Anda di halaman 1dari 8

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2019.2949222, IEEE Journal
of Emerging and Selected Topics in Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1

Parameter-Free Predictive Control of IPM


Motor Drives with Direct Selection of Optimum
Inverter Voltage Vectors
M. Khalilzadeh, S. Vaez-Zadeh, Senior Member, IEEE, M.S. Eslahi

degradation under the motor parametric uncertainties due to its


Abstract— Parameter dependency is a drawback of the model model-based nature. This drawback is one of the main
predictive control of electric motor drives. In this paper, a obstacles of using MPC in plug-and-play drives, where the
predictive current control for an interior permanent magnet controlled motor parameters are unknown [11]. Many kinds of
motor independent of the motor parameters is presented. The
motor current time-derivatives (slopes) are expressed as
research have been dedicated to enhancing the robustness of
functions of the phase angles of the inverter basic voltage vectors. the method against parameter mismatch. Various disturbance
The slopes are then predicted independent of the motor observers are adopted for predicting stator currents and
parameters and are used in selecting the optimum inverter modifying the voltage reference under parameter mismatch
voltage vectors. In addition, a method is used for avoiding the [5], [12], [13]. However, observers are prone to convergence,
time-consuming evaluations of a cost function to select an parameter tuning, and stability issues. In addition, most of
optimum inverter voltage vector. By this method, the reference
current slopes are used for a direct optimum voltage vector
these approaches are related to the modulator-based predictive
selection. As a result, the control performance is improved under control method.
the parametric uncertainties and the execution time of the The effects of model parameter mismatch on the
control code is shortened in comparison with the conventional performance of FCS-PCC of a three-phase inverter are
predictive method. The effectiveness of the proposed method and investigated [14]. As a solution, a model-free FCS-PCC is
its superiority over the conventional and a recently-presented presented that uses a look-up table for storing the current
predictive current control method are confirmed through
simulation and experimental results.
vector components variations provided by application of
inverter basic voltage vectors over a switching interval [15].
Index Terms— Interior permanent magnet motors, inverters, The method thus suffers from the problem of look-up table
motor drives, model predictive control, model-free, robustness. updating, which is denoted as “stagnation of current-variation
update”. A solution is proposed to overcome the problem [16].
However, applying some non-optimal voltage vectors degrade
I. INTRODUCTION the motor drive performance in the latter method. The
updating problem is further dealt with by the reconstruction of
N owadays, the predictive control method in motor drive
applications has emerged as a promising alternative for
conventional control schemes. This control method can be
current components variations for all inverter voltage vectors
using current variations provided by three most recent inverter
implemented with or without a voltage modulator [1]-[3]. switching states [17]. Although the method is effective, it
Among the modulator-based predictive control methods, suffers from an excessive computational burden that needs a
deadbeat has been widely investigated [4]-[7]. The predictive large sampling period that deteriorates the performance of the
control without a voltage modulator is known as finite control FCS-PCC. As another solution to parameter dependency of
set (FCS) model predictive control (MPC) [8]-[10]. Based on FCS-MPC, the errors between the sampled and predicted
the motor control goals, predictive control is divided into current components are stored in look-up tables as prediction
predictive current control (PCC) [3], [5], [8] and predictive errors for inverter voltage vectors. The errors are used in
torque and flux control [6], [9], [10]. Despite many modifying the next step prediction of the current [18] or the
advantages, model predictive control is subject to performance torque [19]. Although the method needs proper gain
adjustment of a compensator, a guideline for selecting the
Manuscript received Jul 8, 2019; revised Sep 13, 2019; accepted Oct 13, gains is lacking. Furthermore, the method uses old data for
2019. error compensation in predictions used in cost function
Support of INSF through the Chair of Wireless and Contactless Power
Transfer is acknowledged. calculations, when a voltage vector is not applied for a couple
M. Khalilzadeh and M.S. Eslahi are with the School of Electrical and of sampling periods. In addition, all of the mentioned methods
Computer Engineering, University of Tehran, Tehran, Iran (e-mail: evaluate a cost function for all inverter switching states, which
M.khalilzadeh @ut.ac.ir, mseslahi@ut.ac.ir).
S. Vaez-Zadeh is with the Advanced Motion Systems Research
is time-consuming. Alternatively, the electromagnetic torque
Laboratory, and the Center of Excellence on Applied Electromagnetic slope is utilized for improving the control performance of
Systems, School of Electrical and Computer Engineering, College of motor drives [20]. It is also used for robustness improvement
Engineering, University of Tehran, Tehran, Iran (e-mail: vaezs@ut.ac.ir ).

