You are on page 1of 6

Voltage Controller for Flux Weakening Operation

of Interior Permanent Magnet Synchronous Motor


in Automotive Traction Applications
Tobias Huber, Wilhelm Peters, Member, IEEE, Joachim Böcker, Senior Member, IEEE

Abstract—In electric automotive traction drives an optimal has to be maintained. This can either be done in an open-
utilization of the DC-bus voltage in the wide flux weakening loop manner, leading to a suboptimal DC-bus utilization for
range is crucial. At the same time a voltage margin is required most operating points. Alternatively, a voltage controller can
to ensure the stability of the inner current control loop. An
adequate trade-off between these conflicting objectives is ob- be employed. In this case, a much better trade-off between the
tained by employing a superimposed voltage controller that is conflicting objectives voltage margin and DC-bus utilization
activated during flux weakening operation. In this paper, the can be achieved.
design of such a voltage controller is presented. A simplified In several publications, the necessity to control the voltage
voltage controller plant model is identified from step response demand of different motors in flux weakening operation is
measurements. Due to variations of the plant parameters, the
voltage controller is designed as a gain-scheduling controller addressed and suitable control schemes are proposed. In [1]
with sufficient robustness towards plant model inaccuracies. and in [2], additional voltage controllers are employed to
The performance of the voltage controller is demonstrated by adjust the voltage demand of an induction motor. In [1],
test-bench measurements based on an electric traction motor the voltage controller represents the outer control loop of a
typically employed in sub-compact electric vehicles. cascaded structure with flux and current controllers serving as
Index Terms—Interior permanent magnet synchronous motors inner control loops. In [2], the voltage controller is arranged
(IPMSM), traction motors, electric vehicles, voltage control, flux in parallel to the flux controller and the switching between
weakening operation constant flux and constant voltage operation is carried out
implicitly by minimum selections. In [3], the principle of
I. I NTRODUCTION lookup-table-based voltage control is presented, which is also
utilized in [4], [5] as well as in this contribution. However,
In today’s electric and hybrid electric vehicles interior per- in [4], [5], the design of the voltage controller based on a
manent magnet synchronous motors (IPMSM) are commonly thorough plant analysis is not described. Moreover, the critical
employed as traction drives. For that specific application, transition from constant torque to flux weakening range is
the motors are designed to obtain high power and torque completely neglected. Both aspects are discussed in this paper.
densities causing a considerably different behavior compared
to industrial drives. The main characteristics are a wide
II. C ONTROL S CHEME OVERVIEW AND VOLTAGE
speed range, a wide constant-power operation range and high
C ONTROLLER S PECIFICATION
magnetic saturation. Since the available energy is limited due
to the battery capacity, the efficiency and the weight of the uDC
drive train components are crucial. Thus, a suitable control Control Scheme
uDC
scheme is required to ensure a high efficiency as well as a ua sa
i∗d Field-Oriented
high utilization of the installed components. T∗ Operating Point ub sb
i∗q Current PWM
Typically, a field-oriented control is utilized, which is Selection uc sc
Controller
composed of an inner current control loop and a superimposed
open-loop torque control. In the flux weakening range, a ψmax,Δ ia
ib
suitable voltage margin is required to ensure the stability of ic
the current control loop on the one hand and to provide a good a
Voltage
utilization of the DC-bus voltage on the other hand. Increasing Controller uDC
IPMSM
the DC-bus voltage utilization allows torque demands to be
 Position Signal
generated at higher voltage and lower current levels. That ωel Processing
way, higher maximum torque outputs can be achieved at given
speeds without violating the current limit. At the same time, T ,n
the drive efficiency can be increased, since for most motors the Fig. 1. Control scheme overview
current magnitude has a higher impact on the total losses than
the voltage magnitude. Due to modelling inaccuracies in the In Fig. 1, a suitable control scheme is presented [6]. The
open-loop torque control, a sufficient voltage safety margin field-oriented current control consists of a decoupling measure
and two independent current controllers for the current com-
T. Huber, W. Peters and J. Böcker are with the Department of Power
Electronics and Electrical Drives, Paderborn University, 33098 Paderborn, ponents id and iq [6], [7]. To fully utilize the DC-bus voltage
Germany, e-mail: {huber, peters, boecker}@lea.uni-paderborn.de for transient operation point changes, the current controller

