Analysis of Multi-Phase Permanent-Magnet Synchronous Machines

T.J.E. Miller and M.I. McGilp

Abstract Large AC machines are sometimes fed by multiple inverters. This paper presents the complete steadystate analysis of the PM synchronous machine with multiplex windings, suitable for driving by multiple independent inverters. Machines with 4, 6 and 9 phases are covered in detail. Particular attention is given to the magnetic interactions not only between individual phases, but between channels or groups of phases. This is of interest not only for determining performance and designing control systems, but also for analysing fault tolerance. It is shown how to calculate the necessary self- and mutual inductances and how to reduce them to a compact dq-axis model without loss of detail. Fig. 1 Duplex 3-phase and 2-phase windings Index Terms multi-phase machines, permanent-magnet machines.

I. INTRODUCTION A. Multiple phases The advantages of polyphase machines and systems are well known and long established. For example the 3-phase 3-wire system conveys 50% more power with only 87@9% of the conductor area, compared with a 2-phase 3-wire system, [1]. In electrical machines the winding factors can be increased, and the torque ripple reduced, by using multiple phases. Space-harmonics in the airgap flux are also reduced. B. Multiplex windings A multiplex winding of plex x has x m-phase sets of balanced polyphase windings, giving mx phases. For example Fig. 1(a) shows a 6-phase winding that is a duplex 3-phase winding, with m = 3 and x = 2. Fig. 1(b) shows a 4-phase winding with m = 2 and x = 2. A triplex winding with m = 3 would have 9 phases. Most conventional machines have simplex windings with x = 1, including the 5-phase machines studied by Parsa and Toliyat [4,5]. Even for induction motors there appears to be very little published work on multiplex machines [6,7]; and evidently none on PM machines. In practice the displacement " between the sets must be chosen so that the fluxes in common sections of the magnetic circuit make sensible utilization of the available cross-section, and to make the slot-fill factors even. We shall see later that the mutual coupling between sets imposes further constraints on ". C. Reasons for using multiplex windings The increased phase number extends the advantages of the simplex machine mentioned earlier, but additional advantages can arise with very large machines.

For example, it may be more economical to use, say, x inverters of 1/xth the rating, rather than one huge inverter of 100% rating. With x separate inverters instead of 1, there is a chance of being able to operate even when one of the inverters is malfunctioning or switched off. It is also possible to operate as many as x independent channels, by connecting the sets of windings to different buses. This can be done with pure AC systems, or with a DC/inverter system, as in Fig. 2. We should be careful to differentiate these machines from fault-tolerant machines, most examples of which (in the research literature) are simplex machines with a high degree of magnetic isolation between phases. The machines considered here have normal magnetic coupling between phases, and additional magnetic coupling between sets of phases, which is the main subject of this paper. II. BASIC ANALYSIS The windings of the 6-phase machine in Fig. 1 are labelled a1b1c1 and a2b2c2. Each 3-phase winding is wyeconnected and is balanced in itself. In general all six phases are mutually coupled. If the rotor rotates anticlockwise, the voltages and currents in set 2 lead those in set 1 by the phase angle ", which is the angular displacement between the two 3-phase windings.

Fig. 2 Duplex wye/delta system; DC-fed; 6 phases

The rotor d-axis is at the angle 21 relative to the axis of phase a1, and at 22 relative to the axis of phase a2, and " ' 22 21 .
(1)

By similar reasoning for the q-axis coils, Mq1q2 ' Lmq mF12q ( " ) .
(8)

The q-axis leads the d-axis by B/2 electrical radians. The dq transformation applied to a1b1c1 gives vd1 ' R1 id1 p Rd1 T Rq1 ; (2) vq1 ' R1 iq1 p Rq1 T Rd1 . Normally with only 3 phases, there is just one d-coil and one q-coil and the flux-linkages are given by Rd1 ' QMd1 Rq1 ' Lq1 iq1 , Ld1 id1 ;
(3)