2168-6777 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2019.2949222, IEEE Journal
of Emerging and Selected Topics in Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2

of the FCS predictive torque control against parameter interval as an optimum choice.
variations [21]. The latter method evaluates a cost function for
B. Problem of Parameter Mismatch
only three inverter voltage vectors instead of seven. However,
it needs an estimation of flux linkage, which is prone to the As mentioned before, for a two-step prediction approach,
effects of variation of winding resistance. the model equations of (1) and (2) are used twice in each
In this paper, a parameter-free FCS-PCC is proposed for sampling period. Hence, if the model is not accurate enough,
IPM motor drives. The predicted slopes of current vector there will be degradation in the control performance of the
components are used in the predictions and voltage vector machine. This is studied in this subsection.
selection. The motor parameters are not included in the It is well-known that an IPM motor is a highly nonlinear
predictions. The proposed method does not adopt look-up system. The iron saturation and cross-saturation may affect the
tables for inverter switching configurations. Hence, it does not model parameters. To include these effects to the model, a
face the stagnation problem. Furthermore, evaluations of a non-linear form of stator flux linkage is given as [24]:
cost function are avoided by using a direct selection of
optimum inverter voltage vectors. For doing so, a reference d ( id , iq ) = Ld ( id , iq ) id + m , (5)
q ( id , iq ) = Lq ( id , iq ) iq ,
voltage vector is calculated independent of the motor
(6)
parameters. The results ensure the performance superiority of
the method over the conventional and a model-free predictive
where the apparent inductances may vary due to the saturation
current control method [16].
and cross-saturation effect. Temperature rise and load
condition may lead to variation of winding resistance and
II. PROBLEM FORMULATION
permanent magnet flux. Under the parameters variation, the
A. Conventional Model Predictive Current Control slopes of current components are expressed as:
Time-derivatives (slopes) of stator current vector
components of an interior permanent magnet synchronous
=
(
did v d − ( Rs 0 + Rs ) id + r Lq 0 + Lq ( id , iq ) iq
,
) (7)
ld 0 + ld ( id , iq )
motor in the rotor flux reference frame are expressed as [22]:
dt
did
=
1
(v d + r Lq iq − Rs id ) , (1) diq
=
(
v q − ( Rs 0 + Rs ) iq − r Ld 0 + Lq ( id , iq ) iq )
lq 0 + lq ( id , iq )
dt Ld dt (8)
diq
= (v q − r m − r Ld id − Rs iq ) , ( )
1
(2) − r ( m 0 + m ) / lq 0 + lq ( id , iq ) ,
dt Lq

where, the conventional notations are used. In a conventional where “  ” subscribe denotes variation of a parameter from its
FCS-PCC, the linear model of the motor is used, and the model value designated by subscribe “0”. Besides,
effects of iron saturation or parameter variations are not ld = ld 0 + ld and lq = lq 0 + lq are differential inductances
included in the control [23]. The current components of which are derived from differentiating (5) and (6) as:
[k+1]th sampling instance should be predicted using the
measured currents of [k]th sampling instance and their lx = x ( id , iq ) i x , x  d , q  . (9)
predicted variations provided by the applied inverter voltage
vector in the [k]th control period as: The cross-differential inductances are ignored due to their
negligible effect on the performance of FCS-MPC. The
iˆx  k + 1 = i x  k  + iˆx  k  , x  d , q  , (3) calculated current slopes by considering the variations of the
parameters, as in (7) and (8) are denoted by Sd and S q
where iˆx  k  is the predicted variation of a current respectively, while the slopes without considering variations
component. It is calculated using the discrete forms of (1) and of parameters are denoted by Sd 0 and Sq 0 . The error
(2) over one sampling period. This prediction is necessary for introduced in the slope of a current vector component due to
compensating the one-step delay of the digital processor. The the parametric uncertainties could be defined as:
current components should also be predicted for [k+2]th
sampling instance for all switching states using one-step ahead
time-shifted form of (3). These predicted values are evaluated
Sx = Sx − S x 0 = k1x v x + k 2x , x d , q  , (10)
in a cost function which is commonly defined as:
where S x is the error of the slope of a current component.
The terms k1x and k 2x are the voltage-independent terms of
CFn = id* − iˆd  k + 2n + iq* − iˆq k + 2n , n  V 0 ~ V6  , (4)
S x . In FCS-MPC only one of the basic voltage vectors of
where the command values are expressed using a star the inverter with a magnitude of V m = 2V dc / 3 is applied to the
superscript. The voltage vector that minimizes the cost motor over one sampling period (except for the null voltage
function is applied to the motor during the [k+1]th sampling