l-))) 
output voltages are limited to the hexagon in Fig. 2 resulting and the voltage limit in the current control loop. Since the
from the possible switching states of a B6 bridge inverter voltage control loop dynamics can be much slower than the
[6]. In an operation point selection scheme, suitable current current controller dynamics to correct the motor’s voltage
reference values i∗d and i∗q are chosen to set the requested demand during steady state operation, a simplified plant
torque T ∗ with high efficiency considering the maximal model is acceptable. Thus, it is convenient to approximate
flux magnitude ψmax,Δ . The reference values are chosen the dominant plant dynamics by a first-order delay transfer
according to the MTPC (Maximum Torque per Current) or function with variable parameters according to (2).
the ME (Maximum Efficiency) operating strategy depending
Kp (θ1 )
on the motors loss characteristics [8], [9]. The maximum flux Gp (s) = (2)
ψmax,Δ results from the actual speed ωel and the DC-bus 1 + sTp (θ2 )
voltage uDC and is corrected by the voltage controller. That Before the identification of the plant parameters Kp and Tp is
way, a high utilization of the DC-bus voltage can be realized performed, their dominant dependencies of the parameter sets
while at the same time maintaining a sufficient voltage margin θ1 und θ2 have to be considered. Afterwards, the parameter
to ensure the proper operation of the current controllers. The identification procedure can be set up.
voltage controller is activated once the motor enters the flux The proportional gain of a first-order delay corresponds to
weakening range. Since the voltage controller is realized in a the output-input-ratio in steady state condition. This ratio can
cascaded control structure with the current controller as inner directly be calculated from the voltage equation of an IPMSM
control loop, its dynamics should be slower than the dynamics in steady state operation. Due to the low resistive stator
of the current controller [10]. That way, the voltage demand of winding values of the investigated motor, ohmic voltage drops
the motor is limited to the inner circle of the hexagon in Fig. 2 can be neglected in this context. Thus,
 the motor’s voltage
during steady state operation. During transient operation, on demand |u| and the flux linkage ψ  interrelate according to
the other hand, the current controllers can utilize the entire (3) with the output of the voltage controller in flux weakening
hexagon inFig. 2 (including edges) without being constrained operation as can be seen in Fig. 3.
by the voltage controller.  
ψ  = |u| = ψmax,Δ = ψmax + ψΔ (3)
uq uβ
√1 uDC = umax
|ωel |
3
Solving (3) for ψΔ (4) results.
ud
u |u| |u| − umax Δu
ψΔ = − ψmax = = (4)
2
|ωel | |ωel | |ωel |
 3 uDC
|u|
uα With the definition of the modulation rate a = umax the
modulation rate variation Δa with respect to a variation
of the flux correction value Δψ can be calculated. Thus,
the plant gain Kp depends on the exogenous parameters
a= |u|
=1 θ1 = [ωel , uDC ]T and can be calculated according to (5).
umax

Δa |ωel | |ωel | 3
Fig. 2. Utilizable Voltage Kp = lim = = (5)
t→inf ψΔ umax uDC
Due to nonlinearities and modelling inaccuracies of the lookup
III. P LANT PARAMETER I DENTIFICATION tables in the operation point selection the plant gain Kp can
For the voltage controller design, a plant model is required. deviate from the analytically calculated value, thus (5) will be
The voltage control loop scheme is shown in Fig. 3. The validated by experiments within the identification of the plant
deviation of the actual modulation rate a from its reference parameters.
value a∗ is the input of the voltage controller. The actuating Since there are no delays in the operation point selection,
variable is the flux correction value ψΔ , which is added to the plant time constant directly results from the current
the feed-forward flux ψmax = uωmax el
with umax = u√DC 3
controller closed-loop dynamics. The current controller gains
determining the flux limit ψmax,Δ in the operation point are not adapted to the variation of the current dynamics caused
selection. Appropriate operation points are chosen according by magnetic saturation. In consequence, the dominant time
to the requested torque T ∗ and the flux ψmax,Δ and are constant of the current control loop Tp varies depending on
applied to the current controller. The voltage at the current the operation point (id , iq ). Additional nonlinearities are in-
controller output is used to calculate the modulation rate a, troduced by the actuating variable limitations in Fig. 3, which
which is the controlled variable. The resulting plant transfer have an impact on the dominant plant time constant as well.
function (1) In this context, especially the voltage limitation at the current
controller outputs has to be considered. The impact of that
a(s) Δa(s)
Gp (s) = = (1) limitation depends on the current controller dynamics and the
ψmax,Δ Δψmax (s) actual modulation rate a, since the available voltage margin
contains several nonlinearities due to the nonlinear torque for current controller operation varies with the modulation
equation, the nonlinear current dynamics caused by magnetic rate. Thus it is assumed, that the dominant plant time constant
saturation, several limitations in the operation point selection Tp depends on the parameter set θ2 = [a, id , iq ]T .


