Professional Documents
Culture Documents
net/publication/3483728
CITATIONS READS
72 466
All content following this page was uploaded by Cheng-Kai Lin on 06 August 2015.
ISSN 1755-4535
Abstract: A novel, advanced position controller design for an interior permanent magnet synchronous motor
control system is proposed. The input – output linearisation technique is used to transfer the system model
into a linearised system model. Then, based on the linearised system model, an H1 controller is designed to
achieve robust performance of the position control system. To improve the system performance, a load
estimator is used to compensate the external load and the influence of the parameter variations as well. In
addition, a maximum torque/ampere control is applied to increase the output torque of the motor. A
digital signal processor, TMS 320LF2407, is used to execute the speed-loop and position-loop control
algorithms. As a result, the hardware circuit is quite simple. Several experimental results show that
the proposed system has fast transient responses, good load disturbance rejection responses and good
tracking responses.
14 / IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26 & The Institution of Engineering and Technology 2008
doi: 10.1049/iet-pel:20070177
www.ietdl.org
drives [1–6], only a few researchers have focused on the proposed nonlinear controller is more intuitive than
feedback linearisation technique to control an IPMSM. other controllers. The reason is that the nonlinear
As we know, the feedback linearisation technique has controller design is deeply rooted in the physics of the
many advantages. For example, this method uses plants. Recently, the DSP has become very powerful;
differential geometric transformation to transform as a result, the applications of nonlinear control have
nonlinear system dynamics into a linearised system. As become more and more popular as well. The paper is
a result, many well-developed linear control organised as follows. The mathematical model of
techniques can be applied. The feedback linearisation the IPMSM is discussed first. Then, the details of the
technique has been successfully used in helicopters, input-output linearisation technique applied in the
aircraft, robots and biomedical devices [7]. Several IPMSM are explained. In addition, the design of an
researchers have applied the feedback linearisation H1 controller based on the linearised system is
technique to different motor drive systems. For presented. Next, the implementation of the proposed
example, Zribi and Chiasson [8] proposed an exact system is described. After that, several experimental
linearisation method to control the position of a PM results are shown. Finally, some conclusions are given.
stepper motor. Chiasson [9] used nonlinear differential
geometric techniques for control of a series DC motor.
Chiasson [10] proposed a dynamic feedback 2 Mathematical model
linearisation control of an induction motor. In Assuming that the IPMSM is of three-phased with
addition, Boukas and Habetler [11] implemented an balanced windings and no saturation, the voltage
exact feedback linearisation with state and state equations of the IPMSM in the d–q synchronous
derivative feedback for an induction motor speed frame can be expressed as
control. Panda and Dash [12] studied the application
of a feedback linearisation approach to switched did
reluctance motors. The published papers used the vd ¼ Rid þ Ld Po vr Lq iq (1)
linearisation feedback technique to different motor dt
drive systems [8 –13]; Recently, Chiasson [13] has diq
proposed a linearisation feedback technique applying in vq ¼ Riq þ Lq þ Po vr Ld id þ Po vr lm (2)
dt
IPMSM. However, the external load and parameter
variations are not considered. In addition, only where vd is the d-axis voltage, R the stator resistance, id
simulation results are shown and no experimental the d-axis current, Ld the d-axis inductance, d/dt is the
results are included [13]. The paper proposes some differential operator, Po the pole pair of the motor, vr
improvements on previous paper [13]. First, a load the motor speed, Lq the q-axis inductance, iq the q-axis
estimator is developed to estimate the external load current, vq the q-axis voltage and lm the flux linkage.
with the variations of the motor parameters. In
addition, an H1 controller is used to obtain better The electromagnetic torque can be described as
performance, including fast transient responses, good
load disturbance responses and good tracking 3
responses. Moreover, a maximum torque/ampere Te ¼ Po [lm iq þ (Ld Lq )id iq ] (3)
control of the IPMSM is used in this paper as well. To 2
the authors’ best knowledge, no researcher has done
the realisation of the feedback linearisation technique where Te is the electromagnetic torque of the motor.