2168-6777 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2019.2949222, IEEE Journal
of Emerging and Selected Topics in Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3

vector). Therefore, (7), (8), and (10) become sinusoidal 6

Current slope (kA/s)


functions of the voltage phase angle in the rotor reference 4 Without mismatch (Sd0)
2 S n 50% error in inductances (Sd)
frame at any instance. The variation of a current vector d Sdn + 5
0 Sdn +1
component in one sampling interval could be calculated as: -2 Sdn + 4
-4 Sdn + 2 Sdn + 3 n:Voltage vector number
i x = Ts S x , x d , q  , (11) (a)

15% error in inductances

prediction error (%)


Normalized current
60
where T s is the sampling period. As mentioned before, the 45
30% error in inductances
50% error in inductances
variations of the current components are used in the 30 65% error in inductances
15
predictions by (3) and (4). If Sd 0 and Sq 0 , instead of Sd and
0
S q , are used for the predictions through (11), the current -15
-30
(b)
predictions come with errors due to the error defined in (10).
150

prediction error (%)


Normalized current
Since the errors are different for any switching state, adding 100
the errors related to the voltage vector of the previous 50
sampling interval to the predictions of the existing sampling 0
interval in a feed-forward compensation form, as proposed in -50
-100
[25], may not be a desirable solution.
-150
The slope curves of the d-axis current with and without 0
1
90 180 270 360
inductance mismatch versus phase angle of a voltage vector ( Voltage vector phase angle in rotor frame (Deg.)
(c)
 ) with a fixed-magnitude of 2V dc / 3 are depicted in Fig. Fig. 1. Investigation of inductance mismatch effects. (a) The d-axis
1(a). The simulated motor parameters are given in Table I. current slope. Normalized errors of current components variations over
a sampling period for: (b) q-axis, (c) d-axis.
The motor operating point is 500 rpm under the nominal load.
It is seen that there are differences between Sd and Sd 0 A. A Parameter-free Prediction Method
curves as expressed in (10). These differences, which are As a first step to propose a method for overcoming the
dependent on the  are shown in the figure for inverter active effect of parameter uncertainties on the FCS-PCC method,
voltage vectors. They introduce errors into the predictions of another form of the cost function is derived for directly using
the current components in the PCC method through (11). The the predicted current slopes. From (4) one can obtain:
errors of the d- and the q-axis normalized current variation 1 * ˆ
CFn = id − id  k + 1 + iˆd  k + 1 − iˆd  k + 2n
over one sampling period under 15%, 30%, 50%, and 65% Ts
inductance mismatch are shown in Figs. 1(b) and 1(c). Large (12)
1
mismatch values happen in the case of a plug-and-play motor + iq* − iˆq  k + 1 + iˆq  k + 1 − iˆq  k + 2n .
Ts
drive with unknown parameters. The errors are normalized to
the maximum possible current variation over a sampling A command of current component slope ( S x* ) can also be
period. It is seen that there are considerable errors in the defined as:
predictions when the inductance mismatch becomes more than
30%. The errors lead to inaccurate current predictions through
(3). These wrong predicted values are used for the voltage
( )
S x* = (1/ Ts ) i x* − iˆx  k + 1 x  d , q . (13)
selection in (4), resulting in inappropriate decisions and
Substituting (11) and (13) into (12), one can derive a new
deterioration of current THD and torque ripples.
form of cost function as:
III. PROPOSED CONTROL METHOD
A control method is proposed in this section to overcome CFn = Sd* − Sˆd  k + 1n + Sq* − Sˆq  k + 1n , n  V 0 ~ V6  (14)
the problem of the previous section without using look-up
tables. In addition, by using a reference voltage calculated in Although the proposed cost function directly uses the slopes of
every sampling period fully independent of the motor current components, it is still parameter-dependent. Thus, a
parameters, the cost function evaluations are omitted. method is proposed to predict the slopes without using motor
parameters. The slopes of current components in (7) and (8)
TABLE I are rearranged in compact forms as:
PMS MOTOR SPECIFICATIONS
Parameters Values
Number of pole pairs 2 Sd = 1 cos  + 2 , (15)
Permanent magnet flux 570 mWb
S q = 3 sin  + 4 , (16)
Stator resistance 18.4 Ω
d-axis inductance 140 mH
q-axis inductance 310 mH where 1 ~  4 are defined in Appendix. For a null voltage
Rated power 330 W
vector where v m = 0 , S d =  2 and S q =  4 . The coefficients

2168-6777 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2019.2949222, IEEE Journal
of Emerging and Selected Topics in Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4
4
Sd (is = inom ) Sd ( No-load )
1 and  3 are dependent on the motor parameters, and the DC