ψmax T∗ Operating i∗dq ∗
Current udq idq
a∗ Voltage ψΔ ψmax,Δ Point IPMSM
Controller
Controller Selection
a
abs umax = u√
DC
3

Fig. 3. Voltage Control Loop Scheme

The parameters of the transfer function (2) were identified 45


from open-loop input-output measurements of the operation
40
point selection and the current control loop. Therefore, the
flux correction value is varied stepwise and the step response 35

Vs
of the modulation rate Δa is measured at the test bench. The

1
30

Kp in
transfer function parameters are identified in different oper-
25
ation points and have only local validity in the environment
of the respective operation point. From the composition of 20

the parameter sets θ1 and θ2 , it is reasonable to understand 15


operation points as 5-tuples (ωel , uDC , a, id , iq ). To reduce the 10
resulting number of experiments the following assumptions 4 6 8 10 12
are made: The modulation rate reference is set to a∗ = 0.953, n in 1000/min
the influence of the DC-bus voltage on Kp is sufficiently Fig. 5. Fitted (red) and analytically calculated (green) plant gain Ki
described by (5) and does not need to be separately identified over speed. The blue points represent the aproximated gains from different
by experiments, a rough estimate of the plant time constant identification experiments.
and a controller that is robust to parameter variations is
sufficient to meet the moderate dynamics requirements of the
controller with small gain, then the integrator value was
voltage controller. Furthermore, the steady state current values
frozen and a negative flux correction value step was applied.
(id , iq ) are determined as a function of ωel , uDC , a and the
Because a variation of the flux correction value ψΔ causes a
torque reference T ∗ by the lookup tables in the operation point
variation of the currents id and iq and thus a variation of the
selection. Thus, it is sufficient to identify the plant parameters
dominant plant time constant Tp , different steps were applied
at different combinations of torque and speed values (T ∗ , ωel ).
to sufficiently estimate the range of Tp . Flux correction
value steps of ψΔ = −5 mVs and ψΔ = −10 mVs were
1 1 1 1
used. Based on these experiments, the plant parameters were
K̂p = 30.3 K̂p = 28.8
Vs Vs identified utilizing the prediction error method [11]. In Fig. 4,
0.8 T̂p = 1.3 ms 0.8 T̂p = 2.0 ms measured and approximated modulation rate step responses
a

are shown for a couple of exemplary operation points. In


0.6 0.6
Fig. 5, the analytically calculated plant gains according to
0.4 0.4 (5) are compared with the actual gains from the identification
−10 0 10 20 30 −10 0 10 20 30 experiments. Since (5) approximates the actual gains quite
t in s t in s accurately, it can directly be used in the controller design.
(a) 10 Nm at 8000 min−1 (b) 65 Nm at 8000 min−1 It can also be seen from the identification experiments that
1 1 1 1
the dominant plant time constant varies within a range of
K̂p = 43, 1 K̂p = 43, 2
Vs Vs
0.8 . . . 3 ms in the relevant operation range.
0.8 T̂p = 0, 8 ms 0.8 T̂p = 2, 7 ms
a

