,3]. In [ l ] and [2], Miller focuses on the description of the
J-
dt2
+ D-d6
dt
+ P(6) = Pioa,j.
synchronization. The critical load torque or maximum load J is the inertia of the shaft, D is the damping constant, Plead
torque that can be synchronized by a given motor, is derived is related to the load torque (constant) and P(6) is a linear
by simulation with a classic dq model [2]. In [3], the critical function of the synchronous torque and the reluctance torque.
load torque is calculated analytically. This analytical method 6 is the load angle. The load angle is linked to the electric
assumes that the relationship between the slip and the load an- speed w , the synchronous electric speed U,,and the slip s as
gle during the final synchronization process can be described follow :
by a sinusoidal function. Due to this assumption, the contribu-
tion of the reluctance torque is not fully considered. d6
During the experiments, it came out that the synchroniza-
-
dt = SW, = w, -W.
tion capability of the tested prototypes are dependent on many Four torques are to be considered here : the synchronous
parameters such as the temperature of the different parts of the torque T,,the load torque Tl,the braking torque from the
motor, the strength of the mains for example. Accurately pre- magnets Tb, and the asynchronous torque Tan. The last three
dicting the transients seemed to be an unreachable goal. For torques are to be submitted to linearization in order to fit in
example, the transients are temperature depending. Adding the dynamical equation. This is done so that the linearized
time-variant resistances to take into account the temperature model is a worse case for the synchronization than reality. The

0-7803-6401-5/00/$10.000 2000 IEEE


424
synchronizationprocess is assumed to occur once the speed is where EOis the no-load voltage, R, the stator resistance, and
higher than wm the speed at which the asynchronous torque is Xd, and X,,the d and q synchronous reactances respectively.
maximum. Linearizing the braking torque between wm and w, yields (fig-
ure 1) :
A. Expressions of the different torques
The synchronous torque depends on the load angle 6. The
expression of this torque has two components due to the pres- Tb E Tb(ws)+ Tb(wm)sm
- Tb(ws)s = TbO + Tbls. (9)
ence of the reluctance torque as well as the magnet-current
torque : It should be noted that this linearization is valid only if wm
is higher than the speed at which the braking torque is maxi-
T, = T , l ~ i n ( d+) T82sin(26). (3) mum. This is usually the case but it was not true for one of our
(see appendix 8.2 for expressions of T,1, and Ts2). prototypes. The speed w s / 2 was used instead of wm for that
The classical expression for the asynchronous torque can be case. A first order linearization does not fulfill our require-
written as follows : ments of worse case than reality as the braking torque is then
underestimated.
T 3pv2x; &s
as - 2 w , a1 + bls + C I S ~ (4)

where V is the mains voltage, p the number of poles, X,


the magnetizing reactance and R, the rotor resistance. The
expressions of al, b l , and c1 are given in Appendix B.
The electric speed wm at which the asynchronous torque is
maximum is given by :

wm = smws =
/:;
-w,.

The synchronizationis more difficult if it starts from a speed


far from the synchronous speed. Then, the synchronous torque
must be stronger to give the energy required to synchronize [2].
Therefore, we assume that the asynchronous torque is linear
between wm and w,. The simplified expression is :

Tas(wm)- Tas(ws)
Tas - s = Tals.
~

(6)
sm Fig. 1. Braking torque of the magnets and approximations
Different load torques are usually considered to be constant, shown for wm = 0 . 6 ~ ~ .
linear with the speed or related to the square of the speed or
a combination of the three assumptions. A linearization is re-
quired if a term proportional to the square of the speed is to B. Dynamic equation
be taken into account in the study of the synchronization. The
The dynamic equation with the different torques can be writ-
linearized load torque, described as follows, is higher than the
ten as follows :
one given by the original model for a speed higher than wm :

Due to the fact that Cl =


P
and9 = sw, = ws - w and
with (7), (6),(9), (3), the following equation is obtained, which
Due to non-synchronous speed, the synchronous speed w8 is similar to (1) :
and the rotor speed w do not match during the start. Hence,
the ’synchronous’part of the motor sees a short-circuit instead
of the mains [ 5 ] . This creates a braking torque that can be
expressed as :
The damping factor of the drive is identified as