Current slope
Sq (is = inom ) Sq ( No-load )
2

(kA/s)
link voltage. Therefore, their variations are slow and
negligible during a couple of short sampling periods (e.g. 0

T s = 75s ). However,  2 and  4 , not only depend on the -2


motor parameters, but also on its operating point. Due to the 0 90 180 270 360
rotor inertia, the variation of the motor speed is not Voltage vector phase angle in rotor frame (Deg.)
considerable during a few sampling intervals. Unlike that, the
current varies faster during transients. Fig. 2 shows the slopes Fig. 2. The effect of current variation on the current components
slopes.
of current components for the simulated motor at 1000 rpm
under no-load and nominal load. The transient between these
Δix[k-2] Δix[k-1] Sx[k] Sx[k+1]
two conditions takes about 2 ms, which is about 26 sampling Sx[k-2]=
Ts
Sx[k-1]=
Ts i x*
cycles for T s = 75s . As it is seen from the figure, current Δix[k-2]
Δix[k-1]

variation during a feasible transient does not affect ix


Ts S x*
considerably its components slopes through  2 and  4 .

ix[k+1]
V  k − 2 V  k − 1 V k  V  k + 1

ix[k+2]
ix[k-2]

ix[k-1]

ix[k]
Therefore, it is reasonable to assume that the curves are almost   k − 2   k − 1  k    k + 1
constant during 2~3 sampling intervals. The data of two points
from each curve is enough for estimation of the curve for a Fig. 3. A time sequence diagram of the proposed FCS-PCC method.
whole range of voltage phase angle ( 0    360 ). In other
words, the coefficients 1 ~  4 can be estimated by using the reference slopes using (3) and (13). The reference slopes along
stored values of the current slopes of [k-1]th and [k-2]th with Sˆx  k + 1 are used in the cost function calculations of
sampling intervals and their corresponding voltage vectors as: (14).
B. Voltage Selection
Sd  k − 1 − Sd  k − 2
ˆ1 = , (17) In order to eliminate the need for evaluation of a cost
cos   k − 1 − cos   k − 2
function, S x* can be useful for generating a reference voltage
vector ( v * = v *  * ) for the [k+1]th sampling interval. If
ˆ2 = Sd  k − 1 − ˆ1 cos   k − 1 , (18) this reference voltage could be applied at the beginning of the
sampling interval, the current would reach its command at the
end of the interval. To calculate the magnitude and the phase
Sq  k − 1 − Sq  k − 2 angle of the reference voltage vector, one can use (15) and
ˆ3 = , (19)
sin   k − 1 − sin   k − 2 (16) as:

S d* =V pu* ˆ1 cos  * + ˆ2 , (22)


ˆ4 = Sq  k − 1 − ˆ3 sin   k − 1 . (20)
S q* =V pu* ˆ3 sin  * + ˆ4 , (23)
In (17)-(20), the slopes are calculated by using the two
consecutive current samples and (11). It can be seen that none
of the motor parameters are used in the estimation. The where V pu* = v * /V m is the normalized magnitude and * is
estimated coefficients are then used in (15) and (16) to predict the phase angle of the reference voltage vector in the rotor flux
the current slopes, (i.e., Sˆx  k  and Sˆx  k + 1 ) for the frame. The phase angle is calculated using (22) and (23) as:
corresponding voltage vectors. A time sequence diagram is
depicted in Fig. 3 for more clarification. The variations of a
current component in four consecutive sampling intervals are
 ˆ1 S q* − ˆ4
 = tan 
* −1
( )  . (24)
shown in the figure. The dashed lines show the predicted
 ˆ S * − ˆ
 3 d 2 ( ) 
variations. The predicted current slope ( Sˆx  k  ) is used for the
prediction of current components variations of [k]th sampling After the calculation of * , the normalized magnitude could
interval provided by the inverter voltage vector that is being also be calculated using either (22) or (23). Fig. 4 represents
applied in the same interval using the following equation: this reference voltage together with the inverter basic voltage
vectors. The dashed lines divide the area of the hexagon into
iˆx ( n ) = Ts Sˆx ( n ) , x  d , q  & n  V 0 ,..., V6 . (21) six well-known sectors. Since applying the calculated
reference voltage is not possible by using only one switching
This prediction is necessary for compensating the inherent configuration over a sampling interval, the nearest basic
delay of the digital microprocessor and calculation of the voltage vector to the calculated reference voltage should be

2168-6777 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2019.2949222, IEEE Journal
of Emerging and Selected Topics in Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5

fig
iq* Sq*
r* +_ PI +
_ 1/Ts Voltage
Switching
commands
v2 Sd* vector
v3 v* id* +
_
1/Ts
selection

ˆ1 ~ ˆ4
v0 v1
v4  *
iˆq [k + 1] Estimations
+ _ ia
Z -1
id dq
and 1/Ts
iˆd [k + 1] S _ ib
predictions d 1/Ts Z -1 abc
r Sq + ic

v5 v6 r iq 
r
d/dt

r = 0.5Vm =Vdc 3 Fig. 5. A block diagram of the proposed PCC method.