0.6 0.6 IV. VOLTAGE C ONTROLLER D ESIGN


0.4 0.4
−10 0 10 20 30 −10 0 10 20 30
t in s t in s ψmax,Δ ∗
ψlim
(c) 10 Nm at 12000 min−1 (d) 50 Nm at 12000 min−1 min i∗d
Tmax (ψ, ω)
Fig. 4. Measured (blue) and approximated (red) modulation rate step
f1 (ψ, T, ω)

responses for flux correction steps of ψΔ = −10 mV


ω ω

Based on these assumptions, the following experiments i∗q


were performed for different operation points in the flux weak- ∗
T∗ ψopt (T ∗ , ω) Tlim
ening range: The DC-bus voltage was set to a nominal value in
the center of the voltage variation interval. A modulation rate
of a = 0.953 was adjusted by an empirically designed integral Fig. 6. Operation Point Selection Scheme


An important aspect of the voltage controller design is the disturbance transfer function (7) results.
realization of the controller output limitation along with a 1
Gp (s) Kp K s
suitable anti-reset windup scheme. From the operation point Gz (s) = = (7)
selection scheme in Fig. 6, it can be seen, that an optimal flux 1 + Gi (s)Gp (s) 1 T
1 + K s + Kp s2
value ψopt is estimated according to the utilized operation The disturbance transfer function has a second-order band-
strategy and compared to the flux limit ψmax,Δ , which is pass filter characteristic with the generalized transfer function
manipulated by the voltage controller. Then, the minimum of (8) including the damping factor d and the characteristic
those flux values is used to select an appropriate operation angular frequency ω0 .
point. It becomes obvious, that the voltage controller is only
active in the flux weakening range, in the constant torque KBP ω2d0 s
GBP (s) = 2d 1 2
(8)
range the voltage control loop is not closed and the modulation 1+ ω0 s + ω02
s
rate a is permanently smaller than the reference value a∗ .
The controller output is limited through positive and negative By equating the coefficients in (7) and (8) the damping factor
flux correction value ψΔ +
and ψΔ−
. Those limits are chosen by and the characteristic angular frequency of the voltage control
means of a worst-case estimation of the expected modelling loop can be calculated according to (9).

inaccuracies in the operation point selection. The positive limit 1
d = √1
T
+ T0 = = Kp (9)
ψΔ ensures, that the drive is not operated in flux weakening 2 KTp ω0
mode at modulation rates a smaller than the reference value Since the voltage controller is employed in a cascaded

a∗ , the negative limit ψΔ ensures, that the voltage demand control structure, its dynamics should be slower than the
of the motor can be sufficiently reduced to maintain the inner current control dynamics to avoid potential oscillation
desired voltage margin during flux weakening operation. In excitations between both controllers. In addition, the voltage
the following, the controller design and the realization of the controller is supposed to adjust the motor’s voltage demand
anti-reset windup scheme will be de described. in steady state operation. However, after a transient operating
point variation the voltage controller has to prevent the current
ψe controller to be operated in saturation for too long. Since the

decoupling of the current components is inoperable during
a ψΔ a
Gi (s) Gp (s) current controller output limitation, the currents can deviate
from their reference values and even exceed the allowed
current limit of the motor. A good controller performance is
gained with aperiodic damping d ≤ 1. With the poles of the
Fig. 7. Simplified voltage control loop
disturbance transfer function (10)
1 1   
An integral controller is chosen as voltage controller mainly p1,2 = − =− d ± d2 − 1 (10)
for two reasons: First, as shown in Fig. 3, the modulation T1,2 T0
rate a is directly calculated from the actuating variable of the dominant time constant of the voltage control loop results
the current control loop u∗dq . A proportional controller gain to (11) considering (9).
leads to a noisy flux correction values ψΔ , while a smooth 1 2d
variation is obtained by employing an integral controller. Tv = = √ Tp (11)
min (|Re{p1,2 }|) d − d2 − 1
Second, the voltage controller should only adjust the motor’s  
voltage demand during steady state operation while allowing α

