A Study of Brushless DFIMs

A Study of Brushless Doubly-Fed (Induction) Machines
C ONTRIBUTIONS IN M ACHINE A NALYSIS , D ESIGN AND C ONTROL
DISSERTATION SUBMITTED FOR THE DEGREE OF
D OCTOR
OF
P HILOSOPHY
Paul C. Roberts
September 2004
Emmanuel College University of Cambridge
(revised January 2005)
To God, Ruth and my parents.
I will lift my eyes to the mountains; From where shall my help come? My help comes from the LORD, Who made heaven and earth. Psalm 121
Abstract
The Brushless Doubly-Fed Machine (BDFM) shows commercial promise as a variable speed drive or generator. However, for this promise to be realised the design of the machine must be improved beyond that proposed to date. This dissertation contributes towards this goal through machine analysis, design and control. A generalised framework is developed for a coherent and rigorous derivation of models for a wide class of BDFMs, of which machines with nested-loop design rotors are a subset. This framework is used to derive coupled circuit, d-q axis, sequence components and then equivalent circuit models for the class of machines. Proofs are given for all derivations, exploiting the circulant properties of the mutual inductance matrices. The coherence between the different models allows parameters calculated for the coupled circuit model to provide parameter values for the other models. A method of model order reduction is proposed for the class of BDFMs with nested-loop rotors, and examples given of the efcacy of the procedure. The reduction method allows parameter values to be computed for a simple equivalent circuit representation of the machine. These calculated parameter values, and those for other BDFM rotor designs are veried by experimental tests on a prototype BDFM. The signicance of particular equivalent circuit parameters is investigated from the model. Series rotor inductance terms are found to have a signicant and direct effect on machine performance. These terms are shown to relate directly to the design of the rotor, and are quantied using the previously developed framework. Seven different rotor designs, including a new BDFM rotor design, are considered to show how the values of these parameters change. An experimental method of parameter estimation is developed for the equivalent circuit model, and the relationship between these parameters and the parameters in other forms of the model derived. The experimental method is shown to be applicable both to standard induction machines and to BDFM machines, yielding accurate results in each case. A synchronous reference frame model for the class of BDFMs is derived and is used to analyse the stability of the machine via a linearized model. Practical methods for the design of PID controllers are proposed to stabilise the machine using voltage source inverters. Results are presented from experimental implementations which show a signicant improvement in performance over previously published results. The non-linear control technique, feedback linearization, is applied to the BDFM and shown to have some robustness to modelling errors, in a realistic simulation. An initial attempt at implementation of the scheme is reported. Preliminary results are encouraging, and warrant further investigation. Keywords: ac machines, BDFM, Brushless Doubly Fed Machines, control, coupled circuits, dq axis, equivalent circuits, feedback linearization, model reduction, parameter estimation, synchronous reference frame
i
Acknowledgements
First, and foremost, I would like to express my gratitude to my supervisors doctors Richard McMahon, Jan Maciejowski and Tim Flack whose insight and enthusiasm have greatly improved this dissertation. I would like to thank colleagues past and present for making work much more enjoyable. Some pleasant memories include, the tea-time Bridge sessions, the Poker evenings, lunch with Dan and hitch-hiking to and from an international conference. I would like to gratefully acknowledge the help of Ehsan Abdi-Jalebi and Xiaoyan Wang in the operation of the BDFM test rig. In particular their assistance in the collection of the experimental data used in gures 5.8, 5.9, 5.10, 5.11, 6.3 and 6.4. Thanks are also due to Ehsan for proof reading this dissertation. I am most grateful for assistance of Mark Barrett, John Grundy, the electricians, carpenters, welders and other support staff in preparing and setting up the prototype BDFM test rig. Thanks are also due to Davor Dukic, Iskandar Samad and George Makrides for their assistance in the development of the instrumentation hardware for the test rig. I am very grateful for the assistance of computing support staff, particularly John Sloan and Patrick Gosling. This work has been funded by the Engineering and Physical Sciences Research Council. A highpowered inverter was generously donated by Semikron UK Ltd. Cambridge Silicon Radio (CSR) Ltd. kindly supplied a number of Bluetooth models. FKI Energy Technology particularly its subsidiaries Laurence, Scott & Electromotors (LSE) Ltd. and Marelli Motori SpA provided the prototype machine frame and manufactured the prototype rotors described in chapter 5. I would like to thank these institutions for their support. In particular, thanks to Peter Tavner, formerly of FKI and Terry Gallant of LSE for their continued interest in the BDFM, and to David Hargreaves of CSR. I would like to thank my parents for a lifetime of support and for continually fuelling my enthusiasm for tinkering with anything I could get my hands on. Finally, I would like to thank my wife, Ruth, for her unagging support throughout, and seemingly endless patience with me.
iii
iv
Acknowledgements
As required by the University Statute, I hereby declare that this dissertation is not substantially the same as any that I have submitted for a degree or diploma or other qualication at any other university. This dissertation is the result of my own work and includes nothing which is the outcome of work done in collaboration, except where specically indicated. This dissertation is 64,277 words in length, and contains 83 gures. Paul Roberts Emmanuel College September 2004
Contents
Abstract Acknowledgements Notation & Terminology 1 Introduction 1.1 1.2 Evolution of the BDFM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Description of the BDFM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 1.2.2 1.2.3 1.3 1.3.1 1.3.2 1.3.3 Machine Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Synchronous mode of operation . . . . . . . . . . . . . . . . . . . . . . . . Potential Applications for the BDFM . . . . . . . . . . . . . . . . . . . . . BDFM Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . Use of BDFM models to investigate performance, new rotor designs and the estimation of machine parameters . . . . . . . . . . . . . . . . . . . . . . . Analysis of stability of the BDFM and the development of control strategies . 12 13 15 15 16 18 19 20 21 24 25 27 28 28
v
i iii xi 1 1 7 7 8 10 10 11
Approach of this work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 BDFM Coupled Circuit Modelling and Parameter Calculation 2.1 2.2 2.3 2.4 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . General Electrical Machine Coupled-Circuit Model . . . . . . . . . . . . . . . . . . 2.3.1 2.4.1 2.4.2 2.4.3 2.5 2.5.1 2.5.2 Torque Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magnetic Flux Density due to Single Coil . . . . . . . . . . . . . . . . . . . Mutual (and self) Inductance of Single Coils . . . . . . . . . . . . . . . . . Calculation of spatial harmonic components of mutual inductance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Calculation of rotor and stator inductance matrices . . . . . . . . . . . . . . Leakage inductance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Calculation of Parameters for Electrical machines . . . . . . . . . . . . . . . . . . .
Mutual Inductance of Machine Windings
vi
CONTENTS
2.5.3 2.6 2.6.1 2.6.2 2.7
Resistance calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mutual inductance between two stator windings . . . . . . . . . . . . . . . . Mutual inductance between two stator windings where coils groups are not in series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29 29 33 36 39 40 42 43 44 47 49 49 50 54 64 65 66 75 78 83 87 87 88 95 98
Some properties of machine windings . . . . . . . . . . . . . . . . . . . . . . . . .
BDFM Rotor Mutual Inductance terms . . . . . . . . . . . . . . . . . . . . . . . . . 2.7.1 2.7.2 Rotor-rotor mutual inductance matrices . . . . . . . . . . . . . . . . . . . . Rotor-Stator mutual inductance matrix . . . . . . . . . . . . . . . . . . . . .
2.8 2.9
Effect of Slotting on Mutual inductance terms . . . . . . . . . . . . . . . . . . . . . BDFM Coupled-Circuit Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 d-q Transformed Model 3.1 3.2 3.3 3.4 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The d-q-0 state transformation matrix . . . . . . . . . . . . . . . . . . . . . . . . . Transformation to d-q-0 axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 3.4.1 3.4.2 3.5 3.6 Determination of d-q model rotor current from bar currents . . . . . . . . . . Model Reduction Techniques . . . . . . . . . . . . . . . . . . . . . . . . . Model order reduction for Nested-loop rotor . . . . . . . . . . . . . . . . . . . . . . New BDFM Rotor State Reduction Technique . . . . . . . . . . . . . . . . .
Simulation comparison of different BDFM model reduction techniques . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 Equivalent Circuit Model and its Implication for BDFM Performance 4.1 4.2 4.3 4.4 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conversion to Symmetrical Components . . . . . . . . . . . . . . . . . . . . . . . . Steady-state Equivalent Circuit Representation . . . . . . . . . . . . . . . . . . . . . Equivalent circuit for the BDFM with a single set of rotor coils . . . . . . . . . . . . 4.4.1 4.4.2 4.4.3 4.5 4.6 4.7 4.8
Physical interpretation of parameters in the per-phase equivalent circuit model 100 Development of Torque Equations from the Equivalent Circuit . . . . . . . . 104 Phasor diagram for rotor branch circuit . . . . . . . . . . . . . . . . . . . . 108
Equivalent Circuit Numerical Simulation . . . . . . . . . . . . . . . . . . . . . . . . 108 Equivalent Circuit Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Magnetic Loading for the BDFM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 119
5 Possible Rotor Designs and Evaluation 5.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
CONTENTS
vii
5.2
Rotor Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 5.2.1 5.2.2 5.2.3 5.2.4 5.2.5 5.2.6 5.2.7 Rotor 1 - the nested-loop design rotor . . . . . . . . . . . . . . . . . . . . 121 Rotor 2 - the new double layer design rotor . . . . . . . . . . . . . . . . . . 122 Rotor 3 - isolated loop rotor . . . . . . . . . . . . . . . . . . . . . . . . . . 124 Rotor 4 - isolated loop rotor with coils removed . . . . . . . . . . . . . . . . 126 Rotor 5 - 6 bar squirrel cage rotor design . . . . . . . . . . . . . . . . . . . 127 Rotor 6 - wound rotor design . . . . . . . . . . . . . . . . . . . . . . . . . . 127 Rotor 7 - standard squirrel cage rotor . . . . . . . . . . . . . . . . . . . . . 128
5.3 5.4 5.5
Experimental and Calculated Torque-Speed Curve Results . . . . . . . . . . . . . . 129 Calculated Rotor Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 5.5.1 5.5.2 Experimental Torque-Speed Curves . . . . . . . . . . . . . . . . . . . . . . 132 Calculated machine parameters from tables 5.1 and 5.2 . . . . . . . . . . . . 134
5.6
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 137
6 BDFM Parameter Identication 6.1 6.2
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Parameter Extraction Optimization Method . . . . . . . . . . . . . . . . . . . . . . 138 6.2.1 6.2.2 General Optimization Problem . . . . . . . . . . . . . . . . . . . . . . . . . 138 Application to BDFM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 Comparison of data obtained using estimated parameter values with experimental data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 6.3.2 6.3.3 Comparison of estimated to calculated parameter values . . . . . . . . . . . 147 Comparison of estimated parameter values to manufacturers parameter values 149
6.3
Parameter Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 6.3.1
6.4 6.5
Relationship of extracted parameters to the d-q axis model . . . . . . . . . . . . . . 151 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 155
7 Modelling for Control of the BDFM 7.1 7.2
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 Synchronous reference frame model . . . . . . . . . . . . . . . . . . . . . . . . . . 156 7.2.1 7.2.2 Transformation from the rotor reference frame to the synchronous reference frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 Evaluation of component matrices in the synchronous reference frame . . . . 160
7.3 7.4
Synchronous Reference Frame Model Equilibrium Conditions . . . . . . . . . . . . 163 Linearization of the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 7.4.1 7.4.2 Conclusions for the Linearized Model . . . . . . . . . . . . . . . . . . . . . 168 Simplication of Linearized Model . . . . . . . . . . . . . . . . . . . . . . 168
viii
CONTENTS
7.4.3 7.5 7.6 7.7 7.8 7.9
Linearization Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Simulated Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Open-loop Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Closed-loop stator 2 phase angle control . . . . . . . . . . . . . . . . . . . . . . . 178 7.7.1 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 2 dt = ( p1 + p2 )r 1 dt . . . . . . . . . . . . . . . . . . . . 180 Control when
Future work on linear model based control . . . . . . . . . . . . . . . . . . . . . . . 181
7.10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 8 Feedback Linearization for the BDFM 8.1 8.2 8.1.1 8.2.1 8.2.2 8.2.3 8.3 8.3.1 8.3.2 8.4 187
Feedback Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188 BDFM model in terms of ux linkages . . . . . . . . . . . . . . . . . . . . 191 Control strategy 1: Speed Only Regulation . . . . . . . . . . . . . . . . . . 193 Control Strategy 2: Speed and Flux Regulation . . . . . . . . . . . . . . . . 194 Zero Dynamics and Idealized FBL Stability . . . . . . . . . . . . . . . . . . 196 Practical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Application to the BDFM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Towards A Practical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 205
9 Conclusions and Future Work 9.1 9.2
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 9.2.1 9.2.2 9.2.3 Analysis of the BDFM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 BDFM Machine Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 Stability analysis and control, including parameter estimation . . . . . . . . 210 213
A Mathematics
A.1 Trigonometric Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 A.2 Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 B Prototype Machine Stator and Rotor Design Details 229
B.1 Prototype machine frame details . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230 B.2 Prototype Machine Stator Windings . . . . . . . . . . . . . . . . . . . . . . . . . . 230 B.2.1 B.2.2 B.3.1 Machine Winding Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . 230 Stator Winding details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Rotor-rotor inductance terms . . . . . . . . . . . . . . . . . . . . . . . . . . 237
B.3 Rotor 1: Nested-loop Rotor Design Details . . . . . . . . . . . . . . . . . . . . . . . 237
CONTENTS
ix
B.3.2 B.4.1 B.4.2 B.5.1 B.6.1 B.7.1 B.7.2 B.8.1 B.8.2 B.9.1
Rotor-Stator inductance details . . . . . . . . . . . . . . . . . . . . . . . . . 239 Rotor-rotor inductance terms . . . . . . . . . . . . . . . . . . . . . . . . . . 239 Rotor-Stator inductance details . . . . . . . . . . . . . . . . . . . . . . . . . 241 Rotor inductance terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 Rotor inductance terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 Rotor-rotor inductance terms . . . . . . . . . . . . . . . . . . . . . . . . . . 247 Rotor-Stator inductance details . . . . . . . . . . . . . . . . . . . . . . . . . 248 Rotor-rotor inductance terms . . . . . . . . . . . . . . . . . . . . . . . . . . 249 Rotor-Stator inductance details . . . . . . . . . . . . . . . . . . . . . . . . . 250 Rotor-rotor inductance terms . . . . . . . . . . . . . . . . . . . . . . . . . . 251
B.4 Rotor 2: New Double Layer Rotor Design Details . . . . . . . . . . . . . . . . . . . 239
B.5 Rotor 3: Isolated loop rotor design . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 B.6 Rotor 4: Isolated loop design rotor with one set of loops removed . . . . . . . . . . . 245 B.7 Rotor 5: 6 bar cage rotor design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
B.8 Rotor 6: Wound Rotor Design Details . . . . . . . . . . . . . . . . . . . . . . . . . 249
B.9 Rotor 7: Standard Squirrel Cage Rotor Details . . . . . . . . . . . . . . . . . . . . . 251 B.10 Machine slot utilisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 C Leakage Inductance and Effective Air Gap C.1.1 C.1.2 C.1.3 C.1.4 257
C.1 Leakage Inductance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Slot and Tooth-top Permeance . . . . . . . . . . . . . . . . . . . . . . . . . 257 Overhang Permeance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258 Zig-zag Permeance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259 Leakage ux per coil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
C.2 Effective air gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262 D Previous BDFMs E Experimental Apparatus E.1.1 E.1.2 E.1.3 E.1.4 E.1.5 E.1.6 E.1.7 E.1.8 267 269
E.1 Apparatus Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 xPC Target PC and peripheral boards . . . . . . . . . . . . . . . . . . . . . 270 Torque Transducer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 DC load motor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 Inverter Output Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 Inverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 Position and Speed Measurements . . . . . . . . . . . . . . . . . . . . . . . 273 Analogue to Digital Converters . . . . . . . . . . . . . . . . . . . . . . . . 274 DC Machine load resistors and Inverter dump resistors . . . . . . . . . . . . 274
CONTENTS
Bibliography Author Index
287 289
Notation & Terminology