425
Tbl is positive so it decreases the value of the damping fac- Depending on additional properties of the Lyapunov function
tor : the braking torque of the magnets is to be reduced if the (its domain of definition is one of them), the stability can be
synchronizationprocess needs to be improved. The term Tl1 is further qualified.
negative if the load is linear or quadratic to the speed. There- If B is a ball around the equilibrium point where a Lyapunov
fore, the worst case is to synchronize a load which is constant function can be defined, any trajectory starting in B remains
with the speed. in this area and converges to the equilibrium point. The ball
The terms P(6) and 6 are linked to the components of the B is called a domain of attraction. For the synchronization
torques and the number of poles p by P(6) = $Ts and Pl = process, it means that if the slip at a given load angle is low
5 (Tho + Go). enough (within the ball), the synchronization will inevitably
take place. The idea is then to study the Lyapunov function and
111. EXPERIMENTS
A N D SIMULATIONS define its largest domain of attraction. This domain should be
The start and synchronization of a 4 pole-LSPM motor were as close as possible to the true stable region of the system, and
recorded. A position sensor gave the position every 0.2 ps. rather smaller. A criterion of synchronization is obtained when
With adequate filtering, the speed is obtained and therefore one can define a condition so that all the trajectories enters the
the slip. The load angle is also calculated by performing a attraction domain. The critical load torque can then be defined
dq transformation of the voltages. The initial value of the load as the maximum load torque for which the criterion is fulfilled.
angle is also taken from the measurements. The simulation However, the first main difficulty is to construct the Lya-
with the classic dq model starts at w = 0 whereas it starts at punov function. In this study, the Lagrange-Charpit method is
w = sa,,ws (see equation 16) for the simplified model pre- used as Miyagi and Taniguchi did in 1981 [6] and Senjyu and
sented above (figure 2). Uezato in 1998 (for the reluctance motor) [4].

A. 1 Construction of the Lyapunov function


1.4

The dynamic equation (1) is rewritten in the following form


1.2 - . . . . . . . . . . . . . . . . .

&El = 22
dt
= w*s
08- 9 = -IC22 - h(z1) (12)
::
06-
with 6 = 21+ 60, 3 = 22, IC = D / J , and h ( z l ) =
04-
(P(z1 + 60) - Pi) / J where 60 is the equilibrium point to be
qualified. The function constructed with the Lagrange-Charpit
02- method [6] is given by :

-02
0

8
5
simplifiedmodel

10
,
15
-v
m 25
V(Z1, 22) =
... +
f [ z 2 + a.K(z1)12+ ...
JOzl h(2)da:+ ...
load angle (fad)
... + +(l - a)[K(21)I2+ ... (13)
+2 So"' Ja(1 - a)kK(z)h(z)dz
Fig. 2. Simulations and test of the start of a 4-pole LSPM
motor. where K(z1) = k z l , 0 5 a 5 1 and a is an arbitrary
constant. The value of a is discussed later in the article. As this
The measurements and the simulations are in good agree- function fulfills the requirements to be a Lyapunov function
ment, the synchronization being harder with the developed [7], the equilibrium point 60 is stable. The asymptotic stability
model than in reality, as expected. The model has been val- can also be proved by choosing a function equal to the sum of
idated, and now the study of the synchronization criterion can V for a = 0 a n d a = 1.
be conducted.
A.2 Stable region and limit cycle
Iv. LYAPUNOV
FUNCTION A N D ONE CRITERION OF
SYNCHRONIZATION The evolutions of the slip versus the load angle or trajectory
in the phase plane is a good tool to study what happens dur-
A. Lyapunovfunction and stability ing the synchronization. Each trajectory is obtained by solving
Lyapunov's direct method can be used to estimate the sta- numerically the dynamic equation ( 1 ) using the Runge-Kutta
bility of linear or non-linear systems without solving the sys- method for a given set of initial values for the slip and the
tem's differential equation [7]. If a Lyapunov function can be load angle. Trajectories with different initial conditions are
defined around an equilibrium point, the stability of this equi- presented in the phase-plane in a case of successful synchro-
librium point can be asserted by the Lyapunov direct method. nization (figure 3).

426
06-

04-

02-

L1
B 0-

-02 -

-04 -

-10
1 5 0 5 10 15
I I
10 12 I4 16 I8 20 22
x l (rad) load angle (rad)

Fig. 3. Phase plane in a case of successful synchronization. Fig. 4. A successful synchronization with the attraction do-
main for cy = l.
It can be seen that for certain initial conditions (for exam-
ple when the initial slip is higher than 0.35 in figure 3), the
synchronization occurs for 21 = 2n or even 4n but not at the 0 14 -
studied equilibrium point $1 = 0. This phenomenon makes the
study more complicated because the defined function V is not 012 -
a Lyapunov function for all the phase plane (it is not positive -
01
definite everywhere).
The limit of the attraction domain related to V is obtained 008 -
by calculating the critical energy level V, which is :
oca -
-
n
vc= V(ZC = 6, - 60,O) (14) II)