the current controller to utilize the entire voltage region during The proportionality factor α in (11) describes the relative
transients (hexagon in Fig. 2). Since the voltage controller dynamics between the inner current and outer voltage control
has to follow the modulation rate to a constant reference loop. From (11), the damping factor d can be expressed as a
value of a∗ = 0, 953, not the reference response but the function of the relative dynamics α according to (12).
disturbance rejection of the modulation controller is crucial. A α
simplified scheme of the lookup-table-based voltage control d= √ (12)
2 α+1
loop is shown in Fig. 7 whereas modelling inaccuracies in
the operation point selection are represented by an additive If suitable values for d and α are found, the controller gain
flux error ψe at the integral controller output. In contrast to K can be calculated from (13).
the plant time constant, the plant gain Kp (|ω| , uDC ) can be 1 α−1 1 1
K= = (13)
compensated in terms of maintaining constant voltage loop Tp α 2 Tp 4d2
dynamics by adjusting the integral controller gain according Since the dominant plant time constant varies for different
to the controller transfer function (6). operation points in the range 0.8 . . . 3 ms, the controller gain
K should be calculated based on the maximal occurring value.
ψΔ (s) Ki K That way, it is ensured, that the relative dynamics between
Gi (s) = = with Ki = (6)
Δa(s) s Kp (ωel , uDC ) the inner and outer control loop do not exceed a predefined
minimal value. Since the controller gain is calculated based on
While the gain Kp in (6) is scheduled according to (5), nominal values for the time constant and the dynamic range
the constant gain K has yet to be determined. With (6) the (e.g. α = 5) in (13), the actual value of alpha can be calculated


by rearranging (13), while assuming Tp to be the actual time
200
constant.
2
α=  (14)
1 − 1 − 4KTp 0

id , iq in A
In addition to the controller design, an anti-reset windup
(ARW) scheme and a limitation of the integral controller -200
action is required to avoid integrator windup when the motor
is operated in the constant torque region and in order to -400
ensure a smooth transition from constant torque region into
flux weakening operation. The realization of the anti-reset id iq i∗d , i∗q
-600
windup scheme is shown in Fig. 8. It is assumed, that the
+ − 0 1 2 3 4
integral action is inherently limited to ψΔ and ψΔ . For the 0

ψΔ in mVs
design of the ARW scheme mainly two scenarios have to -2
be considered. First, in the constant torque range the voltage -4
control loop is not closed due to the minimum selection in the -6
operation point selection scheme in Fig. 6. Thus, the actual -8
modulation rate a is constantly smaller than the reference a∗ 0 1 2 3 4
and due to the positive control error the integrator action 1
+
saturates in the positive limit ψΔ . In case of a transition 0.9

a
into the flux weakening region, the integrator action has
0.8
a a∗
to be reduced by a negative control error, which may take
some time. Meanwhile, the voltage output of the current 0 1 2 3 4
controller remains saturated and thus the decoupling of the t in s
current components is not working properly. If this condition
Fig. 9. Alternating Torque Requests in Flux Weakening Range at a Speed
lasts too long, excessive motor currents may occur. To avoid of 8000 min−1
this circumstance, the integrator action is initialized with a
predefined value if the integral controller is at the positive
+
n in 1000/min

limit ψΔ and starts to decrease due to a negative control 12


error. In the second scenario, the motor is operated close to 10
the flux weakening range. The integral controller action is 8
+
assumed to be just before reaching the positive limit ψΔ when 6
a dynamic transition into flux weakening operation occurs. 4
In that case the ARW scheme just described above does 0 1 2 3 4
200
not go into effect, causing delayed voltage control actions,
id , iq in A

0
which may in turn be critical for keeping the currents within -200
their valid operating limits. To prevent such conditions, the id iq i∗d , i∗q
-400
integral controller action is also reset if the modulation rate
-600
a exceeds a predefined threshold ath . A low-pass filter is 0 1 2 3 4
utilized to ensure, that the integrator reset is not triggered 20
by transient voltage peaks. The low-pass filter time constant
ψΔ in mVs

10
and the threshold value ath are adjusted empirically on the 0
test bench. -10
0 1 2 3 4
|ωel | Kp
uDC 1
0.9
a

a∗ Integral ψΔ
Controller 0.8 a a∗
a 0.7
”‘Hold & Reset”’ 0 1 2 3 4
t in s
LP ARW +
≥ ath = ψΔ Fig. 10. Transition from Constant Torque Range to Flux Weakening Range
Logic