Notation
Numbers
R C Z Z N j X {X } , {X } eld of real numbers eld of complex numbers set of integers set of non-negative integers natural numbers (positive integers excluding zero) the imaginary unit, i.e. 1 denote the complex conjugate of X C of which are in the set or eld F. denotes real, imaginary part of X C
Fnm
denotes that a set comprised of matrices of dimension n m, the elements
xi
xii
NOTATION & TERMINOLOGY
Vectors
xy
x
scalar (dot) product of x and y Given a vector function, f Rn , which is a function of the vector x Rm , and possibly other vector or scalar variables, then
x
2-norm of a vector or vector signal. For x Rn , x
f1 x 1 f = f 2 x 1 . . .
f1 x2 f2 x2 . . .
, which is known as the Jacobian of f when n = m. .. .

2 n i=1
xi2 . Def-
inition 3.1 denes the 2-norm of a continuous function x, which shares the
same notation, however the reader should assume the standard denition unless otherwise indicated. L2 P
i,2
The space of square integrable Lebesgue measurable functions, see denition 3.1. The induced 2-norm of a system P, see denition 3.2
Matrices
X 1 , X X T, X Xxy X indicating inverse, pseudo-inverse of matrix X indicating transpose, complex-conjugate transpose of matrix X element of matrix X at x th row, and y th column estimated value of X zero matrix of compatible dimension the identity matrix of compatible dimension
0 I
In X
the n n identity matrix (also used to denote complex current)
X is a matrix where the rows (or columns) of X span the orthogonal comple-
ment of the subspace dened by the span of the rows (or columns) of X . Whether row space or column space is intended will be determined by the context.
xiii
Mathematical terminology
B 1 . I x . , a hypercube around the origin. x: . I 1
proportional to
the empty (null) set and set B
intersection, e.g. A B denotes the set dened by the intersection of set A dened as
there exists for all such that in, for example q R : q 2 = q., reads there exists a q in the eld of real numbers, such that q 2 = q. end of proof.
end of remark. closed interval from 0 to 10, that is a range from 0 to 10 including both 0 and 10. open interval from 0 to 10, as above but excluding 0 and 10. Hence [0, ) implies, from left to right, e.g. A B reads A implies B. is the interval from 0 to excluding , but including 0.
[0, 10]
(0, 10)
iff
implies, from right to left, e.g. A B reads B implies A. implies A and A implies B.
implies, from right to left and from left to right, e.g. A B reads B if and only if much greater than, for example a K > 0. the dual of . b a > b + K for some (large)
xiv
Electrical Machine Notation

X s1 , X s2 , X r Xd , Xq , X0 X
e
indicating a stator 1, 2, or rotor quantity X indicating direct (or d) axis component, quadrature (or q) axis component, zero (or 0) sequence X indicating an equilibrium value of quantity X indicating the value of quantity X in the synchronous reference frame resistance, mutual inductance, self inductance, impedance an inductance-like term arising from the movement between stator and rotor indicates a coupled-coil inductance parameter indicates an apparent (referred) quantity indicates an equivalent circuit inductance parameter instantaneous voltage, current, ux linked ( is also used to denote eigenvalues). stator-rotor phase offset, or magnetic ux. stator 1, stator 2 winding number of pole pairs stator 1, stator 2 supply frequencies BDFM rotational shaft speed frequency of currents in the rotor reference frame in BDFM synchronous mode see (4.26). BDFM angular position stator 1 and stator 2 slip, as dened in equations (4.32) and (4.33), not to be confused with s which denotes the complex variable of the Laplace transform.
X s or X sync R, M, L , Z Q X cc X Xe v, i, p1 , p 2 1 , 2 r s r s1 , s 2
J Te , Tl d w c s ys 0 B g H
current density, or moment of inertia electrical torque, load torque machine diameter, as shown in gure 2.1 machine stack length, as shown in gure 2.1 coil pitch, in radians (see gure 2.2) slot pitch, in radians (see gure 2.2) slot pitch, in metres magnetic ux density permeability of free space, 0 = 4 107 H/m
air gap width (see gure 2.2) magnetic eld intensity, see denition 2.1
xv
Terminology
absolute harmonic A/D balanced three phase supply cascade induction mode D/A FBL FFT I/O inverter see remark 2.6 analogue to digital converter three voltage or current sources supplying K cos(t ), operation of the BDFM when the second stator is short circuited, see section 1.2.1. digital to analogue converter feedback linearization, see chapter 8. fast Fourier transform, a fast implementation of the discrete Fourier transform input / output a device which converts DC to AC, i.e. performs in inverse operation of a rectier. In this dissertation the word always refers to an electronic device. lamination stack length length of the stack of laminations (typically made from electrical steel) used to make a machine stator or rotor. Figure 2.1 shows this dimension on a rotor. LHP LMI LPV LTI Matlab left half-plane, the dual of RHP. linear matrix inequality linear, parameter-varying linear, time-invariant Matrix-Laboratory: computer software for numerical and symbolic computations from the Mathworks, Inc.; essentially a high level programming language design for mathematical computation. MIMO mmf orthonormal completion ODE orthogonal matrix PI PID relative harmonic multiple input, multiple output magneto-motive force, see denition 2.2 a set of mutually orthonormal vectors which span the orthogonal complement of a subspace. ordinary differential equation. a unitary matrix with X Rnn , hence its columns form an orthonormal basis for Rn . Proportional-Integral, a controller with transfer function K p + tion K p +
Ki s Ki s
K cos(t 2/3), and K cos(t 4/3) for some K , , .
Proportional-Integral-Derivative, a controller with transfer funcsee remark 2.6 + K d s.
xvi
RHP
right half-plane - refers to the position of (typically poles or zeros) in the complex plane, RHP means that complex number in question. {X } > 0 where X is the
rms simple induction mode Simulink SISO stack length unitary matrix
root mean square - the square-root of the mean of the square of a quantity. operation of the BDFM when the second stator is open circuit, see section 1.2.1. A modelling environment within Matlab, allowing easy solution of ODEs. single input, single output abbreviation of lamination stack length. that the columns of X form an orthonormal basis for Cn . X Cnn is a unitary matrix if X X = X X = I , which means
Chapter 1
Introduction
This dissertation is concerned with the analysis, design and control of the Brushless Doubly-Fed Machine (BDFM). The BDFM, sometimes referred to as the Brushless Doubly-Fed Induction Machine, is an AC electrical machine which can operate as both a motor and a generator. As the name implies the machine requires two AC supplies, and has no direct electrical connection to the rotor, which removes the need for carbon brushes sometimes found in electrical machines. This chapter reviews the development of the BDFM, then section 1.2 gives a basic description of the machine operation, and explains the modern interest in the machine. Readers unfamiliar with the machine may prefer to read this section rst. We then outline the approach and structure of this dissertation in section 1.3
1.1 Evolution of the BDFM