Fig. 4. The hexagon plane of the inverter voltage vectors.

selected. This results in a minimum current error at the end of


the sampling interval. The dashed lines, as the borders of the
mentioned sectors, are the bisectors of angles between every
two adjacent basic active voltage vectors. Therefore, when the
reference voltage is located in one of the sectors, the
corresponding inverter active voltage vector of the sector
would be the nearest active voltage vector to reference
voltage. A dashed-dotted circle with a radius of 0.5V m is also
illustrated in the figure. When the normalized magnitude of
the reference voltage vector becomes less than 0.5, it is
located inside the circle. In this situation, a null voltage vector
would be selected instead of an active one as the nearest
inverter voltage vector to the reference voltage.
C. Control System
Fig. 6. A picture of the experimental IPM motor drive setup.
A block diagram of the proposed PCC method is illustrated
in Fig. 5. It consists of a PI speed control loop that generates
the q-axis current component command. Whereas, the d-axis obtaining the rotor position. A TMS320F28377D digital signal
command could be generated from a maximum torque per processor (DSP) executes the control code. A sensor board
ampere (MTPA) block. “Estimations and Predictions” block, consisting of current and voltage sensors together with delta-
sigma modulator ICs is used which is immune to noises.
predicts iˆd  k + 1 and iˆq  k + 1 using the estimated slopes of Motor phase currents are oversampled using the sensors with a
the current components, i.e., substituting (21) into (3). The 20 MHz clock frequency. The sampling period in both of the
“Voltage vector selection” block, selects the optimum simulations and DSP programming is 75μs. A three-phase
switching state for the inverter using the estimated current two-level inverter based on an intelligent power module is also
slopes and the calculated slope commands. The selection is used. Fig. 6 shows the experimental setup.
based on the method which was discussed in subsection III.B. The performance of the drive under the proposed PCC is
evaluated and compared with that under two other control
IV. PERFORMANCE EVALUATION methods including a conventional PCC (model-based) and an
The proposed method is evaluated using computer existing model-free PCC [16].
simulations and experimental tests. Simulations are carried out A. Simulation Results
using Matlab/Simulink. The motor parameters are given in
Steady-state conditions of the motor drive at 750 rpm with 2
Table I. This motor is coupled with a synchronous machine
Nm load is studied under the three control methods. The motor
acting as a load. A 2500-pulse incremental encoder is used for
parameters in the controller code are the same as their actual

1.5
1
0.5
ia (A)

0
-0.5
-1
-1.5
0.2 0.3 0.4 0.2 0.3 0.4 0.2 0.3 0.4
Time (s) Time (s) Time (s)
(a) (b) (c)
Fig. 7. Simulation results of steady-state phase currents. (a) Conventional PCC. (b) Existing model-free PCC. (c) Proposed PCC.

2168-6777 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2019.2949222, IEEE Journal
of Emerging and Selected Topics in Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6

1.5
Current (A) 1
iq iq iq
0.5
0 id id id
-0.5

0.2 0.35 0.5 0.2 0.35 0.5 0.2 0.35 0.5


Time (s) Time (s) Time (s)
(a) (b) (c)
Fig. 8. Simulation results of dq components of the current at steady-state. (a) Conventional PCC. (b) Existing model-free PCC. (c) Proposed PCC.

values. The simulation results of the phase current are


illustrated in Fig. 7. It is seen that the phase current under the 2 Conventional method Proposed method
conventional and the proposed PCC are the same. However,

ia (A)
1
applying non-optimum voltage vectors for preventing the 0
voltage stagnation problem results in current waveform -1
degradation under the existing model-free method. The same
0.15 0.2 0.25 0.3 0.35 0.4 0.45
conclusion can be obtained from the dq components of current Time (s)
vector waveforms as shown in Fig. 8. Fig. 9. The phase current of the motor under 50% varied values of
The drive performance under the parameter mismatch is motor inductances (conventional vs. proposed PCC methods).
also investigated. The motor operates at 500 rpm with a 1 Nm
load. The phase current waveform under a 50% variation of
1 Conventional PCC Proposed PCC
the motor inductances is given in Fig. 9. Degradation of the