Fig. 8. Voltage Controller Scheme


utilized 60 kW electrical traction IPMSM for sub-compact
electrical vehicles was employed as test specimen. The speed
was set by a load motor. Exemplary experimental results
V. E XPERIMENTAL R ESULTS are shown in Fig. 9 and Fig. 10 for two different operation
The control scheme presented in Fig. 1 has been im- scenarios. In Fig. 9, the operation of the voltage controller
plemented on a dSpace rapid-prototyping system. A highly at constant speed and alternating torque steps in the flux


weakening range is demonstrated. The motor is operated in R EFERENCES
motoring (iq > 0) as well as generating (iq < 0) mode. The [1] J. Wiesing, “Betrieb der feldorientiert geregelten Asynchronmaschine
modulation rate a is noisy due to the variation of the current im Bereich oberhalb der Nenndrehzahl,” Dr.-Ing. Thesis, Paderborn
controller outputs udq caused for example by current harmon- University, 1994.
[2] J. Böcker, J. Janning, and K. Anbuhl, “Realization of a high-dynamic
ics. However, its mean value is adjusted by the modulation discrete-time controller for pwm inverter-fed induction motor drives,”
controller according to the reference value a∗ with stationary in European Conference on Power Electronics and Applications (EPE),
accuracy. In Fig. 10, the motor is operated at constant torque, 1993.
[3] B.-H. Bae, N. Patel, S. Schulz, and S.-K. Sul, “New field weakening
but the speed is ramped up from 5000 min−1 to 12000 min−1 . technique for high saliency interior permanent magnet motor,” in
At first, the motor operates in the constant torque range, thus Industry Applications Conference (IAS), 38th IAS Annual Meeting,
the voltage controller is not active and the flux correction 2003.
[4] M. Meyer and J. Böcker, “Optimum control for interior permanent
ψΔ is equal to its positive limit of 20 mVs. Afterwards, the magnet synchronous motors (ipmsm) in constant torque and flux
modulation rate a exceeds the reference value a∗ and the weakening range,” in 12th InternationalPower Electronics and Motion
voltage controller is activated. Due to the proper design of Control Conference (EPE-PEMC), 2006.
[5] M. Meyer, “Wirkungsgradoptimierte Regelung hoch ausgenutzter
the anti-reset windup measure, the integral controller action is Permanentmagnet-Synchronmaschinen im Antriebsstrang von Automo-
reinitialized with a suitable flux value and the modulation rate bilen,” Dissertation, Paderborn University, 2010.
a is properly adjusted to its reference value a∗ . In both figures, [6] W. Peters, T. Huber, and J. Böcker, “Control realization for an interior
permanent magnet synchronous motor (ipmsm) in automotive drive
the current controller reference values idq are calculated by the trains,” in International Exhibition and Conference for Power Elec-
operation point selection and are set with stationary accuracy tronics, Intelligent Motion, Renewable Energy and Energy Management
by the current controllers. (PCIM Europe), 2011.
[7] W. Peters and J. Bcker, “Discrete-time design of adaptive current
controller for interior permanent synchronous motors (ipmsm) with high
VI. C ONCLUSION magnetic saturatuion,” in 39th Annual Conference of IEEE Industrial
Electronics Society (IECON), 2013.
The presented voltage controller enables a good utilization [8] W. Peters, O. Wallscheid, and J. Bcker, “A precise open-loop torque
of the DC-bus voltage while maintaining a suitable voltage control for an interior permanent magnet synchronous motor (ipmsm)
considering iron losses,” in 38th Annual Conference of IEEE Industrial
margin to ensure the stability of the current control loop. Electronics Society (IECON), 2012.
The voltage control loop model is approximated by a first- [9] W. Peters, “Wirkungsgradoptimale Regelung von permanenter-
order delay and the model parameters are identified based regten Synchronmotoren in automobilen Traktionsanwendungen unter
Berücksichtigung der magnetischen Sättigung,” Dr.-Ing. Thesis, Pader-
on measured step responses in different operation points. born University, 2015.
An integral controller is employed as voltage controller and [10] M. A. Laughton and M. G. Say, Electrical engineer’s reference book.
designed to be robust to plant parameter variations. Since Elsevier, 2003.
[11] R. Isermann and M. Münchhof, Identification of dynamic systems.
the voltage controller is not active in the constant torque Springer, 2011.
range, an adequate anti-reset windup scheme is required to
keep the voltage controller output in a predefined range and
to enable fast controller actions in case dynamic transitions
between constant torque and flux weakening range take place.
Test bench results based on an automotive electric traction
motor for sub-compact electric vehicles demonstrate the good
performance of the presented control scheme.



You might also like