The history of the modern BDFM can be originally traced back to a patent taken out by the Siemens Brothers and Mr. F. Lydall in 1902 for a self-cascaded machine [97]. At the time it had become clear that the supply distribution standard was changing from DC to AC. The adoption of AC brought the induction machine into industrial service. The induction machine offered a simple and robust construction, which was desirable, however industry required variable speed operation which led to a signicant problem as the fundamental operating speed for an induction machine is xed by the mains frequency. At the turn of the 20 th century the common method of controlling the speed of induction machines was the introduction of series resistance to the rotor, connected via slip rings, as the modern luxury of power electronic frequency converters (inverters) had not even been conceived. The use of series resistance (rheostats) connected to the rotor is inherently lossy, and thus there was a strong desire to design machines which could be operated at different speeds without rheostatic loss [97].
1
Introduction
It was well known at the time that connecting two induction machines together allowed one to achieve three different speeds of efcient operation [48, p. 648]. Such an arrangement was known as cascaded induction machines. The Lydall patent covered the incorporation of two induction machines with air gap elds of different pole numbers in one frame. Lydall realised that the resulting machine would mimic the behaviour of two cascaded (but physically separate) machines because unlike pole number elds do not couple. His machine had two stator windings and two rotor windings each brought out on slip rings. Three different efcient speeds of operation could be achieved, the synchronous speed of winding 1, / p1 , the synchronous speed of winding 2, / p2 , and the cascade synchronous speed, putting two machines in the same frame. At the same time as Lydalls work, others were attempting to create a self-cascaded machine, but their attempts used suitable spacing of the stator windings to ensure non-linking of the stator elds, rather than different pole number elds. Prof. Silvanus P. Thompson took out a patent in 1901 [49, p.406] for such a machine, Steinmetz in the USA for a similar machine in 1903, and Meller in Germany in 1904 [48, p.651]. In 1907 Hunt [48] further developed Lydalls original idea, although it appears he was initially unaware of Lydalls work [49, p. 407]. Hunt realised that the slip rings were unnecessary, that is, if the cascade connection is made rotor to rotor, rather than rotor to stator, then no slip rings are needed, and three speeds of operation are still possible. Hunt also contributed to the development of the machine and showed that with suitably designed stator and rotor windings it was possible to reduce copper losses signicantly. The single rotor winding was designed to couple to both air gap elds and needed signicantly less cross-sectional area of copper than Lydalls design, leading to smaller slots and a more practical rotor design. Hunt developed the machine from 1907 to 1914, and he describes some renements in his 1914 paper along with experiences gained from industrial applications, principally in the mining industry [49]. Further improvements were made to the rotor and stator designs by Creedy in 1920 [27]. Creedy also rened some of Hunts rules for choosing suitable pole number combinations, thus providing greater choice of machine congurations. Most signicantly Creedys rules for choosing pole numbers led to the possibility of using a machine with 6 and 2 poles, which was 50% faster than the fastest machine available using Hunts rules. However, in fact, Creedys rules for choosing pole number combinations were still incomplete. The self-cascaded machine met with some commercial success. Hunt reports a number of machines which found application in areas including an air compressor in a colliery, and as an air compressor for refrigeration [49, p. 426]. However, after Creedys paper no further publications appeared until Smith published in 1966 [102]. Smiths machine was in the category of the Thompson/Steinmetz/Muller machines, that is, the non-linking of stator elds was achieved by spatial separation rather than different pole number
. p1 + p 2
Lydalls work was to achieve self-cascaded operation by, in effect,
elds. Smiths initial motivation appears to have been for the cascade mode of operation, and he presents, for the rst time, a model of the steady-state operation of the machine. The following year Smith published on the synchronous performance of his machine, and noted that a lower power frequency converter could be used to extract slip power from the rotor by induction, or if supplied from the second winding, the machine would behave as a synchronous machine [103]. This was a signicant development as it was the rst time that the full synchronous mode was noted, however the Smith machine was not a true BDFM, rather two magnetically separate machines in the same frame. Furthermore, in his 1967 paper Smith presents an equivalent circuit for his machine, and analyses its performance in the synchronous mode of operation. Cook and Smith went on to publish on stability of the synchronous mode of operation, [24, 25], however, as mentioned, his work concerns essentially two separate induction machines in cascade, which happen to be put in the same frame. Nevertheless Smith must be credited with rst noting the synchronous mode of operation which is achieved by a double feed. In 1970 an unquestionably signicant contribution was made by Broadway and Burbridge [17]. Indeed, physically, the modern BDFM is essentially the same as that proposed by Broadway and Burbridge. It seems that they were unaware of the work of Smith, however. Broadway and Burbridge returned to the Hunt machine, and made signicant contributions to the design of the rotor. They reasoned that a wound rotor, which the Hunt and Creedy designs required, would lead to higher losses and lower durability. They sought to design a cage-type rotor which could be made by the same means as a squirrel cage rotor, yet did not suffer from excessive leakage reactance. They have erroneously been credited with rst realising that a cage rotor for a BDFM must have the number of bars given by the sum of the pole pairs of the stator windings; in fact Hunt noticed this [27, p. 534]. Nevertheless Broadway and Burbridge were the rst to formalise this concept. Their main contribution was to propose the nested-loop design rotor, which has been adopted in most subsequent work on the BDFM. Broadway and Burbridge presented an equivalent circuit for the BDFM, and some performance aspects were analysed in steady-state. They also noted that the second winding can, not only extract slip power, but also be used to run the machine in a synchronous mode. Yet unlike Smith, they only consider the synchronous mode in the case that the second stator supplies DC. Broadway published again in 1971 on what is, in essence, the modern Brushless Doubly-Fed Reluctance Machine (BDFRM) [15]. In this paper it is noted that doubly-fed operation is possible with different supply frequencies, through the use of a variable frequency inverter. In 1974 a further publication by Broadway appeared, with a valuable discussion of the effects of two pole number elds sharing the same iron circuit, and the effects of saturation [16]. Four years later Kusko and Somuah [57] presented work on a BDFM with two 3 phase windings on the rotor, although they conceded that the Broadway/Burbridge design would have been an improvement. They are, perhaps, the rst to have noticed that the BDFM equivalent circuit has similarities with that of the synchronous machine in synchronous operation, however they were principally
Introduction
concerned with slip power recovery rather than true double feed - the latter requiring a fully controlled converter. In 1983 and 1987 Shibata and Kohrin and Shibata and Taka published on the Broadway machine, rstly in the cascade mode of operation and then in the synchronous doubly-fed mode [94, 95]. In the mid-1980s as part of US Department of Energy contracts, interest developed in the BDFM at Oregon State Univserity [65]. Wallace, Spe, Li and others at Oregon studied BDFMs extensively, and indeed the name BDFM originated in these publications. Their work used the Broadway/Burbridge rotor, and developed a coupled-circuit dynamic model for their prototype machine. They used this model to investigate performance [110, 104]. They then reconsidered the rotor design, and presented some analysis and proposals for renement [109]. This work led towards a patent led by Lauw in 1993 [59]. Hunt, Creedy and Broadway/Burbridge all proposed a single stator winding which allowed both pole number elds to be achieved without the need for separate stator windings. However Rochelle et. al investigated this winding conguration and concluded that it can lead to circulating currents [88]. Therefore, later work at Oregon used separate windings. Li et. al. proposed a d-q (two axis) dynamic model for their prototype machine, and presented performance results [64, 63]. In these works the authors use separate three phase windings, although as they comment, under their modelling assumptions the multi-tap single winding is effectively the same. They used this d-q model to derive an equivalent circuit for their prototype BDFM, and considered steady state performance of the machine in the synchronous mode, [60]. This work was furthered by Gorti et. al [41]. Boger et. al. generalised the Li d-q model to any pole pair conguration, however their starting point was an assumed stator and rotor conguration [11, 12]. Also at Oregon, Ramchandran et. al. offered frequency-domain methods of extracting most parameters for the d-q model described in [64]. Their work was initially presented in simulation only, but then partially veried by experiment in a later paper [79, 80]. At this stage, work on the BDFM at Oregon concentrated on control aspects and applications of the machine, which will be discussed later. However further contributions were made on the modelling of the BDFM by Williamson et. al at Cambridge University. They presented a generalised harmonic analysis of the BDFM which is capable of predicting the steady-state performance of any BDFM with a nested loop rotor with any stator winding, allowing the harmonic contributions to be specically analysed [115, 114, 33]. Williamson and Ferreira also used nite element analysis to verify their results and investigate the effect of iron loss and saturation [33, 34]. Williamson and Boger also used the harmonic analysis model as a starting point to investigate inter-bar currents in the BDFM. Signicantly they found that, unlike in a conventional cage rotor induction machine, the performance of the BDFM is signicantly impaired by inter-bar current, to the extent that they recommend insulation of the rotor bars [12, 117]. Nevertheless, this does not mean that it would be impossible to cast a rotor, as Koch et. al. present methods by which this can be achieved [55].
As previously discussed, Cook and Smith noted the presence of unstable operating modes in the doubly-fed single frame cascade induction machine [24, 25]. Similar lightly damped, and unstable regions were evidently of concern to the work at Oregon State University from the outset [104, p. 742]. The rst attempt, and indeed the only to date, to analyse the open-loop (that is without any feedback, speed or otherwise) stability of the BDFM was undertaken by Li et. al. in 1991 [62], the same work was also published four years later [61]. Reference [61] (and [62]) used the d-q axis model developed in [64], and linearized this model about an operating point. The approach adopted used the rst term of the Taylor series, as usual when performing a linearization. However the process is complicated by the choice of reference frame for the d-q axis model, which is synchronous with the rotor. In such a reference frame the equilibrium conditions are not constant, but sinusoidally varying. This rendered the linearised system periodic, and Li et. al. used Floquet theory (see [28]) to numerically analyse the stability, and present an algorithm for doing so. In 1992 Brasseld et. al. published a direct torque control algorithm for the BDFM [13] (also published as [14]). This is believed to be the rst control algorithm proposed for the BDFM. The algorithm was developed from the d-q model developed in [64], but using a vector ux quantity instead of rotor current states. The algorithm seeks to regulate the torque and control winding ux derivatives, and then outer PI (proportional-integral) loops use these variables to drive the controlled outputs to their reference set points. However, the algorithm, in its most basic state, requires complete knowledge of the stator and rotor currents, speed and position in real time. To overcome this limitation an estimator was proposed to estimate the ux variables, leaving only measurement of the stator voltages and currents, and the speed and position. The results were presented in simulation only. It is interesting to note that the development of the direct torque control algorithm in [14] is, in fact, closely related to the application of feedback linearization to the BDFM, although the authors do not appear to have made this connection. Feedback linearization is a non-linear control design technique where full state feedback is used to remove non-linearities in a model. The application of feedback linearization to the BDFM will be discussed in chapter 8. In reference [128] a variation of the direct torque control algorithm of [14] was presented where the outer loop PI regulation is replaced with an adaptive control law. Simulation results were presented. In 1993 Zhou et. al. presented experimental results from a so-called scalar control algorithm for the BDFM. The algorithm comprised of two PI regulation loops, one for the shaft speed and the other for the grid-connected stator power factor, by varying the instantaneous magnitude and phase of the converter-fed stator. The proposed algorithm requires a current source inverter [127]. The scalar control algorithm requires measurement of the grid-connected winding voltage, current and the rotor speed.
Introduction
In reference [127] it was also noted that the direct torque control algorithm was being implemented in hardware, however to date this is the last mention of the direct torque control algorithm, suggesting that there were practical problems associated with its implementation. One problem mentioned was that of computational complexity and associated hardware limitations, but there may have been more fundamental problems with the algorithm. A signicant development took place in 1994 when Zhou and Spe noted that the d-q model of [64] could be rotated into a reference frame in synchronism with the machine d-q currents, rather than synchronous with the rotor itself [130]. While this reference frame lacks a concrete physical interpretation it is has the advantage that the d and q quantities become constant quantities in the steady state. In reference [130] version of the d-q model in the rotor reference frame supplied with a current source inverter (rst presented in [42]) was converted into the synchronous reference frame. From the synchronous reference frame model a rotor eld orientated control scheme was developed, using the dynamic BDFM equations to decouple control winding ux and torque. The method assumes that the ux is a constant parameter, and then estimates the remaining quantities required to derive a demanded control winding current. The control input is the desired electrical torque, which is used to regulate the speed using a PI regulator. The effect of all stator resistance terms is neglected, as are leakage inductances. The control algorithm required measurement of the power winding voltages, position and speed. The control winding currents are the control outputs, which are fed into a current source inverter. Reference [125] (also published as [129]) presents an experimental implementation of the control algorithm. Principally the implementation is as described in [130], with the same control inputs, however the internal calculations were manipulated for easier implementation, using a synchronous load angle as the internal variable. Two simplications of the proposed eld orientated scheme were presented. The rst, [126], removes the need for position measurement, by assuming a xed value of the synchronous load angle during transients. This was reported as degrading the step response for a 100rpm step speed change from 0.4s to 1.2s. The second simplication proposed in [124] is conceptually similar to the original eld orientated scheme. However, no attempt is made to estimate the electrical torque, instead the synchronous load angle is regulated. Although the new algorithm constitutes a conceptual simplication, the performance of the resulting control law was improved: 0.3s for a 100rpm step change. The authors attribute this improvement is due to an increase in controller bandwidth made possible by the reduced controller complexity. An implementation of the original eld orientated scheme ([125]) was reported in [96]. The results are for a much larger machine (30kW), and although the dynamic performance is not as good (100rpm step response of 1.3s), it is likely that the speed of response was limited by inverter rating, rather than the control algorithm. More recent publications have appeared for the cascaded doubly-fed machine, which is electri-
1.2 Description of the BDFM
cally similar to the machine of Lydall, but literally comprises of two machines, in two separate frames with the rotors coupled both mechanically and electrically. A stator ux orientated control scheme and combined magnetizing ux control scheme were presented. It is claimed that these schemes would apply equally to the BDFM, although results are only presented for the cascaded machine [45, 46].

