Direct Torque Control of PM Brushless AC Motors Having

Non-sinusoidal Back-emf Waveforms

Y. Liu, Z.Q. Zhu and D. Howe
Department of Electronic and Electrical Engineering,
University of Sheffield, Mappin Street,
Sheffield SI 3JD, UK

Keywords: Direct torque control, PM brushless motor, control strategy together at a system level. The paper
sliding mode observer, non-sinusoidal back-emf describes the application of direct torque control (DTC) to a
PM brushless AC motor which does not have a perfectly
Abstract sinusoidal back-emf waveform, and compares its performance
with that which results when the motor is assumed to have a
This paper describes the application of direct torque control sinusoidal back-emf waveform. Although various back-emf
(DTC) to a PM brushless AC machine whose back-emf observers may be employed, accurate machine parameters are
waveform is non-sinusoidal, and compares its performance often required. In practice, however, variations in the winding
with that which is achieved when the back-emf waveform is resistance and inductance are inevitable, and can degrade the
assumed to be sinusoidal. It also demonstrates the utility of control performance significantly. Thus, a sliding mode
employing a sliding mode observer to estimate the back-emf observer, which uses the polarity of the output error rather
waveform, and, thereby, to calculate the instantaneous than the absolute value and which increases the robustness to
electromagnetic torque and to minimise torque ripple. The system disturbances, parameter variations and noise, is
effectiveness of this approach is validated by both simulations employed. The effectiveness of this approach is validated by
and measurements. both measurements and simulations.

1 Introduction 2 Electromagnetic torque estimation

It is desirable for a permanent magnet brushless AC motor to The electromagnetic torque of a BLAC motor with an ideal
have sinusoidal back-emf and current waveforms in order to sinusoidal back-emf waveform can be expressed in the
maximize the efficiency and torque capability and to stationary a,B-reference frame as:
minimize torque ripple. Various design features may be 3 p
employed to obtain a sinusoidal back-emf waveform [1-3]. T= (sajsf8is-s,isa) (1)
For example, the stator slots and/or rotor magnets may be where Vs,sa',s is, sa and isfi are the stator flux-linkages and
skewed, a distributed stator winding may be employed, the
magnets might be appropriately shaped or magnetised, etc. stator currents in the a- and ,-axis, respectively, andp is the
However, while such methods reduce the harmonic content in number of poles.
the back-emf waveform, and, thereby, also reduce torque However, for a machine with a surface-mounted magnet rotor
pulsations, they reduce the average torque and increase and a non-sinusoidal back-emf waveform, the
manufacturing complexity. Further, in practice, it is inevitable electromagnetic torque can be calculated from [6]:
that harmonics exist in the back-emf waveform. 3 p dO di1 /3 (2)
Various vector control techniques have been adopted to T = --[ r isa + is,] (2)
2 2 dOe dOe
minimize torque pulsations by optimising the q-axis reference
current waveform [4-5]. By measuring the non-sinusoidal where Oe is the rotor electrical angle, Y'ra , /r/3p are the rotor
back-emf waveform off-line, appropriate harmonics can be flux-linkages in the a- and ,-axis, respectively. The two
added to the q-axis reference current to meet a constant torque differential items in (2) can be derived from the ratio of the
demand. However, in order to accurately track the reference back-emf to the electrical angular velocity, since they change
current, a very high current loop bandwidth is essential, much slower than the other electrical variables. Hence,
which significantly increases the complexity of the system. divfra d/VIra 1 d/Ira ea.
Moreover, parameter variations should be considered for high dOe d(WOet) Woe dt WOe
performance control.
Direct torque control (DTC) provides a more attractive d / rj dYirf 1 d V/rij eg (4)
approach since it can improve the steady-state and dynamic dOe d(cWet) We dt We
performance without the need to transform the control where coe is the rotor electrical angular velocity, and ea and
variables from torque to current. It directly controls the flux-
linkage and the electromagnetic torque, and considers the e8 are the a- and ,¢axis back-emfs, respectively.
electrical machine, the power electronic inverter, and the