OM -
and solving :
OM -

0- A :
1 :c
6, is the other load angle corresponding to the equality of attractiondomain
8
; ;
,

-
the load torque and the synchronous torque, 60 being the sta-
002 t% r I
1
ble solution and 6, the unstable solution (the derivative of the
synchronous torque to the load angle is negative for 6,).
O M cI
10 15 20 25
1 I
load angle (rad)
Figure 4 presents an example of trajectory when the syn-
chronization succeeds with the Lyapunov domain of attraction Fig. 5. One non synchronization trajectory (limit cycle) and
calculated for cy = 1. attraction domain for a = 0 .
In figure 5 , an example of trajectory when the synchroniza-
tion fails with the Lyapunov domain of attraction calculated for
cy = 0 are shown. The slip doesn’t reach a value low enough to ‘A.3 Synchronization capability
enter the domain of attraction and converge to the equilibrium The idea is to compare the average slip of the limit cycle sa,
point 60. It should be noted that the stable region depends on with a slip related to the domain of attraction defined with the
the load and the motor parameters. Lyapunov function. In [4], the authors use the slip SO, given by
If the motor doesn’t synchronize, the speed pulsation is a the equation V, = V(0,sows) so that :
cyclic function of the load angle 6 with the period 2n. Integrat-
ing the dynamic equation (1) from an arbitrary point (&, sa) to
+
(6, 2n,s a ) by 6 gives the average slip of what can be called
so =
a
-.
Ws
a limit cycle. This average slip is denoted sa, and is equal to :
Therefore, the point (&,so) is the frontier of the attraction
domain. Using the criterion sa, < SO for different values of cy
was considered. However, these solutions were rejected as for

427
high values of inertia, this criterion proved to give an overesti-
mated value of the critical load torque compared to the results < 7 0

obtained with the simulation of the transients with the simpli-


fied model. Instead, it was chosen to use the mean value of
the slip corresponding to the positive part of the attraction do-
main (part with A, B and C figure 5 ) , denoted S d o m and called
domain slip in the following.
In order to have only one value for each load angle (and have
an easier way to calculate S d o m ) , a = 0 has to be chosen. It
can be seen that the attraction domain calculated with a = 1
doesn’t fulfill the requirement as for some load angle two pos-
itive slips can be found on the frontier of the attraction domain
(figure 4). The choice of a = 0 corresponds to the minimum
of the function S d o m ( ( Y ) so it guarantees that the defined attrac-
tion domain is smaller than the stable region. It also simplifies
the expression of the Lyapunov function as follows : 0 20 40 60 80 100 120
load torque (Nm)

Fig. 6 . Domain slip, average slip and critical load torque.

1
vo = + f(z1). (18)
The expression of f(z1)is given in the appendix B. The
Lyapunov function for a = 0 is proportional to the total me-
chanical energy of the rotor, the first term corresponding to the
kinetic energy. The critical energy level VCois then defined by

KO= f ( z c ) = f(6m - 60). (19)


The value of S d o m can be calculated as :

with 61 the load angle corresponding to point A in figure 5 .


The criterion S d o m < sa,, gives a maximum load torque that Fig. 7. Influence of the mains voltage on the critical load
can be synchronized (figure 6). torque.
Varying the mains voltage of the motor, the critical load
torque is calculated with the presented criterion and with nu-
merically solving the dynamical equation (1) and checking the says that the voltage should be 26% higher than the measured
steady state (figure 7). one to reach the limit. With the transient simulation of (l), the
The defined criterion gives a critical load torque that is un- critical voltage is 24% higher with the measured value. Both
derestimated compared to the value given by the simulation of methods gives an underestimation of the critical synchroniza-
the dynamic equation (1). The critical load torque obtained tion conditions as expected. The transient simulation give a
with the criterion is about 60% of the one calculated by sim- better answer but it requires 60 times more computing time
ulating the transients with the simplified differential equation than the criterion.
(1). This is due to the fact that the attraction domain is smaller Simulations with the dq model says that the limit of syn-
than the real stable region. chronization happens for the nominal voltage of 400 V. To ob-
Synchronization problems were encountered when tests tain that value, 3000 times more time is required than with the
were conducted on a four-pole three phase 7.5kW motor criterion and the results lead to an overestimated value. It is
loaded by a DC generator. The axis of the drive had an inertia thought that this is due to the fact that the temperature of the
that was 17 times higher than the one of the motor itself. It rotor varies during the transients, phenomena that is not taking
was measured that with a mains phase-to-phase voltage of 426 into account. One could probably obtain an underestimated
V, the drive was at the limit of synchronization. Our criterion value with the dq model by choosing the worst case for each