The contemporary BDFM is single frame induction machine, without any brushes, which has two 3-phase stator supplies (hence doubly-fed), of different pole numbers. Typically the two stator supplies are of different frequencies, one a xed frequency supply connected to the grid, and the other a variable frequency supply derived from a power electronic frequency converter (inverter), as illustrated in gure 1.1.
3 phase grid, 50Hz
Fractionally rated frequency converter
p1 PSfrag replacements
p2
BDFM
3 phase variable freq.
Figure 1.1: BDFM concept: a brushless, doubly-fed induction machine
1.2.1
Machine Concept
The machine may be thought of, conceptually, as two induction machines, of different pole numbers (and hence different synchronous speeds, for the same supply frequency) with their rotors connected together both physically and electrically. Physically the machine is very similar to the self-cascaded machine proposed by Hunt [48], the main distinction is that the BDFM is explicitly a doubly-fed machine. This combination of induction machines is similar to the traditional cascade connection of induction machines. In a traditional cascade connection, the rotor of one machine was connected to the stator of the next (via slip rings). However it was Hunt who realised that if both machines were in the
Introduction
same frame, then it was advantageous to connect the rotors together, rather than one machines stator to anothers rotor, as it removed the need for slip rings [48]. Hunt also showed that the rotor-rotor connections amounted to essentially the same system as stator-rotor connections. If stator 1 has p1 pole pairs, and stator 2 p2 pole pairs, then the BDFM can be operated as an induction machine of either p1 pole pairs or p2 pole pairs, by connecting stator 1 or stator 2 respectively, and leaving the other supply unconnected in each case. In the sequel this mode of operation will be referred to as simple induction mode. The characteristics of the BDFM in this mode are the same as those of a standard induction machine, except that the performance will be poor. The reasons for this will be addressed in chapter 4. However, if the non-supplied stator winding is short-circuited, then the behaviour of the machine is like that of a cascaded induction machine. A cascade induction machine formed from p 1 and p2 pole pairs. This will be discussed in more detail in chapter 4, and in the sequel this mode will be referred to as cascade induction mode. The previous two modes are both asynchronous modes of operation, that is, the shaft speed is dependent on the loading of the machine, as well as the supply frequency. However in doubly-fed mode the BDFM has a synchronous mode of operation which is the desirable operating mode, and PSfrag replacements the one for which the design of the machine is to be optimised. pole pair induction machines has characteristics which resemble an induction machine with p 1 + p2
1.2.2
Synchronous mode of operation
We shall now introduce the synchronous mode of operation of the BDFM. The synchronous mode of operation of the BDFM relies on cross-coupling between the stator and rotor [115], so does, in fact, the cascade induction mode of operation. Cross-coupling means the coupling of the eld produced by stator 1 to stator 2, and vice-versa. By design, this cross coupling cannot occur directly between stator 1 and 2 as they are chosen to be non-coupling - in the simple case this means that the pole number of each eld must be different. A full discussion of this point is given in chapter 2.
Stator 1 p1 1 f1 = 2 V1 1 p1 Rotor r r1 r2 2 p2 Stator 2 p2 2 f2 = 2 V2
Figure 1.2: BDFM concept: speed of stator and rotor fundamental elds If stator 1 has p1 pole pairs and is supplied with a three phase supply at 1 rad/s, then the stator current wave, and hence air gap magnetic ux density will rotate at 1 / p1 rad/s. Similarly if stator 2 has p2 pole pairs and rotates at 2 rad/s then the waveforms will rotate at 2 / p2 rad/s. Therefore the
fundamental magnetic ux density due to stators 1 and 2 may be expressed as follows (ignoring any xed angular offsets): b1 (, t) = B1 cos (1 t p1 )
b2 (, t) = B2 cos (2 t p2 ) ence. where B1 , B2 are the peak magnetic ux density and is the angular position around the circumferIf the rotor is rotating at r rad/s, then the fundamental ux density equations may be written in the rotor reference frame, = + r t: b1 ( , t) = B1 cos (1 r p1 )t p1
b2 ( , t) = B2 cos (2 r p2 )t p2 Therefore the magnetic ux densities, in the rotor reference frame, are travelling waves of frequency (1 r p1 ) and (2 r p2 ) respectively. Cross coupling, then, requires currents induced by b 1 in the rotor to couple with b2 and vice-
versa. By couple we mean, in this context, that a constant torque be produced, i.e a torque with a non-zero mean. Torque can be expressed as the integral from 0 to 2 of the product of a current wave and magnetic eld wave. The induced current wave due to a magnetic eld will be of the same frequency in the steady state. Therefore if a constant torque is to be produced by cross coupling then (1 r p1 ) = (2 r p2 ). Taking the negative condition and rearranging gives: r = 1 + 2 p1 + p2 (1.1)
Equation (1.1) gives the requirement on rotor speed for cross-coupling to occur in steady-state. supply may be fed with DC. This operational condition was noted by Broadway and Burbridge [17]. The condition when 2 = 0 will be referred to as natural speed in the sequel, and is given by: n = 1 p1 + p2 (1.2) It is noteworthy that 2 = 0 is a perfectly legitimate BDFM synchronous speed, that is, the second
Furthermore there must be spatial compatibility for cross coupling to occur. That is, the current induced in the rotor by b1 must produce a magnetic eld containing a p2 pole pair eld harmonic component for cross coupling to occur; and of course, vice-versa. This requirement can be satised by a special design of rotor, one example of which is the nested-loop rotor rst proposed by Broadway and Burbridge [17]. Further consideration of rotor design is given in chapter 5. When the synchronous conditions are satised then the machine produces torque, the value of which is controlled by a load angle, like that of a synchronous machine [115]. Also, the power supplied to the machine is approximately split between the two stator supplies in the ratio 2 : 1 ,
10
Introduction
therefore if only a small speed deviation is required away from the natural speed, then it is likely that a variable speed drive installation could be designed with an inverter rating a fraction of the total drive power, perhaps as small as 30% of the total [104]. Furthermore, it can be shown that the power factor of the machine can be controlled in this synchronous mode, and, subject to inverter capacity, the grid-connected winding may run at leading power factor [115].
1.2.3
Potential Applications for the BDFM
The BDFM then has some very attractive features: it is brushless in operation, offers high power factor operation when operating as variable speed drive, and can achieve variable speed operation with a fractionally rated inverter. The fractional rating of the inverter will lead to signicant economic benets, as typically the cost of a variable speed drive is dominated by the cost of the inverter [127]. The most promising applications therefore are applications requiring variable speed operation, preferably over a limited speed range, and in environments where high power factor and lack of brushes are highly advantageous. Wind power generation is a likely area of application at present, the advantages of fractional inverter and high power factor have already prompted the use of the doubly-fed induction machine [75]. The BDFM maintains these advantages but also achieves brushless operation, which particularly for off-shore installations, would be of considerable benet; increasing the time between services [8, 19]. Other applications that have been considered, include pump drives [10]. Again the motivation is the reduced inverter rating, the typically low starting torque required for uid loads means that this advantage can be maximised. There is, however, likely to be a cost penalty for using a BDFM as compared to a conventional induction machine-based drive. The rotor is likely to be more complex, hence manufacturing costs will be slightly higher, and also it is likely that the machine itself may be slightly larger for the same output torque. However the benets will certainly outweigh the costs in an appropriate number of application areas.
1.3 Approach of this work

The evidence of the previous discussions leads to a number of unresolved issues which prevent the commercial adoption of the BDFM. The BDFM is seen as a brushless replacement for certain electrical machines in particular applications, however to date no machines are in commercial service and there is no industrial experience of machine design or control. It is therefore desirable for the machines manufacturers to quantify the performance penalty incurred with the new technology (if any), as this decides the economic success, or otherwise. The performance of the machine is of commercial
11
concern in: its physical size, efciency, the extent to which it draws harmonic currents, the required inverter rating for machine operation and the dynamic stability of the drive system. Initially the focus of this work was on the dynamic stability of the machine, however it became clear that existing analysis of the machine was incomplete and provided only a partial solution. Therefore this dissertation addresses the following areas: The development of a coherent set of models for a general class of BDFM machines which accurately predict both steady state and dynamic performance (Chapters 2, 3, 4, and 7); The use of these models to investigate the steady-state performance of the machine and the stability of steady-state operating points, and to propose a method of quantifying machine performance in terms of machine equivalent circuit parameter values (Chapters 4 and 5); The estimation of machine parameter values by experiment (Chapter 6); The consideration of new rotor designs to improve the steady state performance (Chapter 5); The consideration of control strategies to stabilize and improve the damping of steady-state operating points (Chapters 7 and 8). The specic context of these areas will now be discussed.
1.3.1
BDFM Model Development
A spectrum of modelling techniques has been successfully employed in the study of electrical machines, ranging from the elegant simplicity of the equivalent circuit model to the detail and precision of nite elements modelling. Both ends of this spectrum, and methods which fall in between, have their place in BDFM modelling and have been successfully used. Finite elements modelling, from the users perspective, essentially allows experiments to be performed on detailed numerical models of electrical machines which are dened by their physical dimensions and electrical connections. This approach allows dynamic and steady-state modelling and can include the effects of saturation. This technique has been applied to the BDFM in [34, 33], and gave very good predictions. However, the complexity of the model makes the computational burden signicant, typically of the order of 1 hour of computation time to simulate 1 s, using a Pentium 4 class processor. Furthermore, the nite element approach is very much a computer-based method of experimentation rather than an analysis tool, that is, it provides little insight as to how designs may be improved, and machines must be investigated on a case by case basis. For these reasons nite elements modelling is not considered further in this dissertation. While the nite elements technique leads to accurate results, it can be difcult to gain insight into machine operation. A generalised harmonic analysis for the BDFM with a nested-loop design rotor
12
Introduction
was presented in Williamson et. al. [115, 114]. Harmonic analysis performs a harmonic decomposition of the magnetic ux density, and couplings between constituent motor circuits, which can be vital in the optimization of machine design. The model presented in [115, 114], while not able to account for saturation, can model any conguration of nested-loop rotor BDFM (although not any other rotor design) in combination with any realistic stator winding. However, as is typical with harmonic analysis, it can only be used to model the steady-state operation of the machine, which means it cannot be used investigate stability of the machine. Other modelling techniques applied to the BDFM include, dynamic coupled-circuit models, although these were applied to a specic example[110, 104]. d-q axis dynamic models have been presented [64, 63, 11, 12], however these models either apply to a specic machine, or to machines with nested-loop design rotors, and furthermore the connection between these models and the coupledcircuit model is not made clear. Equivalent circuit models have been proposed by a number of authors [17, 60, 65, 59, 41]. However, these have generally been either given without derivation, or have been for a specic machine conguration. Again the connection between these equivalent circuit models and either the d-q axis models or coupled circuit model has not been made clear. There is a need for a generalised dynamic and steady-state modelling framework which can be applied to a wide class of BDFM machines, including, but not limited to, those with a nested-loop rotor. A rigorous derivation of other models is needed from this framework to give a d-q axis model, and equivalent circuit model which are consistent and for which parameter values may be calculated and their physical meaning made explicit. This dissertation provides such a framework and rigorous derivations.
1.3.2
Use of BDFM models to investigate performance, new rotor designs and the estimation of machine parameters
The majority of previous work on the BDFM has been analysis detached from any performance objectives. While some work has been done on the basic machine operation, [41, 60, 114, 10], it has not sought to identify key machine parameters which an ideal design would optimize. Therefore it is difcult to know what design compromises will lead to good machine performance without actually designing a machine and testing it. This dissertation provides a partial solution by investigating which machine parameters in the derived equivalent circuit model have the greatest impact on machine performance. Since the work of Broadway and Burbridge the rotor design has remained largely static for the BDFM [17]. Although some investigations into improving the performance of the nested-loop design of rotor proposed by Broadway and Burbridge was undertaken [109], the design of the rotor has not been considered since. However the design of the rotor cannot be considered a closed area of research as Broadway and Burbridge themselves concede [17]. Therefore in this dissertation new rotor designs are considered and their performance evaluated using the derived equivalent circuit model
13
and by experimental means. The issue of parameter estimation for the BDFM has received relatively little attention. Only two publications appear in the literature [80, 79]. These methods nd parameter values for a d-q axis model. However little, or no, experimental evidence was presented supporting the efcacy of the approach. We therefore propose a new method of parameter estimation. The method gives parameters for the derived equivalent circuit model which are related back to parameter values for the d-q axis model, and for which considerable experimental evidence is presented for its effectiveness.
1.3.3
Analysis of stability of the BDFM and the development of control strategies
The only publication to address the stability of the BDFM in its synchronous mode of operation was presented by Li et. al. [61] (also published as [62]). However, because of the model used, Li et. al. were forced to analyse the stability of a periodic system, which therefore, did not admit the use of well-established controller design tools for LTI systems. Furthermore, the analysis was for a specic BDFM machine. There is a need for a rigorous stability analysis which covers a wide class of BDFM machines. This dissertation develops a general linearized model which admits local stability analysis by standard eigenvalue techniques, and from which controllers may be designed. We also investigate the quadratic stability of the electrical dynamics of the class of BDFM machines considered. Furthermore although a number of publications have appeared on control aspects of the BDFM [125, 129, 130, 126, 124], all of the publications containing experimental results were for control schemes which required a current source inverter. Therefore we propose two practical control strategies using a voltage source inverter. Recognising the limitations of linear controller design techniques we consider the nonlinear controller design technique feedback linearization. We give two possible methods of application to the BDFM and outline initial attempts at implementation of such schemes on a prototype BDFM.
14
Introduction
Chapter 2
BDFM Coupled Circuit Modelling and Parameter Calculation