applied to an IPMSM drive. This reason motivated us
to study this topic. The idea of this paper therefore is The dynamic equation of the speed is
new and interesting. In addition, by using the DSP to
execute the controller, several experimental results d vr 1
show the feasibility of the proposed nonlinear control ¼ (Te TL Bvr ) (4)
dt Jm
algorithm. Although the proposed nonlinear in the
paper is complicated, it has several advantages. First, The dynamic equation of the position is
the proposed nonlinear controller can handle the
nonlinearities in a large-range operation directly. As
we know, when a linear controller is used to control a dur
¼ vr (5)
system, it requires to neglect the nonlinear terms. dt
The linear controller’s accuracy thus quickly degrades
as the speed of motion increases. The proposed In this paper, both the inertia and viscous friction
nonlinear controller can solve the difficulty. In coefficient are considered. As a result, a more general
addition, the proposed nonlinear controller description of the mathematical model can be
intentionally considers the nonlinearities; as a result, obtained. The effect of B is negligible in some papers
the model uncertainties is tolerated. Moreover, the because of its small effect.
& The Institution of Engineering and Technology 2008 IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26/ 15
doi: 10.1049/iet-pel:20070177
www.ietdl.org
3 Nonlinear position controller Next, comparing (10) and (6), we can obtain
design 2 3
Rid Po Lq iq vr
According to (1) – (5), the dynamic equation of the þ
motor can be expressed as 6 Ld Ld 7
6 7
6 P v L i Ri P v l 7
2 3 6 o r dd o r m
q 7
_id 6 7
f (x) ¼ 6 Lq Lq Lq 7
6 _i 7 6 7
6 q 7 6 3Po iq lm þ 3Po (Ld Lq )id iq v B TL 7
6 7 6 r m
7
4 v̇ r 5 4 5
2Jm Jm Jm
u̇ r vr
2 3 (11)
Rid Po Lq iq vr
6 þ 7
6 Ld Ld 7 and
6 P v L i Ri P v l 7
6 o r d d
q
o r m 7 T
6 7
¼6 Lq Lq Lq 7 x ¼ id iq vr ur
6 7
6 3Po iq lm þ 3Po (Ld Lq )id iq v B T 7 T
6 r m L7 g1 ¼ 1=Ld 0 0 0
4 5
2Jm Jm Jm T
vr g2 ¼ 0 1=Lq 0 0
2 3 2 3
1=Ld 0
6 0 7 6 1=L 7 In fact, the real system includes an unknown external
6 7 6 q7 load TL , and the deviations of the motor parameters
þ6 7v d þ 6 7v (6)
4 0 5 4 0 5q which include DJm and DBm [16]. Then, the total
0 0 uncertainty can be expressed as
d vr
As you can see, (6) is a nonlinear dynamic equation. As a Te ¼ Jmo þ Bmo vr þ Td (12)
result, the linearisation technique, the external load dt
estimator and the H1 controller can be designed. The dv
details are discussed as follows. Td ¼ DJm r þ DBm vr þ TL (13)
dt
3.1 Input– output linearisation technique where DJm ¼ Jm Jmo and DBm ¼ Bm Bmo are the
Lie derivate is a well known mathematical operator to deviations of the motor parameters, and Jmo , Bmo are
obtain a feedback linearisation system [7]. The technique the nominal parameters of the motor. By substituting
is called input–output linearisation technique. In this (12) and (13) into (4), we can obtain
paper, the authors use Lie derivative to transform a
