Академический Документы
Профессиональный Документы
Культура Документы
shown in (1) for a voltage transformation. For a machine with Fig. 1. Reference Frames, Modified from [4]
a floating star point, the zero sequence component i0 is zero.
1 − 21 − 21 450
r
2
√
3
√
3
vαβ0 = 0 − 2 vabc (1) 400
3
2
√1 √1 √1 350
2 2 2
Inductance (µH)
All of these reference frames are shown diagrammatically in Fig. 2. Inductance Variation as a Function of Rotor Position
fig. 1.
IV. M AGNETIC S ALIENCY
ensure that insignificant torque generation results as a conse-
The motor used in the development of the sensorless tech- quence of the injection process. The resulting winding currents
nique is an interior magnet permanent magnet synchronous are measured using Hall effect transducers. The currents are
machine designed for use in a mid-size fuel cell powered converted from the three-phase abc representation to the two-
vehicle [19]. The design is based around an embedded mag- phase αβ form. As a consequence of the magnetic saliency,
net rotor construction and a concentrated wound stator. The the locus of the αβ currents is ellipsoidal, with a significantly
embedded magnets in the rotor result in an inherent magnetic greater current magnitude in the d-axis compared to that in the
saliency that produces a detectable variation in the inductance q-axis (due to the fact that Lq > Ld ). Fig. 3 illustrates these
of the stator windings as the rotor turns, as illustrated in fig. 2. ellipsoidal current loci obtained from static current injection
Therefore, if the inductance of the rotor (or any property of tests where the rotor has fixed angular displacements of 60◦
the system that is directly related to the inductance) can be and 90◦ . The two rotor positions result in two distinct orien-
detected, the rotor position can be derived. tations of the ellipse. This property can be used to determine
rotor position.
V. S ALIENCY D ETECTION
Whilst this method of using saliency to detect rotor position
A. Principle has already been proposed [13], the existing techniques gen-
A balanced set of high frequency, low magnitude sinusoidal erally rely upon complicated and computationally expensive
voltages is applied to the machine terminals using a PWM ellipse fitting algorithms [13], where accuracy may be easily
based drive. The frequency and magnitude are chosen to compromised by the presence of electrical noise. Alternatively,
2437
1
shown in (5) and (6).
0.8
cos θr − sin θr Ld 0 cos θr sin θr
Lαβ = (5)
0.6 sin θr cos θr 0 Lq − sin θr cos θr
0.4 L̄ − L̆ cos 2θr −L̆ sin 2θr
= (6)
0.2 L̆ sin 2θr L̄ + L̆ cos 2θr
Normalised iβ
0 where,
-0.2 1
L̄ = (Lq + Ld ) (7)
-0.4
2
1
L̆ = (Lq − Ld ) (8)
-0.6 2
-0.8 The balanced voltage set that is injected into the windings can
be represented by (9)
-1
-1 -0.5 0 0.5 1
Normalised iα − sin(ωi t)
vαβ = Vi (9)
(a) cos(ωi t)
where Vi is the magnitude of the injected signal. Combin-
1
ing (4), (6) and (9) gives the currents that result from voltage
0.8 injection and which contain the rotor position:
0.6
cos(ωi t) cos(2θr − ωi t)
iαβ = I1 − I2 (10)
0.4 sin(ωi t) sin(2θr − ωi t)
0.2 where I1 and I2 are terms dependent upon machine induc-
Normalised iβ
0
tance, and the frequency and magnitude of the set of injected
voltages [17].
-0.2
-0.4
C. Derivation of Rotor Position
To isolate the rotor position information, θr , contained in the
-0.6
second term of (10) the αβ currents are rotated to a reference
-0.8 frame that is synchronous with the injected voltage set and
-1 then passed through a high-pass filter. The resulting rotated
-1 -0.5 0 0.5 1
Normalised iα and filtered currents are represented by (11).