2.1 Introduction
For the purposes of this work it is necessary for the adopted model to predict both dynamic and steady-state performance of a range of different machine designs. Furthermore, in order to make use of the model and to explore potential machine designs it is necessary to calculate parameter values for any such model. As discussed, it has been decided to use a coupled-circuit model for this purpose. Wallace et. al. and Spe et. al. [110, 104] used a coupled circuit technique to model a prototype BDFM, however no generalisation of this model was ever presented. Although Wallace et. al used the model to simulate their prototype machine, [104], the limitations of the model were never addressed, specically, the inclusion of leakage inductance effects, allowance for nite conductor widths and the generalisation of the model to a wide class of machines. The work presented here is similar to their work, but generalised, to allow any realistic BDFM to be modelled, allow the inclusion of some leakage inductance effects and makes allowance for nite conductor widths. The machine is modelled as a set of interconnected coils. This assumption is reasonable in practice, as will be argued, as it allows almost all industrial-type electrical machines to be modelled without further simplication. From this standpoint, relationships between stator and rotor of the machine are derived in terms of mutual inductance matrices. Furthermore the parameter calculation technique adopted is particularly suited for implementation on computer, and when so done allows signicant modication of the design of the machine to be performed with relative ease. This is achieved by calculating inductance parameters in a coil by coil basis, and then combining the results in a manner appropriate to the circuit being analysed. In contrast to [110], two methods are proposed for the calculation of inductance parameters. Firstly the parameters are calculated by direct integration, and secondly by splitting into Fourier
15
16
series prior to integration. The latter method is useful as it allows the harmonic content to be investigated. This is something that will be important in chapter 5. Finally a dynamic model is derived for the BDFM using the calculated parameters. Some of the results presented in this chapter will be of no surprise to readers familiar with electrical machine modelling. However the BDFM is an unusual machine, and great care must be taken in the analysis as not all results applicable to related machines, such as the induction machine, can be directly transfered. For this reason, and because it is anticipated that readers from a system-theory background will have had less exposure to electrical machine modelling, derivations are presented in detail. An attempt has been made to state assumptions at the start of each section and in some cases references are provided to well-respected works which adopt the same assumptions.
2.2 Preliminaries
The following denitions are standard, for further information see, for example, [122]. Denition 2.1. Let there be N turns of wire with current i R A owing in each. Let the magnetic
eld intensity vector be H R3 , dl R3 a direction vector of innitesimal magnitude, J be the current density, and d A R3 a direction vector normal to the surface S of innitesimal magnitude. H dl = J d A = Ni
Amperes Law states:
where S is any open surface bounded by a closed path C. In the case that the surface S contains N conductors each carrying current i, then the right hand side reduces to N i, as shown. A right-hand coordinate system will be used, therefore an anti-clockwise closed path requires the positive direction of current ow to be out of the page. Amperes Law is only valid if the rate of change of electric ux density is negligible. This fact will be assumed. Denition 2.2. [35, p. 132] The magneto-motive force (mmf) of any closed path, C, is dened as the net ampere-turns, N i =
S
J d A, enclosed by that path: FC

S
J d A = Ni
where N , i, S, and J are as given in denition 2.1 Denition 2.3. A magnetic material is said to be linear if:
: B = H, constant
2.2 Preliminaries
17
where is the magnetic permeability of the material, and B is the magnetic ux density. Linearity of the magnetic material implies that there is no saturation, or skin effects. Denition 2.4. The magnetic ux, , passing through a surface, A is dened as:
A
B dA
Denition 2.5. Gausss Law for magnetic elds states that total magnetic ux out of any closed surface, S, is zero:
S
B dA = 0
Denition 2.6. The magnetic ux l , linked by a circuit with conductor density, C, is dened as: l
A2 A1 (A2 )
B d A1 Cd A2
where the surface A 1 is dened by the closed curve due to elemental conductor Cd A 2 within which the magnetic ux density is B. In the simple case of N coincident conductors bounding a surface A the ux linked by the circuit reduces to: l N
A
B dA
Therefore from denitions 2.1, 2.3 and 2.4, in a network of k circuits, if a magnetic material is, or can be assumed to be, linear, the magnetic ux l linked by circuit n k due to magnetic ux produced by circuit m k circuits is linearly dependent on the current owing in circuit m. Denition 2.7. For a network of n circuits with linear magnetic material the Mutual inductance, owing in the n circuits: M Rnn , can be dened in terms of the ux linked by each of the n circuits due to the current l = Mi where M is not a function of i. The ux linked by the j th circuit due to current owing in the k th is M jk . Denition 2.8. Self inductance, L k , is the ux linked in the k th circuit (k n) due to current owing in the k th circuit. Therefore: L k = Mkk i.e. the self inductance terms are the diagonal elements of M. Lemma 2.1. [72] From energy conservation it can be shown that: i, j n : Mi j = M ji (2.1)
18
Lemma 2.2. [72] For any mutually coupled coils of self inductance L i and L j and mutual inductance Mi j : Mi j ux linked by that circuit: v= dl dt Li L j (2.2)
Denition 2.9. Faradays law relates the voltage induced across the terminals of a circuit due to the
where the electro-motive force (emf) induced in the circuit is given by v. From this denition the following are immediate: Denition 2.10. A matrix, A Rnn is said to positive (semi) denite if x T Ax > ()0 x Rn = 0. 1. If A, B are positive denite then A + B is positive denite. 2. If A is positive denite and > 0 where is scalar, then A is positive denite. 3. A 0 0 B is positive denite if and only if A and B are positive denite.
Analogous results follow for positive semi-denite matrices and negative (semi) denite matrices. Lemma 2.3. [72] For any set of mutually coupled coils , M Rnn , with constant inductance
1 T i Mi. 2
parameters, the work done in increasing the current from 0 to i R n , that is the energy stored, is Furthermore from Lenzs law, with the chosen sign convention, this energy must be greater than or equal to zero. We may therefore conclude from def. 2.10 that M is positive denite.
2.3 General Electrical Machine Coupled-Circuit Model

and Rn being the ux linked by each circuit, combining Faradays law and Ohms law: v = Ri + where R Rnn is the matrix of resistances. d dt Given a network of n circuits, each with terminal voltage, v R n , current i Rn owing in them,
(2.3)
This modelling approach is called a coupled-circuit or coupled-coil model.
It should be noted that for any physically realizable system the principle of reciprocity requires that R and M are symmetric, that is that R = R T and M = M T . Further, it is usually the case that R is diagonal.
2.3 General Electrical Machine Coupled-Circuit Model
19
Assuming a linear magnetic circuit, from the denition of mutual inductance, def. 2.7, equation (2.3) may be expanded using the chain rule to give: v = Ri + position, r R. di dM i+M dt dt (2.4)
In electrical machines the mutual inductance term, M, is assumed to vary only with the rotor We dene the rotor angular speed as: r dr dt (2.5)
It is then natural, to express (2.4) in terms of the rotor angular speed: dM di i+M dr dt
v = Ri + r
(2.6)
In systems theory terms an electrical machine can be thought of as a dynamic system with input voltage and output currents. In this case, the system admits a state-space representation where the currents are the system states. From (2.6): dM di = Ri r i +v dt dr
(2.7)
2.3.1
Torque Calculation
The electro-magnetic torque developed by the machine couples the electrical differential equations to the mechanical differential equations. The torque produced by an electrical machine can be determined by considering instantaneous power transfer in the system. Firstly we recall that the energy stored in magnetic eld in a coupled circuit network, or magnetic co-energy, is dened as: 1 Wco = i T Mi 2 Therefore the instantaneous power transfer from the magnetic eld is given by: 1 di T 1 dM 1 di dWco = Mi + i T i + iTM dt 2 dt 2 dt 2 dt di 1 TdM T i +i M = i 2 dt dt Multiplying (2.4) by i T from the left: i T v = i T Ri + i T dM di i + iTM dt dt (2.10) (2.8)
as
Wco is scalar and M = M T dt
(2.9)
20
Therefore from (2.9) with (2.5), and expanding via the chain rule gives: 1 dM dWco 1 dM dWco i T v = i T Ri + i T i+ = i T Ri + r i T i+ 2 dt dt 2 dr dt (2.11)
Equation (2.11) is a balance of the instantaneous power transfer in the system. The input power, i T v is balanced by the Ohmic loss term, i T Ri, the rate of change of stored energy in the magnetic eld,
dWco , dt
and the remaining term corresponds to the mechanical power generated by the machine. 1 dM Te = i T i 2 dr
Therefore the electrical torque generated by the machine is given by: (2.12)
Notice that the torque may equivalently be dened as: Wco r
Te =
(2.13)
2.4 Calculation of Parameters for Electrical machines