nonlinear system into a linearisation system. First, we 3Po iq lm þ 3Po (Ld Lq )id iq vr Bmo Td
define the following operators [7, 14, 15] v̇ r ¼
2Jmo Jmo Jmo
X
n
@h (14)
Lf h ¼ rh f ¼ fi (x) (7)
i¼1
@xi
Then, the system dynamic equation can be rearranged as
Lif h ¼ Lf L(i1)
f h ¼ r L(i1)
f h f (8)
By using the same way, we define [7, 14] x_ ¼ f^ (x) þ g1 vd þ g2 vq þ g3 DTd (15)
Lg Lf h ¼ r(Lf h) g (9)
where
Then, according to (1)–(3), we can rearrange the
dynamics of the system as follows DTd ¼ Td T^ d
T
x_ ¼ f (x) þ g1 vd þ g2 vq (10) g3 ¼ 0 0 1=Jmo 0
16 / IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26 & The Institution of Engineering and Technology 2008
doi: 10.1049/iet-pel:20070177
www.ietdl.org
and and
2 3
f^ 1 (x) 3Po (Ld Lq )iq
6 7 Lg1 L2f^ h1 ¼ (25)
6 f^ 2 (x) 7 2Jmo Ld
f^ (x) ¼ 6
6 f^ (x) 7
7
4 3 5 3P lm þ 3Po (Ld Lq )id
Lg2 L2f^ h1 ¼ (26)
f^ 4 (x) 2Jmo Lq
2 3
Rid Po Lq iq vr
þ Rid Po Lq iq vr
6 Ld Ld 7 Lf^ h2 ¼ þ (27)
6 7 Ld Ld
6 P v L i Ri P v l 7
6 o r dd o r m
q 7
6 7 Bmo
¼6 Lq Lq Lq 7 Lg3 L2f^ h1 ¼ (28)
6 7 2
Jmo
6 3Po iq lm þ 3Po (Ld Lq )id iq v B T^ 7
6 r mo d 7
4 5
2Jmo Jmo Jmo 3Po (Ld Lq )iq Bmo
vr L3f^ h1 ¼ f^ 1 (x) f^ (x)
2Jmo Jmo 3
(16)
3Po lm þ 3Po (Ld Lq )id
where ‘ˆ’ is the symbol of the estimating value. þ f^ 2 (x) (29)
2Jmo
In this paper, we choose the d-axis current and the
rotor position as output variables [7]. The aim of the 1
controller is to maintain the position and the d-axis Lg3 Lf^ h1 ¼ (30)
current in the trajectories as we want. The new state Jmo
variables can be defined as 1
Lg1 h2 ¼ (31)
z1 ¼ ur ¼ h1 (x) (17) Ld
To effectively control the system, the d – q axis voltages
z2 ¼ z_1 ¼ vr ¼ Lf^ h1 (x) (18) can be selected as
" #
vd u1 L3f^ h1 Lg3 Lf^ h1 T_^ d DL3f^ h1 (x)
z3 ¼ L2f^ h1 (x) ¼ D(x)1
vq u2 Lf h2 DLf^ h2
3Po lm iq þ 3Po (Ld Lq )id iq vr Bmo T^ d (32)
¼
2Jmo Jmo Jmo
Where DL3f^ h1 and DLf^ h2 are the parameter variations
(19) of the Lie derivatives. Then u1 and u2 are new control
variables. The D(x) can be expressed as
z4 ¼ id ¼ h2 (x) (20)
L L2 h Lg2 L2f^ h1
D(x) ¼ g1 f^ 1 (33)
In addition, the dynamic equation of the system can be Lg1 h2 0
expressed as
By substituting (32) into (23) and (24), we can obtain
z_1 ¼ z2 (21)
z_3 ¼ u1 þ Lg3 L2f^ h1 DTd DL3f^ h1 (x) (34)
z_2 ¼ z3 þ Lg3 Lf^ h1 DTd (22)
z_4 ¼ u2 DLf^ h2 (35)
z_3 ¼ L3f^ h1 (x) þ Lg1 L2f^ h1 vd þ Lg2 L2f^ h1 vq
Because of the influence of the parameter variation,
þ Lg3 Lf^ h1 T_^ d þ Lg3 L2f^ h1 DTd (23) measuring error and external load, the nonlinear
system cannot match the linear decouple Brunovski
canonical form [14]. As a result, the performance of
z_4 ¼ Lf^ h2 þ Lg1 h2 vd (24) the system is deteriorated. If we just consider the
ideal case, which means that DL3f^ h1 ¼ 0, DTd ¼ 0
& The Institution of Engineering and Technology 2008 IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26/ 17
doi: 10.1049/iet-pel:20070177
www.ietdl.org
and DLf^ h2 ¼ 0, then it is possible to select a linear state solve the problem, a new variable is defined as
feedback control method as follows
...