428
parameter. However, the dq model would still require a huge
100
computing time compared to the criterion. 1
Therefore, the presented criterion is considered to fulfill the
current requirements. Having this criterion, it is possible to
compare designs even though the absolute value of the critical
load torque is not so accurate. An improved criterion may be
required if the optimization procedure produces designs that
present limited synchronization capability. This has not been
the case yet when the motors were tested in the normal applica-
tion conditions. In any case, it is intended to check the starting
transients of the optimized designs with a dq model and finite
element time-stepping analysis before considering a prototype
@I.
, I
m
v. INFXUENCE OF PARAMETERS O E4 103 I50
no-load voltage (V)
250 3w

A criterion that enables to calculate the critical load torque


without running simulation of the starting transients is a suit- Fig. 8. Influence of no-load voltage on the critical load torque.
able tool to make a parametric study. The influence of the
different parameters can be studied and different motors can
be compared. Two four pole 7.5 kW three phase LSPM mo- only proportionally to the no-load voltage, it happens in this
tors are considered. They have the same external dimensions, case that the increase of the synchronizing energy is more than
but motor 2 has an increased amount of magnets to improve compensated by the lower speed at which the synchronization
its steady-state performances. The nominal effiencies are re- occurs. Therefore, an increase of the no-load voltage can lead
spectively 86.2% and 91.2% respectively for motor 1 and mo- to a reduced critical load torque for motor 2 whereas it does not
tor 2 . The nominal power factor is also better for motor 2 for motor 1. An optimum can be found for the no-load voltage
(cos9 = 0.99) than for motor 1 coscp = 0.88. The rotor resis- of motor 2 to maximize the critical load torque.
tance of motor 2 was increased compared to motor 1 in order It should be noticed, however, that increasing the no-load
to improve the early start as the braking torque of the magnets voltage implies to have more magnets. It has the effect to de-
was increased. No consideration were made about the synchro- crease the d-inductance as well. This was not taken into ac-
nization performances during the design. count here as only one parameter is varied at a time, except for
The influence of the no-load voltage and the rotor resistance the influence of the temperature.
are presented on figure 7 and 8 respectively. Table 1 gathers the relative sensibilities of the critical load
A higher voltage leads to a higher critical load torque. The torque related to the different parameters for motor 1 and motor
asynchronous torque being higher with the voltage, the syn- 2 respectively. The mains voltage is the most influent param-
chronization occurs at a lower slip and a lower energy is re- eter on the critical load for both motors. Therefore, an index
quired to perform the synchronization. That is something that of 100 is given to the sensibility related to the mains voltage
should really be taken into account as in most applications it is for each motor. The sensibility of the critical load torque to the
specified that the drive should keep certain performances even mains voltage are 0.48N.mlV for motor 1 and 0.43N.mlV
if the mains voltage is lower than its nominal value. for motor 2. A positive value, as for the mains voltage, means
In figure 8, it is shown that the influence of the no-load volt- that a higher value of the parameter helps the synchronization.
age on the critical load torque is reverse for the two considered After the mains voltage V , several parameters have great
drives. It is believed that this is a very special and interest- influence on the synchronization capability. For motor 2, the
ing result. That the synchronization capability of a motor in- rotor resistance R, has almost the same influence as the mains
creases with the amount of magnets as for motor 1 is easy to voltage but is reverse. As increasing the rotor resistance has a
understand : more magnets means more synchronous torque benifitial effect on the early start, one has to find a compromise
and more energy for synchronization. However, that is not true for the design between the early start and synchronization per-
for every value of the no-load voltage for motor 2. The rotor formances when choosing R,. Increasing the rotor resistance
resistance of motor 2 is very high compared to a conventional increases the slip at which the synchronization occurs. There-
value. This means that the slope of the asynchronous torque fore more energy is needed to synchronize and a lower value
near the synchronous speed Talis exceptionally low. Due to of the critical load torque is obtained.
the presence of the magnet braking torque, which is propor- As well as the no-load voltage, the influence of the d-
tional to the square of the no-load voltage, the effective slope reactance depends on the considered motor. As it was pre-
or the damping constant D has a value that is much dependent viously written, a change of EO affects the value of X d s . In-
on the no-load voltage. As the synchronous torque increases creasing the amount of magnets increases the no-load voltage