Current (A)
0.5
performance under the conventional PCC can be seen.
iq (A)
However, it is improved when the proposed method takes over 0 id (A)
the control at t=0.3 s. In addition, Fig. 10 shows the dq- -0.5
components of the current. It is seen that the current 0.15 0.2 0.25 0.3 0.35 0.4 0.45
components enjoy fewer ripples under the proposed PCC Time (s)

compared to those under the conventional PCC. The better


Fig. 10. The d- and q-axis components of the phase current under 50%
performance is due to the lower prediction errors in the
varied values of motor inductances (conventional vs. proposed PCC
proposed method, which results in a better voltage vector methods).
selection. Fig. 11 shows the prediction errors of [k+1]th
instance for the d- and q-axis stator current components under through the dq-components of the phase current as shown in
different inductance mismatch values. It is worth noting that Fig. 14. It is seen that when the proposed PCC is enabled at
not only there are errors in the predictions of the [k+1]th t=0.15 s, the ripples of i d and i q decrease considerably.
instance, but also those related to the [k+2]th instance. The As mentioned before, the mismatch causes prediction errors
latter is used in the cost function calculations under the of the current components as presented in Fig. 15. The errors
conventional PCC. The errors result in a non-optimum voltage are calculated by subtracting the measured current at [k]th
vector selection and degradation of the control performance. It sampling instance from the predicted current for this instance
is seen in Fig. 11 that, under the proposed PCC, the prediction at [k-1]th sampling interval. It can be seen that after enabling
errors are nearly zero, even in the parameter mismatch the proposed method, the errors vanish.
conditions. Finally, the execution time of the proposed method is
compared with the one for the conventional PCC method. In
B. Experimental Verification the latter method, the cost function is executed seven times in
Firstly, the dynamic performance of the motor drive under each sampling interval by the DSP, which takes about 16 μs.
the proposed method is investigated. While the motor is Nevertheless, there is no need for any cost function evaluation
running at 300 rpm, a speed step command is applied to the in the proposed method since the optimum voltage vector is
motor to increase the speed to 1000 rpm. After a while a selected directly by some simple calculations and
speed-reversal command is applied to the motor drive to comparisons. The execution latency for the conventional and
reduce the speed to -1000 rpm. The waveforms of the motor the proposed PCC methods are 22 μs and 11.5 μs respectively.
speed, torque, and the phase current are illustrated in Fig. 12. It is thus concluded that the proposed method can be executed
The figures confirm desirable motor dynamic performance. 47.8% faster than the conventional one.
The steady-state waveforms of the motor phase current under
the conventional and the proposed PCC methods are then V. CONCLUSION
compared under the parameter mismatch. Fig. 13 shows the Model predictive current control of motor drives suffers
results at 500 rpm, 2 Nm load with a 100% inductance from prediction errors caused by variations of motor
mismatch in the controller code. It is seen that the mismatch
parameters during the operation or unavailability of the exact
affects the current waveform under the conventional PCC,
values of those parameters. The errors lead to inappropriate
while the proposed method is robust against the parameter
selection of inverter voltage vectors and thus the degradation
mismatch side effects. These effects are further investigated

2168-6777 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2019.2949222, IEEE Journal
of Emerging and Selected Topics in Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7

0.2 Conventional PCC


15% error in inductances
0.15
30% error in inductances
iq prediction error (A)

0.1
50% error in inductances
0.05 65% error in inductances
0
Proposed PCC 2 (A/Div.)
-0.05

-0.1

-0.15

-0.2
-0.25
(a)
0.4

0.3 Fig. 13. Steady-state waveforms of the phase current under the
id prediction error (A)

0.2 conventional and the proposed PCC methods with 100% error in the
0.1
motor inductances values.
0

-0.1
0.5 Conventional PCC Proposed PCC 1.4
-0.2 0.3 1.2

iq (A)
id (A)
0.1 1
-0.3
-0.1 id iq 0.8
-0.4 -0.3 0.6
Conventional PCC Proposed PCC -0.5 0.4
(b) -0.7 0.2
Fig. 11. The prediction errors of [k+1]th sampling instance for the 0.15 0.3 0.45
current components; (a) d-axis current, (b) q-axis current. Time (s)

Speed (2000 rpm/Div.) Speed reference Fig. 14. Steady-state waveforms of the d- and q-axis components of the
motor current under 100% error in the motor inductances values.