xc1 ¼ T^ d L1 vr (42)
u1 ¼ ur þ k11 ðu€ r z3 ) þ k12 (u_ r z2 ) þ k13 (ur z1 )
Then, it is not difficult to obtain
(36)
and x_c1 ¼ T_^ d L1 v̇ r
_ B 1
u2 ¼ k21 (id z4 ) þ _id (37) ¼ T d L1 mo vr
^ T^
Jmo Jmo d
Equations (36) and (37) show the pole placement 3Po lm 3Po (Ld Lq )id iq
technique applying in the IPMSM drive system. þ i þ (43)
2Jmo q 2Jmo
3.2 External load torque estimator Assuming that Td and T^ d are close enough, we can easily
obtain
In the real world, the motor is used to convert the
electrical energy into mechanical energy. As a result,
B 1 3P l
an external load is added to the drive system. For a x_c1 ¼ L1 mo vr þ T^ d o m iq
fixed sampling interval of the voltage control loop, Jmo Jmo 2Jmo
the external load does not change abruptly. The major
3Po (Ld Lq )id iq
reason is the sampling interval is very short. As a (44)
result, for each sampling interval, we can assume that 2Jmo
the external load Td is a constant, and its derivative T_ d
is equal to zero. According to (3) and letting T_ d ¼ 0, By letting e ¼ Td T^ d , we can obtain
we can obtain
L
2 3 _e ¼ T_ d T_^ d ¼ 1 e (45)
Bmo 1 Jmo
v̇ r vr
¼ 4 Jmo Jmo 5
T_ d 0 0 Td By suitably selecting L1 , it is not difficult to make the
2 3 2 3 estimating error converge. In the real world, we can
3Po lm 3Po (Ld Lq )id iq obtain the estimating load as T_^ d ¼ x_c1 þ L1 v̇ r , which
þ 4 2Jmo 5iq þ 4 2Jmo 5 (38) can be obtained from (43). The dynamic behaviour of
0 0 the external load estimator can be determined by the
gain L1 . In addition, the influence of the load
and disturbance can be effectively compensated for. Fig. 1
shows the block diagram of the external load torque
estimator. The T^ d can be obtained from the input
vr information vr, iq and id and using the parameters of
y o ¼ vr ¼ 1 0 (39)
Td the motor and the gain L1 as well. In the real world,
the estimated load is not equal to the external
Now, we can define the load estimator as load. There exists estimated error. However, the
closed-loop system can effectively compensate the load
disturbance although the estimated error appears.
T_^ d ¼ T_ d þ L1 (v̇ r v_^ r ) (40)
Substituting (40) into (38)
_T^ ¼ L v̇ þ Bmo v þ 1 T^
d 1 r
Jmo r Jmo d
3P l 3Po (Ld Lq )id iq
o m iq (41)
2Jmo 2Jmo
where L1 is the gain of the load estimator.
Unfortunately, in the real world, it is difficult to
obtain v̇ r without causing high-frequency noise. To Figure 1 Block diagram of the external load estimator
18 / IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26 & The Institution of Engineering and Technology 2008
doi: 10.1049/iet-pel:20070177
www.ietdl.org
3.3 H1 controller design robust. As a result, the H1 controller can reach its
desired responses without seriously being influenced
To reduce the influence of the parameter variation, by the perturbations. Then, the proposed controller
measuring error and external load, an H1 is proposed works well. As we know, we can consider the worst
here to achieve the maximum torque/ampere control. case of the H1 controller. The Du1and Du2 are
By suitably adjusting the d-axis current, a maximum related to the nonlinear perturbations. In the real
torque/ampere characteristic can be achieved. In world, it is impossible to obtain Du1 and Du2 .