In the previous section the general electrical machine equations have been derived. In order to proceed with simplifying transformations it is necessary to explore the structure of the inductance and resistance matrices in (2.4). The structure is most simply revealed by calculating values for these parameters, under the assumptions stated in section 2.3. The following additional assumptions will now be made: The permeability of iron can be considered to be innite. The air gap width between the stator and rotor is small enough, relative to the machine length and diameter, that it may be assumed that: The effects of nite machine length can be neglected. The lines of magnetic ux in the air gap are parallel. The machine is assumed to consist of a cylindrical rotor inside a cylindrical stator, with conductors which may each be adequately modelled as a uniform conductor distribution on the surface of either the rotor or stator cylinder within the region containing those conductors (see gure 2.2). That is, the conductors are assumed to be of zero depth, but nite width. For the purposes of mutual inductance calculation the rotor and stator are separated by an effective air gap width, given by Carters Factor. See section 2.8 for full details. All conductors in each slot of the stator (or rotor) are uniformly distributed over the same pitch, although the pitch may differ between stator and rotor. The pitch over which the conductors are distributed is assumed to be the slot mouth width.
21
The effects of leakage (discussed in section 2.5.2) may be satisfactorily modelled by some suitable addition to the self-inductance terms only. The assumptions stated above, combined with those used to derive (2.4), are known to yield a model with satisfactory accuracy in the modelling of induction machines and BDFMs: [115] and [110] present BDFM models derived using these assumptions. [114] and [104] verify these respective models against experimental results. Furthermore, [113] and [116] give examples of models for cage rotor induction machines derived with these assumptions, the results being veried against experimental data. As calculation of parameters for a variety of different machines is required, the approach presented here is motivated by versatility. It is similar to that described in [110], but the assumption of point conductors relaxed, and generalised for any machine for which the stated assumptions are reasonable. We rst calculate the ux density due to unit current owing in a single coil. We then use this result to derive an expression for the mutual inductance between any two coils with a particular class of conductor distributions (to generalise to any conductor distribution is trivial, but of limited application). We then split the machine windings and rotor circuits into single coil elements and compute the mutual inductance between all single coil elements. We then sum the single coil elements taking care with the sign to yield the machine parameters. The calculation of ux density and inductance for a single coil is well-known, and can be found in, for example, [43, ch. 1], although it is usual to express the result as a Fourier series rather than a denite integral, as is required in this case. Furthermore point conductors are usually assumed, whereas, in common with [116] we allow for uniformly distributed nite width conductors. For these reasons, and for the benet of the reader from a systems theory background, a full derivation is included.
2.4.1
Magnetic Flux Density due to Single Coil
We wish to determine the spatial magnetic ux density produced in the air gap due to current i owing in a coil of N turns connected in series. The coil is the basic building block used to create machine windings. From Denition 2.1, we may relate the line integral of the magnetic eld intensity around any closed path to the current owing through the surface bounded by the closed path. Figure 2.1 shows a machine rotor in plan with a section through the rotor. As the air gap is assumed small, the rotor and stator diameter are taken to be the same. Figure 2.2 shows a cross-section through a rotor and stator of an electrical machine. For simplicity, the cylindrical rotor and stator surfaces have been attened, thus the far left and right hand edges of the diagram are coincident. The gure shows two stator slots lled with conductors, and the remaining slots empty. and and out of the page respectively. indicate current ow into
Consider the closed path, C 1 : as we assume that the iron has innite permeability the portion of the line integral of magnetic eld intensity in the iron is zero regardless of the exact path. Therefore
22
PSfrag replacements
A typical slot lled with one or more conductors
frag replacements
Figure 2.1: Cross-section and plan views of a typical machine rotor, showing the rotor diameter, d, and lamination stack length, w.
Stator teeth c
Stator slots
Stator =
H1
H2
ws s
H3
H4
B J

C2 B2 Ni

C1
Rotor =
air gap = 0
Figure 2.2: Magnetic ux and current density for a stator coil, all dimensions in radians except for g, the air gap width.
B1
Ni
23
from Ampres Law (def. 2.1), because J = 0 everywhere, around C 1 : H3 g H4 g = 0 H3 = H4 (2.14) (2.15)
where g is the air gap length as shown in gure 2.2, and Hx are magnetic eld intensities. Whereas, around path C 2 we get: H2 g H1 g = N i where N is the number of conductors each with current i owing in them. From def. 2.3 the total change in air gap magnetic ux density across a slot with N conductors with current i owing in each is: B = B2 B1 = N i0 g (2.16)
Thus equations (2.16) and (2.14) show that changes in magnetic eld in the air gap only occur between teeth, and that the value of this change is given by (2.16). As in [116], it is assumed that the change in magnetic eld is linear across the slot mouth, w s , which therefore means that the current density, J for the two slots shown is as given in gure 2.2. The magnetic ux density, B is shown in blue. In order to calculate B1 and B2 Gausss Law (def (2.5)) may be applied to the cylindrical surface of the machine rotor. By considering the ux density waveform (see gure 2.2), it can easily be shown that:
B2 c + B1 (2 c ) = 0 where c is the coil pitch in radians, as shown in gure 2.2. From (2.17) and (2.16): N i0 c g 2 N i0 (2 c ) B2 = g 2 B1 = where ws the slot mouth width in radians, as shown in gure 2.2.
(2.17)
(2.18) (2.19)
We have now calculated the air gap magnetic eld due to current i owing in a coil, as shown in gure 2.2. We may now write down the equation of the air gap magnetic ux density, B k (), due excitation of the k th circuit as shown in blue in gure 2.3 (the green lines depicting the j th circuit conductor distribution will be used in a later section):
24
Bk () =
B + (B B ) k 1k 2k 1k ws k B2k
1k
k < < k + w s k k + w s k < < k + ck k + ck < < k + ck + w s k otherwise (2.20)
Using (2.18) and (2.19), equation (2.20) may be written in terms of the current, number of turns and slot and pitch geometries of the k th circuit, shown in blue in gure 2.3: ck + k 2 ws k 2 ck N k i k 0 2 Bk () = k ck 2 ck g 2 ws k ck 2 k < < k + w s k k + w s k < < k + c k k + ck < < k + ck + w s k otherwise (2.21)
B2 + (B1 B2 ) k ck k k k ws k B
2.4.2
Mutual (and self) Inductance of Single Coils
From the denition of mutual inductance 2.7 and ux linked 2.6, the mutual inductance between the j th circuit and k th circuit of an electrical machine, M jk , is given by: M jk i k = w d 2
2 0
Bk ()
C j ( )d d
where i k , Bk are the current and magnetic ux density in the air gap in the k th circuit, C j is the conductor density function for the j th circuit (
0
C j ( )d will be referred to as the conductor distribution
function for the j th circuit). In particular, the conductor density function is related to the current density for the j th circuit by the coil current, i j : J j () = C j ()i j As self inductance is a special case of mutual inductance, mutual inductance will henceforth be used to refer to both mutual and self inductance. We have assumed that the conductors in each coil are uniformly distributed across the slot mouth width (ws in gure 2.2). Therefore, as each coil comprises of N series connected turns, the current in each turn is the same, and the current density function will always be of the form shown in gure 2.2, with each rectangle having area N i, as depicted.
PSfrag replacements

0
25
C j ( )d Nj B2k c j ck
Bk
B1k j k
ws j ws k
Figure 2.3: Air gap magnetic ux density due to excitation of the k th coil and conductor distribution for the j th coil Figure 2.3 shows Bk () and d M jk i k = w N j 2
j +ws j j 0
C j ( )d for a typical jk pair. From this diagram we may write: j d + ws j

j +c j j +ws j j +ws j +c j j +c j
Bk ()
Bk ()d+ Bk () j + w s j + c j d ws j
0
(2.22)
where Bk (), the magnetic eld due to the k th circuit is given by (2.21) and ductor distribution function for the j circuit.
th
C j ( )d is the con-
Therefore the mutual inductance between any two coils of differing pitches, and positions may be calculated by considering unit current ow in one coil, and computing Mi using (2.22), because the integral can be analytically solved for the magnetic eld distribution assumed.
2.4.3
Calculation of spatial harmonic components of mutual inductance
Equation (2.22) can be used to calculate space harmonic components of the mutual inductance between two coils. The space harmonic components of the mutual inductance between two coils are the components of mutual inductance due to terms of the Fourier series representation of magnetic ux density. The method described is essentially the same as that used in [113, 116] although the presentation is rather different, and lends itself more directly to calculation by computer. The Fourier series representation of the magnetic ux density for a single coil can be derived by nding the Fourier series of equation (2.21). It is straightforward to show that it is given by: Bk () = 2Nk i k 0 g
n=1
sin(nck /2) sin(nwsk /2) cos n k n nwsk /2
ck 2
w sk 2
(2.23)
26
Thus the n th spatial harmonic component of magnetic ux density for the k th coil is: Bkn () = 2Nk i k 0 sin(nck /2) sin(nwsk /2) cos n k g n nwsk /2
ck 2
w sk 2
(2.24)
Hence the n th spatial harmonic component of mutual inductance linking the j th coil to the k th coils is: d M jk n = w N j 2
j +ws j j 1 Bkn ()
j d + ws j
j +c j j +ws j
1 Bkn ()d+
j +ws j +c j j +c j 1 where Bkn is the Bkn due to unit current ow.
1 Bkn ()
j + w s j + c j d ws j
(2.25)
Alternatively (2.25) can be derived directly from (2.4.2). The conductor distribution function,
0
C j ( )d from gure 2.3 may be written as a Fourier series:

0
C j ( )d =
2N j
sin(nc j /2) sin(nws j /2) cos n j s j /2 ws j /2 n nws j /2
(2.26)
Before proceeding it is necessary to recall the following trigonometric result: Lemma 2.4. Given m, n N and 1 , 2 , R then:
2 0
cos(n + 1 ) cos(m + 2 )d = 0 n = m
Proof. Apply trigonometric identity 2 cos(A) cos(B) = cos(A + B) + cos(A B), then integrate, and the result comes by inspection. Substituting (2.26) and (2.23) into (2.4.2), and using Lemma 2.4 gives:
M jk = w
d 2Nk 0 2N j 2 g
2 0
nN
sin(nck /2) sin(nwsk /2) sin(nc j /2) sin(nws j /2) n nwsk /2 n nws j /2
ck 2
cos n k
w sk 2
cos n j s j /2 ws j /2 d
(2.27)
Remark 2.5. The principle used to derive (2.27) applies more generally to the mutual inductance between any two circuits. From equation (2.4.2) and Lemma 2.4 it can be seen that the mutual inductance between two circuits can only consist of terms from the Fourier series representation of the magnetic ux density distribution, B() and the conductor distribution function,
0
C( )d which
corresponds to the same absolute harmonic. More precisely if Q b , Q c denote the set of absolute harmonics which have non-zero coefcients for the magnetic ux density and conductor distribution function respectively, then only harmonics in Q b Q c contribute to the mutual inductance.
2.5 Mutual Inductance of Machine Windings
27
Remark 2.6. The terms absolute and relative harmonic are used to differentiate between the mathematical harmonics in a Fourier series (i.e. the value of n), and relative harmonics, that is integer multiples of some frequency (or pole number) eld. In other words the fundamental absolute harmonic is n = 1, the second absolute harmonic n = 2 and so on. Whereas the rst relative harmonic is n = k, for some integer k and the second relative harmonic is n = 2k and so on.
2.5 Mutual Inductance of Machine Windings