Torque (5 Nm/Div.)
Conventional PCC Proposed PCC

ia (2 A/Div.) id prediction error


0.4 (A/Div.)
iq prediction error

Fig. 15. Prediction errors of the d- and q-axis components of the motor
Fig. 12. Speed reversal test of the motor drive.
current under 100% error in the motor inductances values.

of the drive performance. The proposed predictive current stagnation of voltage vectors is resolved without degrading the
control method is totally robust against variations of the drive performance.
parameters. Unlike existing model-free methods, the proposed
method does not need look-up tables for storing the current ACKNOWLEDGMENT
variations provided by all inverter voltage vectors. Hence, it The effective cooperation of Dr. M. R. Nikzad in
does not face the stagnation of data-updating, which is a developing the experimental setup is appreciated.
drawback of the existing model-free control methods. In the
proposed method, the time-derivatives (slopes) of current APPENDIX
vector components are predicted using the data of the current
variations of two most recent sampling intervals and their
corresponding voltage vectors. Moreover, using the predicted 1 =V m / ( l d 0 + l d ) , (A1)
current slopes, a direct optimum voltage vector selection
method is used in order to eliminate the need for cost function
evaluations. This leads to a reduction of the computational
2 =
( )
r iq L q 0 + L q ( id , iq ) − ( Rs 0 + Rs ) id
, (A2)
effort. The effectiveness of the proposed method is evaluated l d 0 + l d
by computer simulation and experimental tests. The results
show an improved performance under the parameter mismatch
 3 =V m / ( l q 0 + l q ) ,
between the motor and the controller compared to the (A3)
conventional predictive current control method. In addition,
compared to the existing model-free methods, the problem of

2168-6777 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2019.2949222, IEEE Journal
of Emerging and Selected Topics in Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8

Electronics (PRECEDE), 2017 IEEE International Symposium on,

(
r ( m 0 + m ) + r id ld 0 + ld ( id , iq ) )
2017, pp. 119-124: IEEE.
[18] M. Siami, D. A. Khaburi, A. Abbaszadeh, and J. Rodríguez,
4 = − "Robustness improvement of predictive current control using prediction
l q 0 + l q (A4) error correction for permanent-magnet synchronous machines," IEEE