In general, the currents (and voltages) in the second set must be phase-shifted by " relative to those in the first set. In a duplex wye-connected 6-phase machine, there are only 2 coils on the d-axis and 2 coils on the q-axis, and the two sets of phases will be balanced as long as Md1d2 = Md2d1 and Mq1q2 = Mq2q1, the normal condition of reciprocity. However, consider the extension to a 9-phase machine with 3 coils on the d-axis and 3 on the q-axis. We will have Md1d2 ' Lmd Md2d3 ' Lmd Md3d1 ' Lmd mF 12d ( " ) ; mF 23d ( " ) ; mF 31d ( " ) .

where QMd1 is the flux-linkage produced by the magnet. We are going to transform the second winding a2b2c2 into a second pair of coils d2q2 on the same direct and quadrature axes as the d1q1 coils, and these will be coupled to the d1q1 coils, so we can write Rd1 ' QMd1 Rq1 ' Lq1 iq1 Ld1 id1 Md1d2 id2 ; Mq1q2 iq2 .
(4)

(9)

The model can be completed by adding the equations for the a2b2c2 winding corresponding to (2) and (4): vd2 ' R2 id2 vq2 ' R2 iq2 Rd2 ' QMd2 Rq2 ' Lq2 iq2 p Rd2 p Rq2 T Rq2 ; T Rd2 ,
(5)

If mF12d, mF23d and mF31d are not all equal, it will be impossible to achieve balance between the three sets of windings. Even if they are all supplied with the same currents phase-shifted by 0, " and 2" respectively, the voltages will be unequal because of the differences in the mutual inductances. In general, we can say that a winding of higher multiplicity (x > 2), cannot be balanced unless all the d-axis mutual inductances are equal; and all the q-axis mutual inductances are separately equal. To pursue this further, we need a matrix analysis of the inductances. III. MATRIX ANALYSIS In terms of flux-linkages [R] and currents [i], the dq-axis transformation for plain 3-phase machines is summarized as follows, where [T] is the transform matrix used in [1-3]: [Rabc] ' [Labc] [iabc] ; [idq0] ' [T] [iabc] ; [iabc] ' [T] 1 [idq0] .
(10) (11) (12)

and

Ld2 id2 Md2d1 id1 ; Mq2q1 iq1 .

(6)

The mutual inductance Md1d2 or Md2d1 between the d1 and d2 coils is important, and we must consider whether it depends on the displacement angle ". Later we will also consider the possibility of cross-coupling between the d and q-axes, which is absent from (4) and (6). If " = 0, and if all phases have the same effective number of turns, then phases a1 and a2 will be aligned with the d-axis at the same time. In this case we expect Md1d2 to be equal to Lmd, the magnetizing component of the synchronous inductance Ld, plus another term mF12d arising from mutual coupling in the slots and in the end-windings: Md1d2 ' Lmd mF 12d ( " ) .
(7)

[Rdq0] ' [T] [Rabc] ' [T] [Labc] [T] 1 [idq0] ' [Ldq0] [idq0] .

The matrix [Labc] is well known to contain constant terms and second-harmonic terms in rotor position 2, and the familiar result is Ld [Ldq0] ' Lq L0 where Ld and Lq are the synchronous inductances. To consider the duplex 6-phase machine we use a full square inductance matrix (15) of the following form, of order 6 6, partitioned between the two sets of phases: a1b1c1 a1b1c1 a2b2c2 [A1] [BN] a2b2c2 [B] [A2]
(14) (13)

The mutual inductance term mF12d may include a component due to differential or harmonic leakage, while Lmd is related to the conventional self- and mutual leakage inductances LF and MF, and is also associated with the fundamental ampere-conductor distribution. Any relationship between LF, MF and mF12d is not immediately obvious, but mF12d can be calculated formally by applying the dq-axis transformation to the entire inductance matrix as shown below. If " = B/2, phase a2 will be aligned with a q-axis of the rotor when phase a1 is aligned with a d-axis. If Ld and its components Lmd, LF and MF are independent of the orientation of the d-axis relative to the winding axes, (7) remains valid, although the value of mF12d will be different.