(b) cos(2θr − 2ωi t)
i = −I2
′
(11)
sin(2θr − 2ωi t)
Fig. 3. Examples from Experimental Data of Ellipses Rotating with Rotor
Position (a) θr = +60◦ ; (b) θr = +90◦ The final stage is to rotate (11) through an angle of
θ = −2ωi t rads to give (12) where the angle of the result is
directly related to the rotor angle. A simple Cartesian-to-polar
relatively complex techniques that use a combination of ma- transformation completes the sensorless scheme.
chine and saliency models with PID controllers have also
been proposed [15], [17]. In contrast, the method proposed cos(2θr )
i′′ = −I2 (12)
in this paper uses a simple demodulation technique that is sin(2θr )
computationally simple and fast. In most practical motors however, the current loci obtained
at different rotor positions do not follow the same ellipsoidal
B. Algorithm Basis
shape since, due to such factors as stator slotting, the saliency
The standard model for a permanent magnet synchronous profile is not sinusoidal. This results in considerable error in
machine is shown in (3). the estimated position. The distorted saliency can be modelled
Rs + pLd −ωr Lq 0 by modifying (10) to include additional distorting compo-
vdq = idq + (3) nents [17]. This multiple saliency model is shown in (13).
ωr Ld Rs + pLq ψm ω r
If the angular frequency of the injection signal, ωi , is sufficient, cos (ωi t)
iαβ = I1
the resistance of the windings can be considered negligi- sin (ωi t)
Pn (13)
ble compared to the reactance. Considering only the high- cos (hk θr − ωi t + φk )
+ Ik
frequency components of the system, the machine model may k=2 sin (hk θr − ωi t + φk )
be approximated by [17]: In tracking observer based systems, additional saliency models
vαβ ≈ jωi Lαβ iαβ (4) can be added to the observer for each component in (13)
in order to compensate for the saliency distortion [17]. For
where Lαβ is the result of transforming the dq-axis induc- significantly distorted saliencies, this adds considerable com-
tances into the two phase stationary reference frame [18] as plexity to the already complex tracking observer. In many
2438
7
applications, the requirement is for simple and robust schemes Measured
to remove the possibility of failure. To achieve this goal, an Estimated
6
alternative technique is used to compensate for position error.
Through examination of the results obtained from this and
Motor
Gate
CTs
Drive perfectly sinusoidal saliency, this profile would be circular.
Test results have shown that different fingerprints can be
Resolver
clearly identified for different machines.
0.6
Fig. 4. Test Platform Block Diagram
0.4
0.2
VII. R ESULTS
Normalised i′′y
quency was injected into the windings of the PMSM. Applica- -0.2
tion of the sensorless position scheme without compensation
-0.4
resulted in the position information presented in fig. 5.
The results obtained from the sensorless position measure- -0.6
ment scheme show excellent agreement with those obtained
-0.8
from the resolver. The error plot of fig. 6 shows that the
maximum error over one cycle is 2.75◦ mechanical. -1
-1 -0.5 0 0.5 1
To demonstrate the saliency distortion discussed in sec- Normalised i′′x
tion V-C, the ‘saliency fingerprint’ obtained from the one-and-
a-half slots per pole test machine is shown in fig. 7. For a Fig. 7. Saliency Fingerprint
2439
Analysis of the saliency fingerprint allows the technique to single-dimensional table with a relatively low resolution –
be tuned for a particular motor. This improves the accuracy intermediate points can be found by interpolation. For the
of the position sensing algorithm and enables machines with machine discussed in this paper, the correction table is shown
significantly distorted saliencies to be driven in brushless graphically in fig. 10. Due to the simplicity of the look-up table
AC mode. A simple look-up table can be used to correct
for the errors produced by the distorted saliency. This table 3
1
-1
0.8
0.6
-2
0.4
Normalised i′′y
0.2
-3
0 -3 -2 -1 0 1 2 3
Lookup Table Input (radians)
-0.2
-0.4
Fig. 10. Correction Data
-0.6
1 4
0.8
3
0.6
0.4
2
Normalised i′′y
0.2
0 1
-0.2
0
-0.4 0 2 4 6 8 10 12 14
Rotor Electrical Position (radians)
-0.6
2440
0.06
[7] A. Consoli, S. Musumeci, A. Raciti, and A. Testa, “Sensorless vector
Error in Corrected Electrical Position Estimation (radians)
2441