addition, the H1 controller can compensate for the However, for every plant, the parameter variations
uncertainty caused by the parameter variation, and external load have their limitations. As a result,
measuring error and external load disturbance. it is possible to set upper bounds for Du1 and Du2 . In
this paper, we assume jDu1 j 0:1ju1 j, and
The relationship of the d- and the q-axis currents jDu2 j 0:1ju2 j, which means that the allowed
that can produce the maximum torque/ampere variations of the control inputs are below +10%. If
control is developed in previous papers [6, 17, 18], the variations of the control inputs beyond +10%, the
and can be expressed as performance of the system is deteriorated. Then, new
upper bounds should be selected. However, in the real
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi world, the influence of the uncertainties is not so
lm l2m serious. Generally speaking, the u1 and u2 cannot be
id ¼ þ i2 (46)
2(Ld Lq ) 4(Ld Lq )2 q zero under transient responses and load disturbance
conditions. Then, Du1 and Du2 are not zero as well.
However, when the external load is zero and the
In (46), we assume that the d-axis inductance is not system reaches steady state. The u1 and u2 can become
equal to the q-axis inductance. If the d-axis inductance zero if the parameter variations do not exist. If the
is equal to the q-axis inductance, then the motor is a parameter variations exist, then the steady-state error
surface-mounted permanent synchronous motor. The appears. In this situation, the u1 and u2 are not zero,
d-axis current command is set as zero when the and the Du1 and Du2 appear to force the system
surface mounted motor is operated in the constant reducing its steady-state error. From (43) and (44), we
torque region. Equation (46) therefore cannot be can redefine the new state variables as [13]
applied for the surface mounted permanent magnet
synchronous motor. According to (23) and (24), we
can obtain x_h2 ¼ (A þ DA)xh2 þ (B þ DB)uh2 þ Dwh2 (53)
...
z_3 ¼ u r ¼ u1 þ Du1 (47)
z_4 ¼ id ¼ u2 þ Du2 (48) zh2 ¼ C1 xh2 þ D12 uh2 (54)
and where
2 3
u1 ¼ L3f^ h1 þ Lg1 L2f^ h1 vd þ Lg2 L2f^ h1 vq þ Lg3 Lf^ h1 T_^ d (49) 0 1 þ a12 0 0 0 0
6 0 0 0 0 0 07
6 7
6 7
6 0 0 0 0 0 07
u2 ¼ Lf^ h2 þ Lg1 h2 vd (50) A þ DA ¼ 6
6 1
7
6 0 0 0 0 077
6 7
4 0 0 0 1 0 05
Du1 ¼ DL3f^ h1 (x) (51) 0 0 1 0 0 0
2 3
0 0
Du2 ¼ DLf^ h2 (52) 61 þ b 0 7
6 21 7
6 7
6 0 1 þ b32 7
To design the H1 state feedback controller, we should B þ DB ¼ 6
6 0
7
7
assume that the Du1 and Du2 are uncertainties and 6 0 7
6 7
have their bounded values. In the real world, the 4 0 0 5
nonlinear perturbations are unknown and are not
constant values. However, the H1 controller is quite 0 0
& The Institution of Engineering and Technology 2008 IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26/ 19
doi: 10.1049/iet-pel:20070177
www.ietdl.org
where eu1 is the position error, _eu1 the derivative of the By using the same method, we can design an H1
position error, eu2 the integration of the position error, speed controller. Because of the limitation of the
ei2 the integration of the d-axis current error, ur the scope of this paper, only some important results are
position command and id the d-axis current command. discussed here. First, from (36) and (37), we can obtain
In the real world, the percentages of the parameter
variations a12 , b21 and b32 are unknown. However, u1 ¼ v̈ r þ k31 (v̇ r z3 ) þ k32 (vr z2 ) (58)
they have variation limitations. As a result, according
to the real situations, we can set bounds when we u2 ¼ k31 (id z4 ) þ _id (59)
design the controller. In this paper, we select the
20 / IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26 & The Institution of Engineering and Technology 2008
doi: 10.1049/iet-pel:20070177
www.ietdl.org
Figure 4 Block diagram of the nonlinear speed controller Figure 6 Implemented circuit of the drive system