The full form of this matrix is a1 b1 c1 a2 a1 La1 Ma1b1 Lb1 Mc1b1 Mb1a2 Mb1b2 Mb1c2 Ma1c1 Mb1c1 Lc1 Mc1a2 Mc1b2 Mc1c2 Ma1a2 Mb1a2 Mc1a2 La2 Mb2a2 Mc2a2

to the complete 6-way matrix a1b1c1a2b2c2, we get b2 Ma1b2 Mb1b2 Mc1b2 Ma2b2 Lb2 Mc2b2 c2 Ma1c2 Mb1c2 Mc1c2 Ma2c2 Mb2c2 Lc2 d1q101 d2q202 d1q101 [T1][A1][T1] [T2][BN][T1]
1

d2q202 [T1][B][T2] [T2][A2][T2]

(19)

b1 Mb1a1 c1 Mc1a1 a2 Ma1a2 b2 Ma1b2 c2 Ma1c2

If [B] ' 0, there is no mutual coupling between the a1b1c1 windings and the a2b2c2 windings. In that case the dq0 transformation [T1] = [T(21)] can be applied to [A1], and [T2] = [T(22)] to [A2],1 and the result is two completely independent, uncoupled dq0 systems d1q101 d1q101 d2q202 [T1][A1][T1]
1

We already have the upper-left and lower-right partitions, and we also have the Lmd and Lmq terms in the upper-right and lower-left partitions. The remaining mutual inductance terms are treated independently, by transforming the matrix of mutual inductances between pairs of phases of different channels, excluding the airgap fluxes which give rise to Lmd and Lmq. To do this, define 8 as the mutual inductance between a1 and a2; between a1 and b2; and < between a1 and c2. 8, and < are confined to slot-leakage flux, end-winding flux, and harmonic leakage, and we can consider them to be independent of rotor position. The matrix to be transformed is a1 b1 c1 a2 8 < 8 < < 8 < 8 b2 8 < c2 < 8

d2q202
(16)

a1 b1 c1 a2 b2 c2

[T2][A2][T2]

In practice the d1q101 system is mutually coupled to the d2q202 system, and we have already postulated the form of the mutual inductances in (7), (8) and (9). The situation so far can be summarized by writing the duplex dq0 inductance matrix as follows. d1 d1 q1 01 d2 q2 02 Lmd + mF12d Lmq + mF12q Ld1 Lq1 L01 Ld2 Lq2 L02 q1 01 d2 Lmd + mF12d Lmq + mF12q q2 02

Thus for the upper-right partition, 8 [T1][B][T2]

8 <

< 8 @ [T2]
1

= [T1] @

<

(21)

Without writing out the entire solution, it is instructive to work out and examine just the (1,1)th element, which represents mF12d : [T1][B][T2] 1 (1,1) ' mF12d ' 8 cos (22 cos (22 Since 22 21) 21 120E)
(22)

Lmd and Lmq are the normal magnetizing inductances in the two axes, but the terms in mF12d and mF12q require more analysis. If we apply the transformation d1q101 d1q101 d2q202
1

d2q202
(18)

[T1] [T2]

< cos (22 21 120E) . 21 ' ", we can write this as mF12d ' 8 cos " cos (" 120E) < cos (" 120E) .

(23)

By the same process we find that [T1][B][T2] 1 (2,2) ' mF12q ' ' 8 cos " cos (" 120E) ' mF12d . < cos (" 120E) (24)

The notation [T(2)] means that the elements of [T] are trigonometric functions of rotor position 2, as is known; see [1]-[3].

Fig. 3 Angles for considering 8, and <

This shows the feature we might have hoped to avoid: that although mF12d = mF12q, these mutual inductances appear to depend on ", and further analysis is necessary to find out the constraints that apply to the choice of " for duplex, triplex, and higher multiplex windings. For the duplex winding we have already observed that it does not matter if mF12d and mF12q depend on ", because there is only one mutual inductance on the d-axis and one on the q-axis. For triplex windings we can expect mF12d and mF23d both to be given by (23) directly, because the displacement angle between sets 1 and 2 is the same as the displacement angle between sets 2 and 3, both being equal to ". But the displacement angle between sets 1 and 3 is 2", so we expect mF31d to be given by (23) with 2" substituted in place of ". Now we know that it is necessary to have mF12d ' mF23d ' mF31d
(25)