429
TABLE I Eo no-load voltage at synchronous speed
INDEXES OF SENSIBILITY OF THE CRITICAL LOAD TORQUE J inertia
(CRITERION). R, stator resistance
R, rotor resistance
sensibility motor 1 sensibility motor2 T torque
100 100 V mains voltage
+33 -26 Xds direct reactance at synchronous frequency
10 1 X , magnetizing reactance
2 6 X,, quadrature reactance at synchronous frequency
-3 -3 X , rotor reactance at synchronous frequency
-4 -7 p number of poles
-8 - 16 s slip
-8 -30 6 load angle
-16 - 84 w pulsation
-36 +24 AO, temperature rise of the stator
AO,l, temperature rise of the rotor compared to the stator
and decreases the d-reactance. Therefore, the two effects add. Formulas
For motor 1, increasing the volume of magnet gives a better Synchronous torque :
synchronization capability, it is the contrary for motor 2. T - 3.P .&.V
To obtain an underestimated value of the critical load torque, - 2.W. X d .
T 3.p .v2.(Xds-Xqa)
one should then consider : the lowest mains voltage, the high- s2 - 4.W. xd..xqs

est temperature for the rotor and for the stator and the unsat- Asynchronous torque :
urated value of the d-reactance for motor 1 and the saturated +
a1 = RZ. (R: ( X , X,)2) +
value for motor 2. bl = 2.&.R8.X$
However, these conclusions should be taken with care as al- c1= (X,.X, + +
x,.x, x,.x,)~ R:. (x, xT12 + +
ready with two considered motors, it was found out that the * Part of the Lyapunov function for Q = 0 :
f(z1)= &.[T,l. + +
influence of parameters are not the same for every design. The
rapidity of use of the criterion enables to run a parametric study
easily. Checking the influence of the no-load voltage could be
... + +. (cos (bo) - cos ( 2 1 bo)) ...
+
(cos (2.60) - cos (2(21 do)))] - 9.21
introduced in the optimization procedure as problems of con- Parameters of the 7.5 kW four pole LSPM prototypes
vergence may be encountered if a condition on the critical load
torque is introduced. I Darameters I motor2 I motor 1 I

VI. CONCLUSIONS
A simple method to define the maximum load torque that
can be synchronized by LSPM motors has been presented. The
influence of the electrical parameters on the synchronization
could then be studied with the introduced criterion. The ap-
plied voltage and the rotor resistance are the main parameters J (kg.m2) 0.035 0.035
to be considered. A worst case for the synchronization can sm 1.12 0.15
now be considered to calculate the critical load torque doing a
preliminary parametric study with the criterion. ACKNOWLEDGMENT
It was shown that the tested motor has greater synchroniza-
tion capability than the criterion predicts. It is known from the This investigation is a part of the Permanent Magnet Drives
chosen assumptions that the critical load torque can be quite program within the Center of Excellence in Electric Power En-
underestimated compared to the results from transient simu- gineering, in cooperation with ITT Flygt.
lations. An improved criterion may then be required in the REFERENCES
future if the optimized designs present limitations in their syn-
chronization capability. [ 13 T. Miller, ”Synchronization of line-start permanent-magnet
AC motor,” ZEEE Trans., vol. PAS-103, July 1984, pp 1822-
APPENDIX 1828.
[2] T. Miller, ”Transient performance of permanent magnet AC
Notations machines,” ZEEE ZAS Annual Meeting 1981, Philadelphia, pp
D damping constant 500-503.

430
[3] M. Rahman, A. Osheiba, T. Radwan, ”Synchronization
process of line-start permanent magnet synchronous motor,”
Electric Machines and Power Systems, vol. 25, 1997, pp 577-
592.
[4] T. Senjyu, K. Uezato, ”Pull-in criterion for reluctance mo-
tors determined by Lyapunov method,” Electric Machines und
Power Systems, vol. 26, April 1998, pp 331-345.
[5] U. Herslof, ”Design, analysis and verification of a line
start permanent magnet synchronous motor,” Licentiate thesis,
Royal Institute of Technology of Stockholm, 1996.
[6] H. Miyagi, T. Taniguchi, ”Lagrange-Charpit method and
stability problem of power systems,” Proc. IEE, vol. 128, Mars
1981, pp 117.
[7] J. Slotine, W. Li, Applied Non-Linear Control, Englewood
Cliffs, NJ: Prentice Hall.
[ 81 L. Lefevre, J. Soulard, ”Finite element time-transient start
of a line-start permanent magnet synchronous motor,” Con$
Rec. ICEM, 2000.

43 1

You might also like