( ) (
− iq R s 0 + Rs / l q 0 + l q . ) [19]
Trans. Ind. Electron., vol. 63, no. 6, pp. 3458-3466, 2016.
M. Siami, D. A. Khaburi, and J. Rodríguez, "Torque ripple reduction of
predictive torque control for PMSM drives with parameter mismatch,"
IEEE Trans. Power Electron., vol. 32, no. 9, pp. 7160-7168, 2017.
[20] S. Vaez-Zadeh, "Analysis of a DTC with back EMF oriented voltage
REFERENCES for PMS motor drives," IEEE Trans. Energy Convers., vol. 33, no. 3,
[1] P. Cortés, M. P. Kazmierkowski, R. M. Kennel, D. E. Quevedo, and J. pp. 1594-1596, Sept. 2018.
Rodríguez, "Predictive control in power electronics and drives," IEEE [21] M. Khalilzadeh and S. Vaez-Zadeh, "Computation efficiency and
Trans. Ind. Electron., vol. 55, no. 12, pp. 4312-4324, 2008. robustness improvement of predictive control for PMS motors," IEEE
[2] Z. Ma, S. Saeidi, and R. Kennel, "FPGA Implementation of model J. Emerg. Sel. Topics Power Electron., 2019.
predictive control with constant switching frequency for PMSM [22] S. Vaez-Zadeh, Control of permanent magnet synchronous motors.
drives," IEEE Trans. Ind. Informat., vol. 10, no. 4, pp. 2055-2063, London, U.K.: Oxford Univ. Press, 2018.
2014. [23] J. Rodriguez and P. Cortes, Predictive control of power converters and
[3] F. Morel, X. Lin-Shi, J. M. Retif, B. Allard, and C. Buttay, "A electrical drives. John Wiley & Sons, 2012.
comparative study of predictive current control schemes for a [24] N. Bianchi and S. Bolognani, "Magnetic models of saturated interior
permanent-magnet synchronous machine drive," IEEE Trans. Ind. permanent magnet motors based on finite element analysis," in
Electron., vol. 56, no. 7, pp. 2715-2728, 2009. Conference Record of 1998 IEEE Industry Applications Conference.
[4] A. D. Alexandrou, N. K. Adamopoulos, and A. G. Kladas, Thirty-Third IAS Annual Meeting (Cat. No. 98CH36242), 1998, vol. 1,
"Development of a constant switching frequency deadbeat predictive pp. 27-34: IEEE.
control technique for field-oriented synchronous permanent-magnet [25] J. Stumper, S. Kuehl, and R. Kennel, "Predictive torque control for AC
motor drive," IEEE Trans. Ind. Electron., vol. 63, no. 8, pp. 5167-5175, drives: Improvement of parametric robustness using two-degree-of-
2016. freedom control," 2013 IEEE Energy Conversion Congress and
[5] X. Zhang, B. Hou, and Y. Mei, "Deadbeat predictive current control of Exposition, 2013, pp. 1170-1175.
permanent-magnet synchronous motors with stator current and
disturbance observer," IEEE Trans. Power Electron., vol. 32, no. 5, pp.
3818-3834, 2017.
[6] B. H. Kenny and R. D. Lorenz, "Stator- and rotor-flux-based deadbeat Mohammad Khalilzadeh was born in
direct torque control of induction machines," IEEE Transactions on
Ind. Appl., vol. 39, no. 4, pp. 1093-1101, 2003. Urmia, Iran in 1990. He received his M.Sc.
[7] M. Khalilzadeh and S. Vaez-Zadeh, "Deadbeat current control of and B.Sc. in Electrical Engineering from K.
permanent magnet synchronous motors using a simplified discrete N. Toosi University of Technology and the
space vector modulation," Power Electronics, Drives Systems and University of Tehran, in 2012 and 2014
Technologies Conference (PEDSTC), 2018 9th Annual, 2018, pp. 456-
461: IEEE. respectively. He is currently working toward
[8] J. Rodriguez et al., "Predictive current control of a voltage source the Ph.D. degree at the University of Tehran. His research
inverter," IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 495-503, 2007. interests include power electronics and motor drives.
[9] M. Preindl and S. Bolognani, "Model predictive direct torque control
with finite control set for PMSM drive systems, Part 1: maximum
torque per ampere operation," IEEE Trans. Ind. Informat., vol. 9, no. 4,
pp. 1912-1921, 2013.
[10] M. Navardi, J. Milimonfared, and H. Talebi, "Torque and flux ripples Sadegh Vaez-Zadeh (S’95‒A’03‒SM’05)
minimization of permanent magnet synchronous motor by a predictive- has been a Full Professor at the University
based hybrid direct torque control," IEEE J. Emerg. Sel. Topics Power of Tehran, since 2005. He has co-authored
Electron., vol. 6, no. 4, pp. 1662-1670, Dec. 2018. more than 200 technical papers and
[11] P. G. Carlet, F. Tinazzi, S. Bolognani, and M. Zigliotto, "An effective
model-free predictive current control for synchronous reluctance motor authored Control of Permanent Magnet
drives," IEEE Trans. Ind. Appl., pp. 1-1, 2019. Synchronous Motors, Oxford University
[12] M. Abdelrahem, C. Hackl, Z. Zhang, and R. Kennel, "Robust predictive Press, 2018. He is an Editor of the IEEE
control for direct-driven surface-mounted permanent-magnet Transactions on Energy Conversion and a Subject Editor of
synchronous generators without mechanical sensors," IEEE Trans.
Energy Convers., vol. 33, no. 1, pp. 179-189, March 2018. the IET Renewable Power Generation.
[13] B. Wang, X. Chen, Y. Yu, G. Wang, and D. Xu, "Robust predictive
current control with online disturbance estimation for induction
machine drives," IEEE Trans. Power Electron., vol. 32, no. 6, pp. Mohammadsadegh Eslahi received the
4663-4674, 2017.
[14] H. A. Young, M. A. Perez, and J. Rodriguez, "Analysis of finite- B.Sc. degree in electrical engineering from
control-set model predictive current control with model parameter the University of Shiraz, Shiraz, Iran, 2015,
mismatch in a three-phase inverter," IEEE Trans. Ind. Electron., vol. and the M.Sc. degree in electrical
63, no. 5, pp. 3100-3107, 2016. engineering from the K. N. Toosi
[15] C.-K. Lin, T.-H. Liu, L.-C. Fu, and C.-F. Hsiao, "Model-free predictive
current control for interior permanent-magnet synchronous motor University of Technology, Tehran, Iran,
drives based on current difference detection technique," IEEE trans. 2017. He is currently with Advanced
Ind. Electron., vol. 61, no. 2, pp. 667-681, 2014. Motion Systems Research Laboratory of University of Tehran,
[16] C.-K. Lin, Y.-S. Lai, and H.-C. Yu, "Improved model-free predictive Tehran, Iran.
current control for synchronous reluctance motor drives," IEEE Trans.
Ind. Electron., vol. 63, no. 6, pp. 3942-3953, 2016.
[17] D. Da Rù, M. Polato, and S. Bolognani, "Model-free predictive current
control for a SynRM drive based on an effective update of measured
current responses," Predictive Control of Electrical Drives and Power

2168-6777 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Anda mungkin juga menyukai