to operate a triplex winding balanced. So what we really need, at this stage, is to understand the conditions under which (23) gives the same value for 2" as it does for ". Now consider the mutual inductances 8, , < in terms of the winding harmonics. From Fig. 3 we can see that 8 is a function of the angle ", that is, 8 = 8("). Similarly is a function of the angle (" 120E), and if we assume the same functional dependence, we can write = (" 120E). Finally < = <(" + 120E), again with the same functional dependence on the angle. For each harmonic component of the inductance, let 8 ' 7n cos n " ; ' 7n cos n ( " < ' 7n cos n ( " 120E ) ; 120E ) .

This implies that " = 20E, 40E or even 80E is acceptable for triplex (9-phase) windings, but not 15E or 30E. On the other hand 15E and 30E are acceptable for duplex (6-phase) windings. In general it appears that " = 180k/mxE gives the required result expressed by eqn. (25), at least for duplex and triplex cases. It is of interest to give some thought to the conditions that would make 8 = = < = 0. If this could be achieved, then any value of the displacement angle " would be acceptable with any value of x. If all the mutual coupling between phases of different sets were via the slot-leakage flux, this condition could be satisfied by ensuring that the windings of a1b1c1 have no shared slots with a2b2c2 or a3b3c3. The differential leakage is more troublesome, since it is a function of several spaceharmonics of MMF having different pole-pitches. For example, if " ' 30E there should be no third-harmonic linkage between a1 and a2, since 3 30 ' 90E, rendering a1 and a2 orthogonal and therefore uncoupled for this harmonic. The same cannot be said of the non-triplen harmonics 5th, 7th, etc. One might hope that the differential leakage is small; but the very nature of multiplex windings is to reduce the number of slots per pole per phase, potentially increasing the winding factors of some of the harmonics we would like to eliminate. The conclusion is that it is safest to select " from the preferred values indicated on the previous page, when designing triplex or higher-multiplex windings. Finally let us test the symmetry of the combined dq0 matrix by evaluating the [1,1]th term of [T2][BN][T1] 1, the lower-left partition. The evaluation is similar to that of the upper-right partition [T1][B][T2] 1, but with 21 and 22 interchanged. The result is mF21d ' 8 cos " cos ( " 120E) Thus mF21d ' mF12d
(29)

< cos ( "

120E) .

(28)

and the mutual (off-diagonal) inductances in the dq0 matrix are indeed reciprocal, as would be expected. In order to proceed with manageable calculations, the simplest expedient is to drop mF12d from eqn. (7) and mF12q from its q-axis counterpart, leaving Md1d2 ' Lmd and Mq1q2 ' Lmq .
(31) (30)

(26)

This is, in effect, one term of a Fourier series expansion of the inductances 8, and <. If we substitute this in (23) we get the total nth harmonic inductance mn(") ' 7n [ cos n " cos " cos n ( " cos n ( " 120E ) cos ( " 120E ) cos ( " 120E ) 120E ) ] .
(27)

It can be shown that mn(2") = mn(") for " ' 20E, 40E, 60E, 80E, 100E, 120E etc., but not for " = 15E or 30E.

Alternatively, mF12d and mF12q can be retained in the equations but treated as a user-defined perturbation parameter. This permits an experimental numerical approach to determine how large these parasitic mutual inductances must be to cause significant errors in the calculation (and/or to cause significant imbalance between winding sets in cases of multiplicity 3 or higher). This approach is followed in the remaining sections.

IV. TORQUE The electromagnetic torque is the result of interaction between the currents and the rotational voltages in (2) and (5). For the first set of windings, T1 ' m p ( Rd1iq1 Rq1 id1 ) ,
(32)

Tei ' Tr A '

Tr B '

mp Eq1 Iq1 T mp [ (Xd1 T (Xd3 mp [ ( Xd1d2 T ( Xd2d3

Eq2 Iq2 Xq1) Id1Iq1

Eq3 Iq3 ; (Xd2 Xq2) Id2Iq2 (38) Id2 Iq1 ) Id3 Iq2 ) Id1 Iq3 )] .