Using equation (2.22) the mutual inductance matrix for a machine winding may be calculated. We are concerned with the calculation of mutual inductance parameters for any winding which can be made up of interconnected groups of coils. A group of coils is a number of coils, typically in different positions within the frame, connected in series. In order to calculate the mutual inductance matrix for a machine winding, each group of coils can be represented by a single circuit as it is series connected. Therefore it can be represented by a single self-inductance term with mutual inductance terms for each coupling to other coil groups. Furthermore, these inductance terms are the summation of the inductance terms of the individual coils making up the group: suppose that there are m sets of series connected groups of coils. Hence there are m circuits in total, and the voltage induced at the terminals of each of the m circuits is given by: v = TM di dt (2.28) Suppose that there are a total of N coils, with mutual inductance matrix M R N N . Further
1s, -1s and 0s. Typically T will have only one non-zero element per column, because each coil is only connected to one series set. T can be found by considering the connections between the coils opposite way around. which form each group. When elements of T are 1 this corresponds to the coil being connected the
where i R N is the current vector and T RmN , T is a combination matrix made up of only
Furthermore, if the current owing in each of the m series connected groups of coils is i, then, after some consideration, it can be seen that the current in each of the N coils is T T i. Therefore: M = T MT T where M is the mutual inductance matrix for the m sets of series coils. The procedure outlined above can be used to calculate the mutual inductance matrices for any machine subject to the constraints described at the start of the section. The method has been implemented in Matlab, and has been used to calculate all inductance parameters used in this dissertation. Functions were written which compute the mutual inductance between individual coils, returning the answer as a symmetric matrix, the fact that the matrix must be symmetric facilitates the elim(2.29)
28
ination of programming errors. Functions were written for the direct calculation method, equation (2.22), and for the calculation method using the Fourier series representation, equation (2.27). As each coil must be placed in a slot, the functions were written to take a list of slots which house the out and return conductors for each coil, and from this compute the mutual inductances. When one coil is on the stator of the machine, and the other on the rotor, then the functions take an angular offset parameter. The stator-rotor (and hence rotor-stator) mutual inductance terms will, of course, be parameterised by this angular offset.
2.5.1
Calculation of rotor and stator inductance matrices
When applying the previously outlined procedure to a machine, the transformation matrices can always be partitioned into a block diagonal matrix. By choosing the coil order so that the stator coils appear before the rotor coils, the transformed mutual inductance matrix will be: M= Tss 0 0 Trr Mss T Msr Msr Mrr Tss 0 0 Trr
T
(2.30)
rotor respectively. Tss , Trr are the combination matrices for the stator and rotor respectively. Note that the Msr will be a function of the rotor position at any instant.
where Mss , Msr , Mrr are the coil mutual inductance matrices for the stator, stator to the rotor, and the
Appendix B.2 gives full details of the stator winding used on the prototype machine, including
the coil order, and coil combination matrix used, and thus serves as an example application of this calculation procedure. Appendix B.2 also reviews standard winding factors which are often used in the design of machine windings and can be used to verify inductance calculations for certain winding congurations.
2.5.2
Leakage inductance
Thus far, inductance arising from leakage effects have not been discussed. Inductance arising from leakage effects are inductances due to the magnetic ux not linking stator and rotor conductors. This denition is similar to that found in [31, sect. 5.2], except that for our purposes the harmonic mutual inductance terms are not considered part of the leakage terms, rather as a separate term (contrary to [43]). There are various types of leakage effects, a summary of those used is found in appendix C.1. Typically leakage inductances in a well designed machine are 1-2 orders of magnitude lower than the principal inductance terms. For simplicity it will be exclusively assumed that leakage inductance causes an increase in self inductance only. This assumption will typically lead to only very small errors, and simplies the calculations which follow.
2.6 Some properties of machine windings
29
2.5.3
Resistance calculation
The calculation of resistance is rather more straightforward than inductance calculations. We assume that the resistivity is constant, and therefore the resistance of a coil is given by: R= l A (2.31)
where A is the wire cross-sectional area, l is the total length of wire in each coil, and is the resistivity of the material at the operating temperature. The coils length is approximately given by: l = 2k N (c d + w) 2 (2.32)
where c is the coil pitch, d the diameter of the centre-line of the slots, w the machine stack length, N the number of turns in the coil, and k > 1 a constant to allow for the space required to bend the conductors in the coil, typically k < 1.1. Once the coil resistance has been calculated, the resistance of each group may be computed using the same combination matrices used to compute the group inductance parameters. In this dissertation it is assumed that the resistance of each coil is constant throughout the machine operation. The major source of error introduced by this assumption is the increase in resistance due to temperature. However if resistance values are calculated at a normal running temperature, then the error will be minimised. There is also a small error due to the skin effect, however at the frequencies involved, with copper or aluminium conductors, this error is likely to be negligible.

Having described methods of computation for machine inductance parameters. It will be useful to derive analytic expressions for a class of stator windings. The general approach is well-known, however a complete derivation is included to ensure consistency with the previously used notation, and so that it can be used to draw some conclusions about suitable, and unsuitable, winding arrangements for the BDFM. The ensuing approach shares much with that adopted in [43, chap. 1], although the presentation is different. The following properties will now be assumed for stator windings. These assumptions cover concentrated, and distributed windings both fully pitched and short pitched when series connected: Each coil is of the same pitch. The machine slot pitch is constant for this winding. The coil is located inside the slots.
30
The winding is comprised of groups of series connected coils. Each group comprises of N g coils separated from one another by a constant angle 1 . There are 2 p groups in each phase of the winding. Before proceeding we will recall three trigonometric results which will be required later: Lemma 2.7. Given R, N N: e j (N 1)/2 sin(N /2) N 1 sin(/2) e j (n) = N n=0
N 1 t=0
= 2k, k Z otherwise
Proof. Notice that the left hand side is a geometric progression. Recalling that: xt = 1 xN , x C=1 1x
then, for = 2k, k Z: LHS = 1 e j (N ) e j (N )/2 = 1 e j () e j ()/2 sin(N /2) = e j (N 1)/2 sin(/2) e j (N )/2 e j (N )/2 e j ()/2 e j ()/2
Lemma 2.8. Given p, n, k N then: n = kp, p odd sin(n) = cos(k) n = kp, p even p sin(n ) p 0 otherwise 1
Proof. When n = kp, sin(n ) = 0, and clearly sin(n) = 0, hence the result. p Consider the limit as q n. By LHpitals rule:
qkp
When n = kp we investigate what happens as n approaches kp. Let q = n + where q, R. lim sin(q) cos(kp) cos(q) = 1 = lim = p sin(q ) qkp p cos(q ) cos(k) p p p
Lemma 2.9. Given , R = 2k (k Z), N N: cos()+cos( )+cos( 2)+ +cos( (N 1)) = sin(N /2) cos ( (N 1)/2) sin(/2)
31
Proof.
Applying the trigonometric identity cos(A B) = cos(A) cos(B) + sin(A) sin(B) gives: LHS = cos() [1 + cos() + cos(2) + + cos((N 1))] +
N 1 n=0
sin() [sin() + sin(2) + + sin((N 1))] cos(n) + sin()