425
Thus, for a motor with a non-sinusoidal back-emf waveform, frequency ripple in the estimated variables may be reduced by
the electromagnetic torque can be estimated from equation using low-pass filters.
(2), provided that the back-emf waveform can be estimated.
Although several techniques can be used for the back-emf 3 Simulation and experimental investigation
observer, accurate motor parameters are generally required
for most high performance observers. In practice, however, A block diagram of the DTC BLAC drive is shown in Fig. 1,
winding resistance and inductance variations are inevitable, DTC strategy being implemented on a TMS320C31 DSP-
which will degrade the control performance significantly. based controller, which includes an interface to an
However, a sliding mode observer which uses the polarity of incremental rotor position encoder. The surface-mounted
the output error instead of the absolute value has an inherent magnet brushless motor has the non-sinusoidal back-emf
high gain feature and increases the robustness to system waveforms, shown in Fig.2, its parameters being given in
disturbances, parameter deviations, and noise. Thus, such an Table I. Simulations of the DTC BLAC drive were carried out
observer is employed to estimate the non-sinusoidal back-emf on the MATLAB/SIMULINK platform, and the performance
waveform and, thereby, to calculate the electromagnetic which results with the proposed DTC implementation was
torque. compared to that which results when a sinusoidal back-emf
Assuming that =L5is
L + where yi5 t and is are the waveform is assumed. Fig.3 compares the sliding mode
stator flux-linkage vector, the rotor flux-linkage vector and observer estimated back-emf waveform in the ox,-reference
the current vector, respectively, the a- and ,-axis voltage frame.
equations can be expressed as [7][8]:
dis e
= isaRs +Ls dt (5)
Uau saR +S Sdt+s a
up=is,RS Lsdt e
dis, (6)
where L, is the stator winding inductance.
If only low order harmonics in the back-emf waveform are
considered, it is also appropriate to assume that ea = 0 and
ep8= 0. Hence, choosing x- and 3-axis currents and back-
emfs as the state-variables, the following state-variable
equations can be established: Figure 1 Block diagram of DTC BLAC drive.
x = Aex+ Beu (7) Table I Specification of surface-mounted PM BLAC motor
Y=Cex (8) Number of poles, p 10
where x=[isa, i,,8 ea, eg]T is the vector of state-variables, y = DC link voltage (V) 36
[isa, is/]T is the output vector, and u=[ua, u/8]T is the input Rated speed (rpm) 400
vector, and PM excitation flux-linkage (Wb) 0.0794
Rs 0 0 1 Phase resistance (Q) 0.35
Ls Ls Self-inductance (mH) 3.9
Ae eLRs
= 0 0
I
B= 0
S, CeF1 100
00(9) Mutual-inductance (mH) -0.0023
O 0 0 0 0 0
O 0 0 0 0 0 By way of example of the performance which is achieved,
Figs.4 and 5 show the simulated phase current and
From (7) and (8), the sliding mode observer can be electromagnetic torque, in both time and frequency domains,
constructed as: which result with the conventional and the proposed DTC
x = Aex + Beu + Ksign(y-Cex) (10) strategies, respectively. As will be seen, the flux-linkage loci
I if cY0 -k, 0- are similar, and both are essentially circular. In general, a
with sign(o) = l20 if cs=0 and K= ° kj (11) non-sinusoidal back-emf waveform will result in a non-
-1if (7<0 k2 o circular the stator flux-linkage locus, and in order to maintain
[O k2] constant electromagnetic torque, either the amplitude of the
stator flux-linkage, or the load angle between the stator and
where K is the gain matrix. Due to the symmetry of the a- rotor flux-linkages may be varied [9]. However, with DTC,
and flcomponents in the state-variable equations, K can be the stator flux-linkage command is specified to be constant,
assumed to have the above form. The choice of the gains is a and the rotational speed of the stator flux-linkage vector is
compromise between the operating range of the sliding mode adjusted to increase or decrease the load angle so as to
and the resultant ripple in the observed variables. Large gains minimize the torque ripple. However, the variation in the
lead to a wider observer range but may cause excessive ripple rotational speed of the stator flux-linkage vector does not
in the observed variables, which may lead to the observed affect the shape of the stator flux-linkage locus, i.e. it remains
variables being unusable for control. However, high circular even when the back-emf waveform is non-sinusoidal.

426
As will be seen by comparing Figs. 4(a) and 5(a), the phase 20

current waveforms differ slightly since the proposed DTC

strategy ensures that the current waveform caters for the non- 10
sinusoidal back-emf waveform so as to maintain constant
electromagnetic torque. Hence, the low frequency torque 100
ripple, which results with conventional DTC when a
sinusoidal back-emf waveform is assumed, is eliminated, this -10
will be evident by comparing Figs.4(c) and 5(c), in which it
should be noted that although high frequency torque
components associated with PWM are present with both DTC -20
0.025 0.0375 0.05 0.0625 0.075
strategies, the low-frequency (6th) torque harmonic (viz. the Time (s)
198Hz component which corresponds to operation at 400rpm (a) Simulated
(i.e. 33.3Hz for the 10-pole BLAC motor under
consideration)), which is present with conventional DTC is
eliminated when the sliding mode emf observer is
incorporated with DTC. The corresponding measurements
shown in Figs.6 and 7 confirm the findings from the ,~~
simulations. Hence, the inherent capability of DTC to g
minimise torque ripple has been demonstrated.