Xq3) Id3Iq3 ] ; Xq1q2 ) ( Id1 Iq2 Xq2q3) ( Id2Iq3

where m is the number of phases (3) in the set, and p is the number of pole-pairs. Substituting for the flux-linkages from (3), T1 ' m p [ QMd1 iq1 ( Ld1 Lq1 ) id1 iq1 Mq1q2id1iq2 ] .
(33)

( Xd3d1 Xq3q1) ( Id3Iq1

Md1d2 id2iq1

For the second set of windings the process is the same: T2 ' m p ( Rd2iq2 so that T2 ' m p [ QMd2 iq2 ( Ld2 Lq2 ) id2 iq2 Mq2q1id2iq1 ] .
(35)

Rq2 id2 ) ,

(34)

Md2d1 id1iq2

In each of (33) and (35), the first term is the familiar permanent-magnet alignment torque, while the second term is the familiar reluctance torque. The third and fourth terms arise from interaction between the two sets of windings: daxis flux produced by one set interacts with the q-axis current of the other set. Consequently four additional terms appear for a duplex winding; eight for a triplex winding, and 4(x 1) for an x-plex winding. The torque equations (33) and (35) can be used for transient or steady-state torque. In the steady-state they can be written in terms of phasors (RMS AC quantities): mp T1 ' [ Eq1 Iq1 (Xd1 Xq1) Id1Iq1 T Xd1d2 Id2 Iq1 Xq1q2Id1Iq2 ] ;
(36)

Fig. 4 Phasor diagram for duplex 3-phase machine

V. STEADY-STATE PHASOR DIAGRAM Under AC steady-state conditions the RMS values of the d- and q-axis flux-linkages Rd and Rq in (3) can be combined into a phasor Q1 ' Qd1 j Qq1 ,
(39)

and likewise the current can be expressed as a phasor I1 ' Id1 j Iq1 .
(40)

The voltage phasor is then given by V1 ' Vd1 in which Vd1 ' R1Id1 Vq1 ' Eq1 Xq1Iq1 R1Iq1 Xq1q2Iq2 ; Xd1Id1 Xd1d2Id2 .
(42)

T2 '

mp [ Eq2 Iq2 (Xd2 Xq2) Id2Iq2 T Xd2d1 Id1 Iq2 Xq2q1Id2Iq1 ] .

j Vq1 ' R1 I1

j T Q1 ,

(41)

where T = 2Bf and X = TL. Regrouping the terms of this equation, we can separate the alignment torques and the reluctance torques as
Te i ' Tr A Tr B mp Eq1 Iq1 Eq2 Iq2 ; T mp ' (Xd1 Xq1) Id1Iq1 (Xd2 Xq2) Id2Iq2 ; T mp ' ( Xd2d1 Xq1q2 ) Id1 Iq2 ( Xd1d2 Xq2q1) Id2Iq1 T

(37)

where Tei is the total alignment torque, TrA is the sum of the self reluctance torques of the two sets operating individually, and TrB is the sum of the mutual reluctance torques that arise from mutual coupling between the two sets of phases. With a triplex winding the result is given by (36):

This can be expressed graphically in the phasor diagram, Fig. 4. The terms R1Id1, Xq1Iq1, Eq1, R1Iq1, and Xd1Id1 are familiar from the simplex winding; but the cross-coupling terms Xq1q2Iq2 and Xd1d2Id2 are peculiar to the duplex winding. Without the cross-coupling terms the terminal voltage would be U1 (Fig. 4); but with them the terminal voltage is V1. A similar phasor diagram is obtained for the second set of windings, for which the corresponding voltage equations are eqns. (43). Vd2 ' R2Id2 Vq2 ' Eq2 Xq2Iq2 R2Iq2 Xq2q1Iq1 ; Xd2Id2 Xd2d1Id1 .
(43)

The cross-coupling terms appear in the phasor diagram as additional voltage-drops which tend to limit the current. If " = 0, we have tightly coupled inductances between the two sets, as already observed; and if these sets are fed from a common voltage source the current in each set will be approximately half the current that would flow in one set if the other were open-circuited. This is an important practical point because it implies that in a duplex winding, if one set is open-circuited the current in the other set could increase by a factor approaching 200%, if it were not regulated. Likewise if one set is short-circuited, the impedance of the second set will be reduced and its current could also increase to a high value if it is not regulated.