& The Institution of Engineering and Technology 2008 IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26/ 21
doi: 10.1049/iet-pel:20070177
www.ietdl.org
4 Implemented system
The implemented system is shown in Fig. 5. A digital
signal processor is used to execute the whole control
algorithm. The a- and b-phase currents are obtained
by using Hall-effect current sensors. Then, the A/D
converters are used to convert the analogue signals
into digital signals. Next, the current signals are read
by the DSP. In addition, the shaft position of the
motor is read by the DSP. The d- and q-axis currents
can be computed by using the a-b-c to d-q axis
coordinate transformation. The encoder used in this
paper is 1000 pulses/revolution. However, it increases
to 4000 pulses/revolution after a multiplier circuit is
used. The control algorithm includes the input –
Figure 7 Comparison of transient responses output linearisation technique, the external load
estimator, and the H1 controller, which were
where vr is the speed command. By substituting (60) discussed in Section 3 of this paper. Finally, the DSP
and (41) into (32), and then computing D(x)21, we computes the d- and the q-axis voltages to the
can obtain the control voltages vd and vq, which are inverter. A closed-loop control system with the
shown in Fig. 4. maximum torque/ampere control is thus achieved. As
22 / IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26 & The Institution of Engineering and Technology 2008
doi: 10.1049/iet-pel:20070177
www.ietdl.org
& The Institution of Engineering and Technology 2008 IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26/ 23
doi: 10.1049/iet-pel:20070177
www.ietdl.org
24 / IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26 & The Institution of Engineering and Technology 2008
doi: 10.1049/iet-pel:20070177
www.ietdl.org
& The Institution of Engineering and Technology 2008 IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26/ 25
doi: 10.1049/iet-pel:20070177
www.ietdl.org
[9] CHIASSON J: ‘Nonlinear differential geometric techniques [15] MASSOUM A, FELLAH MK, ABDELKADER M, ET AL.: ‘Input output
for control of a series DC motor’, IEEE Trans. Control Syst. linearisation and sliding mode control of a permanent
Technol., 1994, 2, (1), pp. 35– 42 magnet synchronous machine fed by a three levels
inverter’, JEE, J. Electron. Eng., 2006, 57, pp. 205– 210
[10] CHIASSON J : ‘A new approach to dynamic feedback
linearisation control of an induction motor’, IEEE Trans. [16] KIM KH, YOUN MJ: ‘A nonlinear speed control for a PM
Autom. Control., 1998, 43, (3), pp. 391– 397 synchronous motor using a simple disturbance estimation
technique’, IEEE Trans. Ind. Electron, 2002, 49, (3),
[11] BOUKAS TK , HABETLER TG : ‘High-performance induction pp. 524– 535
motor speed control using exact feedback linearisation
with state and state derivative feedback’, IEEE Trans. [17] EL-REFAIE AM, JAHNS TM: ‘Comparison of synchronous PM
Power Electron., 2004, 19, (4), pp. 1022 – 1028 machine types for wide constant-power speed range
operation’, IEEE Conf. Industry Applications, 2005, vol. 2,
[12] PANDA SK, DASH PK: ‘Application of nonlinear control to pp. 1015 – 1022
switched reluctance motors: a feedback linearisation
approach’, IEE Proc., Electr. Power Appl., 1996, 143, (5), [18] HAN SH, JAHNS TM, GUVEN MK, ET AL.: ‘Impact of Maximum
pp. 371– 379 Back-EMF Limits on the Performance Characteristics
of Interior Permanent Magnet Synchronous Machines’,
[13] CHIASSON J: ‘Modeling and High-Performance Control of IEEE Conf. Industry Applications, 2006, vol. 4,
Electric Machines’ (John Wiley & Sons, New Jersey, 2005) pp. 1962 – 1969
[14] ISIDORI A : ‘Nonlinear Control Systems’ (Springer, [19] BURL JB: ‘Linear optimal control-H2 and H1 methods’
New York, 1995) (Addison-Wesley, California, 1998)
26 / IET Power Electron., 2008, Vol. 1, No. 1, pp. 14– 26 & The Institution of Engineering and Technology 2008
doi: 10.1049/iet-pel:20070177