4 Conclusions
Time (12.5ms/div)
The paper has demonstrated the utility of applyingD direct (b) Measured
torque control to a permanent magnet brushless ac moltor with
a non- sinusoidal back-emf waveform. A slidingr mode Figure 3 Sliding mode observer estimated ca- and I-
observer has been employed to obtain the instantaneouis back- components of back-emf waveform.
emf and, thereby, to calculate the instan taneous
electromagnetic torque. In this way, the low frequenc)ytorque References
ripple which would have resulted with conventionaLI DTC. and W. L. Soong, "Pulsating torque
when a sinusoidal back-emf waveform is assumed, h as been [1] T. M. Jahns
eliminated. The inherent capability of DTC to m inimise minimization techniques for permanent magnet AC motor
drives a review," IEEE Trans. Industrial Electronics, vol.
torque ripple has been confirmed by both simulatic)ns and 43, no. 2, pp. 321-330. (1996)
measurements. [2] J. R. Hendershort, and T. J. E. Miller, 'Design of brushless
... ...
permanent-magnet motors,' Magana Physics Publishing
\Eab and Clarendon Press, Oxford. (1994)
15 [3] T. Kenjo, and S. Nagamori, 'Permanent magnet and
brushless DC motors,' Clarendon Press, Oxford. (1985)
[4] S. Bolognani, L. Tubiana and M. Zigliotto, "Sensorless
control of PM synchronous motors with non-sinusoidal
back-EMF for home appliance," IEEE Int. ConfJ Electric
-15 Machines and Drives, vol.3, pp.1882-1888. (2003)
[5] P. Mattavelli, L. Tubiana and M. Zigliotto, "Torque-ripple
-30 reduction in PM synchronous motor drives using repetitive
0.0125 0.025 0.0375 0.05 current control," IEEE Trans. Power Electronics, vol.20,
Time (s)
no.6, pp.1423-1431. (2005)
(a) Phase and line [6] Y. Liu, Z. Q. Zhu and D. Howe, "Direct torque control of
20 brushless DC drives with reduced torque ripple," IEEE
Ea Trans. Industry Applications, vol.41, no.2, pp.599-608.
\-
10 ><
I\ ' (2005)
[7] C. Edwards and S. K. Spurgeon, "Sliding mode control:
theory and applications," Taylor and Francis, London.
\l
100
to0 /1 +- Xi
(1998)
[8] F. Parasiliti, R. Petrella and M. Tursini, "Sensorless speed
-10 control of a PM synchronous motor based on sliding mode
observer and extended Kalman filter," Conf: Record 36th
IEEE IAS Annual Meeting, pp.533-540. (2001)
20
0 0.0125 0.025 0.0375 0.05 [9] L. Zhong, M. F. Rahman, W. Y. Hu and K. W. Lim,
Time (s) "Analysis of direct torque control in permanent magnet
(b) oc- and ,-components synchronous motor drives," IEEE Trans. Power
Figure 2 Off-line measured back-emf waveforms. Electronics, vol. 12, no.3, pp.528-536. (1997)

427
 2.5 2.5

1.25 1.25

¢0 O

-1.25
-1.25

-2.5
-2.5
0.025 0.05 0.075 0.1
Time (s) 0 0.025 0.05 0.075 0.1
Time (s)
(a) Phase current (a) Phase current
2 _

1.5 1.5 -

I
lhiallial iihiloigiii IlialidilliAml o
1I
IC IIISITLTITfS
G)

0 I- .11.

0.5 0.5
t-

0 0
0 0.025 0.05 0.075 0.1 0 0.025 0.05 0.075 0.1
Time (s) Time (s)

Q#
(b) Electromagnetic torque
i;i;iEE 'ji :fy :E
iS0i iLSLS :0E2 aE.jEAiEEETREHi #iEii E: SiN ~ iSe iR;tHE TEN iE St jg iR C.

(c) Harmonic spectrum of electromagnetic torque (c) Harmonic spectrum eiectromagnetic torque oI

0.1 0.1

0.05 0.05

-0.05 -0.05

-O. I O
~~~~~~~~~~~~~~~~~~~~-0.1
-0. 1 -0.05 0 0.05 0.1 -0. 1 -0.05 0 0.05 0.1

'lsa (Wb) Vsa (Wb)

(d) Locus of stator flux-linkage (d) Locus of stator flux-linkage
Figure 4 Simulated performance with conventional DTC Figure 5 Simulated performance with proposed DTC strategy
strategy (400rpm) (400rpm)

428
 Time (25ms/div) Time (25ms/div)

(a) Phase current (a)I Phase current

I

Time (25ms/div) Time (25ms/div)

(b) Electromagnetic torque

Frequency (250 Hz/div) Frequency (250Hz/div)

(c) Harmonic spectrum of electromagnetic torque (c) Harmonic spectrum of electromagnetic torque

. !4

Oi

Cm
3

W,a (0.05NWb/div) W,a (0.05Wb/div)

(d) Locus of stator flux-linkage (d) Locus of stator flux-linkage
Figure 6 Measured performance with conventional DTC Figure 7 Measured performance with proposed DTC strategy
strategy (400rpm) (400rpm)

429