VI. CONCLUSION The main results of this paper are the phasor diagram of the multiplex permanent-magnet synchronous machine, and the torque equation. Both these tools are highly useable in design work in much the same way as the simpler versions that are familiar with conventional simplex machines, and both of them provide clear physical intepretation of the interaction between the sets or channels. The dq-axis model has additional mutual reactances between the sets, and these can be calculated in the first instance from the full inductance matrix transformed into dq axes. It is shown that the sets of a symmetrical duplex winding are inherently balanced; but with triplex windings and windings of higher multiplicity the displacement angle between sets is critical, and an expression is given for it. Although the multiplex or x-plex machine appears suited to the use of multiple inverters, each of 1/xth the full rating, it is not inherently fault-tolerant because the sets are always magnetically coupled. Therefore, any operation with unbalanced loads between sets must be controlled. Further work on this topic would be useful. The analysis does lead to a practical set of equations for solution, in spite of the lengthy mathematics of their derivation. Future work might well include a more detailed analysis of the mutual reactances between sets (preferably not by finite-element grinding). ACKNOWLEDGMENT

Fig. 5 Parallel inductances (aiding)

The behaviour of the duplex sets is analogous to that of parallel inductances, Fig. 5, with an equivalent inductance L1 L2 L1 L2 M2 2M .
(44)

If L1 = L2 = L, this simplifies to (L + M)/2, and the equivalent circuit is two parallel uncoupled inductances, each of value L + M. When " = 0, M becomes close to L and the equivalent inductance is approximately (L + M)/2 = L. The total current is that which is limited by L, and half the current flows in each set. But if one set is open-circuited, the same total current will tend to flow in one set. The implication is that regulation of the current is essential. We are now in a position to solve the system with any given voltages or currents applied to the terminals of the two sets of windings. In the steady state this is a question of solving eqns. (42) and (43) when either the voltages V1 and V2 are given, or the currents I1 and I2 are given. When the machine is inverter-fed, the applied voltages are PWM waveforms determined by a current regulator. Where we normally solve only (2) for Rd and Rq in a 3phase machine, we must now integrate (2) and (5) to produce updated values of Rd1, Rq1, Rd2 and Rq2 at each time-step, also inverting (4) and (6) for id1, iq1, id2 and iq2 and updating the torque. The extension to triplex and multiplex windings is a straightforward matter of adding additional voltage equations of the form of (2), and additional mutual terms in the corresponding flux-linkage equations of the form of (4). Everything else follows as described above.

The authors thank Dr. Mircea Popescu of Motor Design Ltd. (formerly of the SPEED Laboratory), and H. C. Karmaker of TECO Westinghouse and for their assistance. REFERENCES [1] SPEED's Electric Motors, SPEED Laboratory, University of Glasgow, 2008, pp. 1.66-1.69. [2] J.R. Hendershot and T.J.E. Miller, Design of Brushless Permanent-magnet Machines, to be published, 2010. [3] A.E. Fitzgerald and C. Kingsley, Electric Machinery, McGraw-Hill (2nd edition), 1961. [4] L. Parsa, On advantages of multi-phase machines, Industrial Electronics Society, 2005. IECON 2005. 31st Annual Conference of IEEE, pp. 6-10, Nov. 2005 [5] L. Parsa and H.A. Toliyat, Multi-phase permanent magnet motor drives, Industry Applications Conference, 2003. 38th IAS Annual Meeting, Conference Record, vol.1, pp. 401-408, Oct. 2003 [6] T.A. Lipo, A d-q model for six phase induction machines, ICEM' 80, Athens, Greece, September 1980, pp. 860 867 [7] R.H. Nelson and P.C. Krause, Induction machine analysis for arbitrary displacement between multiple windings, IEEE Trans., Vol. PAS-93, pp. 841 848, May 1974