N 1 n=0
= cos()
sin(n)
From Lemma 2.7, taking real and imaginary parts we may write: LHS = cos() cos((N 1)/2) sin(N /2) sin(N /2) + sin() sin((N 1)/2) sin(/2) sin(/2)
recombining using cos(A B) = cos(A) cos(B) + sin(A) sin(B) gives: LHS = sin(N /2) cos ( (N 1)/2) sin(/2)
Magnetic ux density due to a coil group We consider a group of N g series connected coils of N c turns each, with equal current i g owing and each displaced from the next by constant angle, g . Each coil has a magnetic ux density distribution given by (2.23) with k = {g , g + g , , g + (N g 1)g } for the different coils in the group, summed. It can be shown using Lemma 2.9: Bg () = 2Nc N g i g 0 g Nk = Nc , i k = i g To compute the resultant eld distribution the contribution from each coil is sin(nc /2) sin(nws /2) sin(n N g g /2) n nws /2 N g sin(ng /2) cos n g see [43, Sect. 1.4] for details. Magnetic ux density due to a single series connected phase It is assumed that a single phase comprises of either p or 2 p groups of coils where p N is the number of pole-pairs. If the winding is a double layer winding (where coils are short-pitched), then it is assumed that each phase comprises of 2 p coil groups connected in series with alternating polarity. If the winding is single layer (fully pitched) then it is assumed that each phase has p coil groups, connected in series, all of the same polarity.
c 2
nN
ws 2
(N g 1)
g 2
(2.33)
32
In both cases the total magnetic ux density for a phase is the sum of its component groups. The ux density of each group is given by (2.33), but with different values of the group offset, g . layer winding g = {k , k + , , k + p In the case of the single layer winding g = {k , k +
2(2 p1) }, 2p 2 , p
where k is the angular offset of the k th phase.
, k +
2( p1) }, p
for the double
However in the case of the double layer winding the polarity of connection alternates. Using Lemma 2.9 the magnetic ux density for the k th phase, with current i k , of a single layer winding can be written as:
Bksing. () =
2Nc N g i k 0 p g
nN
sin(nc /2) sin(nws /2) sin(n N g g /2) sin(n) n nws /2 N g sin(ng /2) p sin(n/ p) cos(n( k
c 2
ws 2
(N g 1)
g 2
( p 1) )) p
from Lemma 2.8, assuming p even (modication trivial for p odd): = 2Nc N g i k 0 p g sin(nc /2) sin(nws /2) sin(n N g g /2) cos(n/ p) n nws /2 N g sin(ng /2)
c 2
n= pt tN
(2.34)
cos(n( k
ws 2
(N g 1)
g 2
( p 1) )) p
p
For a single layer winding, the coil pitch, c will generally be c =
[99, sect. 10.7], therefore all
even absolute harmonics vanish. The resulting magnetic ux density is: = 2Nc N g i k 0 p g sin(nc /2) sin(nws /2) sin(n N g g /2) cos(n/ p) n nws /2 N g sin(ng /2) n= p(2t1)
tN
(2.35)
cos(n( k
c 2
ws 2
(N g 1)
. p
g 2
( p 1) )) p
A double layer winding consists of coil groups spaced by
The polarity of connection of these
coil groups alternates, thus a phase of a double layer winding can be thought of as two phases of a single layer winding connected in series, but with one set the opposite polarity to the other. Therefore the magnetic ux density may be written as:
2Nc N g i k 0 p Bkdoub. () = g
sin(nc /2) sin(nws /2) sin(n N g g /2) sin(n) n nws /2 N g sin(ng /2) p sin(n/ p)
c 2
cos(n( k cos(n( k
ws 2
(N g 1)
g 2
( p 1) )) p (2.36)
c 2
ws 2
(N g 1)
g 2
( p 1) )) p p
33
from Lemma 2.8, assuming p even (modication trivial for p odd): = 2Nc N g i k 0 p g sin(nc /2) sin(nwsk /2) sin(n N g 1 /2) cos(n/ p) n nwsk /2 N g sin(n1 /2)
c 2
n= pt tN
cos(n( k cos(n( k
c 2
ws 2
(N g 1)
g 2
g 2
( p 1) )) p
(2.37)
ws 2
(N g 1)
( p 1) )) p p
applying the identity cos(B) cos(A) = 2 sin( A+B ) sin( AB ): 2 2 = 4N g Nc i k 0 p g sin(nc /2) sin(nws /2) sin(n N g 1 /2) cos(n/ p) n nws /2 N g sin(n1 /2)
c 2
n= pt tN
(2.38)
sin(n 2p ) sin(n( k
ws 2
(N g
1) 2g
( p 1) p
)) 2p
notice that the sin(n 2p ) suppresses even multiples of the working harmonic, p. Hence the magnetic ux density for a phase from a double layer winding may be written as: = 4Nc N g i k 0 p g sin(nc /2) sin(nws /2) sin(n N g 1 /2) cos(n/ p) n nws /2 N g sin(n1 /2) n= p(2t1)
tN c 2
(2.39)
sin(n 2p ) sin(n( k
ws 2
(N g
1) 2g
( p 1) p
)) 2p
2.6.1
Mutual inductance between two stator windings
Equation (2.39) gives the magnetic ux density for a single phase of any winding realisable under the assumptions stated. Using a similar method to that used to derive (2.39) the conductor distribution function for each phase may be derived. Equation (2.26) gives the conductor distribution for a single coil, hence the conductor distribution for a phase of a double layer winding can be written as:
0
C j p ( )d =
4Nc N g p
sin(nc /2) sin(nws /2) sin(n N g 1 /2) cos(n/ p) n nws /2 N g sin(n1 /2) n= p(2t1)
tN
sin(n 2p ) sin(n(
c 2
ws 2
(N g 1)
g 2
( p 1) p
)) 2p
(2.40)
From equation (2.4.2) the mutual inductance between two phases is proportional to the integral of the product of (2.40) and (2.39) (if the winding is single layer a similar result ensues). This leads to the following results: Remark 2.10. The mutual (and self) inductance between phases within any double or single layer three phase winding can only comprise of terms from the Fourier series representation of the magnetic ux density (or equivalently the conductor distribution function) which are odd harmonics of the
34
working harmonic (the number of pole pairs). Note that the mutual inductance will include 3rd harmonics of the working harmonic, however these terms can be shown to cancel out if the supply is a balanced three phase set, see for example [43]. Remark 2.11. The individual phases of a three phase winding will have the same self-inductance,
and same mutual-inductance between phases. This is self-evident from considering (2.39) and (2.40): different phases differ only in the value of k and j , which correspond to a phase offset, which has no bearing on the value of the integral (Lemma 2.4). The self-inductance has k = j , and the mutual phase winding the stator-stator mutual inductance matrix will be of the form: L s Ms Ms Mss = Ms L s Ms Ms Ms L s which, from denition A.1 is a symmetric, circulant matrix. Rules for choosing pole numbers for BDFM stator windings Kusko and Somuah [57] give general guidelines for choosing non-coupling pole pair combinations, however they offer no proof, or say for which windings their claims apply. From the preceding work a general statement may be made: Theorem 2.12. The mutual inductance between any two phases of any single layer three phase windings of p1 N and p2 N pole pairs respectively ( p1 > p2 ) will be zero if and only if, p1 is even and p2 is odd, or p1 and p2 are both even providing that p1 / p2 is not odd ( p1 / p2 may not be an integer) and p1 = p2 . Furthermore the condition is sufcient for double layer windings. Proof. Sufciency: A three phase winding of p1 pole pairs produces the set of absolute harmonic Q 1 N and for p2 pole pairs, Q 2 p2 (2t 1), t N. p1 (2t 1), t inductance k = j + 2/3 (for k > j , which can be ensure for all phases). Therefore for a three
If p1 is even and p2 odd then Q 1 therefore contains only even numbers, and Q 2 only odd numbers. Hence Q 1 Q 2 = , hence from Remark 2.5, the mutual inductance will be zero. Q 1 = p1 (2k 1), k N and Q 2 = p2 (2k 1), k N, hence Q 1 / p2 = p1 (2k 1)/ p2
and Q 2 / p2 = (2k 1). As p1 / p2 is not odd then Q 1 / p2 Q 2 / p2 = , as Q 2 / p2 is odd, hence Q 1 Q 2 = . Necessity:
From (2.35) none of the terms will be zero, hence: non-zero. If p1 and p2 are odd then p1 p2 is odd, and thus p1 p2 Q 1 , hence the mutual inductance is If p1 and p2 is even and p1 / p2 is odd, then Q 1 / p2 and Q 2 / p2 are both odd, and hence q
Q 1 / p2 : q Q 2 / p2 , hence Q 1 Q 2 = hence the mutual inductance is non-zero.
35
Remark 2.13. For a double layer winding, the sin(nc /2) term will typically be zero for certain of slots per pole, and q is the number of slots by which the winding is short-pitched. Therefore if, result as all odd 15th harmonics are cancelled. However, practically speaking, because k p2 c , hence Theorem 2.12 is necessary and sufcient in most practical situations. harmonics, by design. It can be shown that this term is equivalent to cos( nq ) where k is the number k say, p1 = 1, p2 = 15, and if the 2 pole winding is wound with q = 1, k = 5 then no coupling will q unless p1 too then there is little chance of being able to cancel out the coupling by careful choice of
Remark 2.14. Notice that Theorem 2.12 is derived from consideration of the mutual inductance between phases of different windings. It can be shown (see e.g. [43]) that when a p pole pair three phase winding of the types assumed, is supplied with a balanced set of three phase voltages, absolute are zero. Therefore the requirement of theorem 2.12 for zero coupling may be relaxed: any p1 and machines, such as [11]. However extra caution should be exercised, as there will be direct coupling between the two different pole number stator whenever the balanced condition is not met (e.g. due to differing end winding resistances in each phase). There is a further requirement which is imposed on candidate stator pole pair combinations, the requirement that there be no unbalanced magnetic pull on the rotor. If there were this would unduly stress the machine bearings. Creedy [27, p. 512] gives necessary and sufcient conditions for this to be so. He points out that the conditions given by Hunt [49, p.408], who rst made this point, are sufcient but not necessary. Creedy states that to avoid unbalanced magnetic pull it must be ensured that, ...at every instant any value of the square of the magnetic [ux] density which appears at any point [around the circumference of the rotor] shall also appear at one or more other points equally distributed around the circumference with respect to the rst. However Creedys interpretation of this rule is still only sufcient. It is worth noting that some authors have cited Broadway and Burbridge [17, p. 1281] as stating that unbalanced magnetic pull can be avoided by ensuring that the pole pairs of the stator winding differ by at least 2, in fact Broadway and Burbridge states that this is a necessary, but not sufcient condition. However it turns out that the condition given in Broadway and Burbridge is both necessary and sufcient: Theorem 2.15. In order not to impose unbalanced magnetic pull on the rotor of a machine with two different pole pair elds in the airgap, p1 and p2 , the pole pairs must be chosen so that | p1 p2 | = 1. Proof. Assume that the air gap eld, at some time instant, ti is given by:
i i B(, ti ) = B1 cos( p1 + 1 ) + B2 cos( p2 + 2 )
harmonics of order 3np, n N of the magnetic ux density and conduction distribution function
p2 such that p1 / p2 = 3k, k N, p1 > p2 . This relaxed requirement allows the usage of 6/2 pole
(2.41)
36
i i where B1 , B2 R are the amplitudes of the ux density waves, and 1 , 2 R are the phase offsets
value, must be balanced. Squaring gives:
at time ti . p1 , p2 N are the number of pole pairs. Then from Creedy [27, p. 512], the square of this
2 i 2 i i i B(, ti )2 = B1 cos2 ( p1 + 1 ) + B2 cos2 ( p2 + 2 ) + 2B1 B2 cos( p1 + 1 ) cos( p2 + 2 )
(2.42)
2 cos(A) cos(B) gives: B(, ti )2 =
Applying the trigonometric identities: cos2 (A) =
1 2
1 + 2 cos(2 A) and cos(A + B) + cos(A B) =
2 B1 B2 B2 B2 i i + 1 cos(2 p1 + 21 ) + 2 + 2 cos(2 p2 + 22 )+ 2 2 2 2 i i i i B1 B2 cos(( p1 + p2 ) + 1 + 2 ) + B1 B2 cos(( p1 p2 ) + 1 2 )
(2.43)
The only way unbalanced magnetic pull can occur is if B(, ti )2 contains a point between 0 and 2 which is not balanced by another point or points. It is reasonably straightforward to see that cos(n) (where n Z) is only unbalanced if n = 1. Therefore B(, ti )2 is balanced iff 2 p1 , 2 p2 , ( p1 + p2 ), ( p1 p2 ) = 1. As p1 , p2 N then this condition reduces to | p1 p2 | = 1.
2.6.2
Mutual inductance between two stator windings where coils groups are not in series
Early BDFM machines used a special tapping on their stator winding to avoid the need to have separate isolated 3 phase windings to produce the two pole number elds, see for example [48, 17]. However it was shown by experiment in Rochelle et. al. [88], and theoretically by Alexander [2], that connections of this nature can lead to windings where circulating currents are induced. Circulating currents lead to additional Ohmic losses, and may lead to poor iron utilisation and knock-on effects in the rest of the machine. Therefore separate 3 phase windings are now generally used. In the preceding derivation of inductance parameters for single and double layer windings, we have assumed that each winding consists of only three distinct circuits. That is, the groups of coils making up a phase are connected only in series. However it is common practice when designing 3 phase electrical machine windings to choose to connect coils within a phase in either series, parallel or some combination of the two, as suits the particular situation (often parallel connections allow a narrower wire gauge to be used which is preferred by machine winders) [91, sect. 6.4]. It will be shown, however, that even when adhering to the criteria of theorem 2.12 it is possible to get coupling between two stator windings when the coil groups are connected in parallel. This fact has been inadvertently veried by experiment on the prototype machine. Using the method outlined in section 2.5.1 the mutual inductance matrix for a series or parallel connected winding may be calculated. Appendix B.2.2 gives the details of the stator winding used in
37
the prototype machine. The windings are a 4 and 8 pole winding with all coil groups connected in series. After transformation by the combination matrix (B.8) the stator mutual inductance matrix is (without any leakage effects added): 0.210 0.0963 0.0963 0 0 0 (2.44)
0.0963 0.210 0.0963 0 0 0 0.0963 0.0963 0.210 0 0 0 Mss = 0 0 0 0.217 0.0974 0.0974 0 0 0 0.0974 0.217 0.0974 0 0 0 0.0974 0.0974 0.217 Recall that: Vs1 Vs2 = Mss di s dt
(2.45) (2.46)
di s 1 Vs1 = Mss dt Vs2
where Vs1 , Vs2 are the stator 1 and 2 supply voltage vectors, i s is the stator current vector. Furthermore
1 from Lemma A.5 as Mss is block diagonal as Mss is block diagonal. Therefore di s dt
may be expressed
as: Mss1 Vs1 di s = dt Mss2 Vs2 where Mss1 , Mss2 can be found from Lemma A.5. Therefore we conclude that there is no direct coupling between the two stator supplies because the rate of change of current in the rst winding is due only to stator 1 voltage supply. An alternative winding was considered for the prototype machine, which comprised of the same coil groups, however connected such that four sets of groups were connected in parallel for each phase. Details can be found in appendix B.2.2. To analyse a winding with parallel paths it is necessary to treat each parallel path as a separate circuit. Therefore the stator mutual inductance matrix is
1 hence Mss ) are not block diagonal. Performing the same analysis applied to the series connected
(2.47)
24 24, and as such too unwieldy to be printed here. However it can easily be shown that M ss (and winding we may write: di s 1 Vs1 p = Mss dt Vs2 p
(2.48)
38
where Vs11 Vs11 Vs11 Vs12 Vs12 = Vs 12 Vs 12 Vs13 Vs 13

Vs13 Vs13
s11
Vs2 p
Vs1 p
where Vs11 is the rst element of Vs1 etc. tively give:
Vs21 Vs21 Vs21 Vs22 Vs22 = Vs 22 Vs 22 Vs23 Vs 23

Vs23 Vs23
s21
(2.49)
Evaluation (2.48) for some balanced three phase voltage source of magnitudes V1 and V2 respec-
di s = dt
Therefore notice that all the coil groups are functions of both V1 and V2 . Therefore there is signicant direct coupling between the two windings. The effect of the cross coupling has been to introduce circulating currents in the other winding. This will lead to poor copper utilisation and efciency. This winding was built and experiment conrmed the presence of circulating currents. The machine was supplied from winding 1 (4 pole) with a balanced three phase supply derived from the 50Hz grid, with the rotor removed. The rotor was removed to eliminate any possibility of coupling via the rotor. The 4 pole winding was excited and allowed to reach thermal equilibrium while the 8 pole winding was left unconnected at the terminals. Once thermal equilibrium was attained then voltage, current and power were measured using a three phase digital power meter. Immediately after disconnecting the power supply the DC resistance was measured of each phase of the 4 pole winding. Table 2.1 summarises the results for both the parallel path winding and series connected winding. It can immediately be seen that an unaccounted extra loss of 29% appears in the parallel path winding whereas there is substantially no such extra loss in the series path winding, thus conrming that there
211.2 V1 cos(t)87.0 V2 cos(t)+86.8 V2 sin(t) 211.2 V1 cos(t)+87.0 V2 cos(t)86.8 V2 sin(t) 211.2 V1 cos(t)87.0 V2 cos(t)+86.8 V2 sin(t) 105.6 V1 211.2 V1 cos(t)+87.0 V2 cos(t)86.8 V2 sin(t) V2 sin(t) 105.6 V cos(t)+182.9 V1 sin(t)+118.7 V2 cos(t)+31.9 V sin(t) 1 cos(t)+182.9 V1 sin(t)118.7 V2 cos(t)31.9 2 105.6 V1 cos(t)+182.9 V1 sin(t)+118.7 V2 cos(t)+31.9 V2 sin(t) 105.6 V1 cos(t)+182.9 V1 sin(t)118.7 V2 cos(t)31.9 V2 sin(t) 105.6 V1 cos(t)182.9 V1 sin(t)31.7 V2 cos(t)118.8 V2 sin(t) 105.6 V cos(t)182.9 V sin(t)+31.7 V cos(t)+118.8 V sin(t) 1 1 2 2 105.6 V1 cos(t)182.9 V1 sin(t)31.7 V2 cos(t)118.8 V2 sin(t) 105.6 V1 cos(t)182.9 V1 sin(t)+31.7 V2 cos(t)+118.8 V2 sin(t) 200.1 V1 cos(t)83.1 V1 sin(t)+40.2 V2 cos(t) 200.1 V1 cos(t)+83.1 V1 sin(t)+40.2 V2 cos(t) 200.1 V1 cos(t)83.1 V1 sin(t)+40.2 V2 cos(t) 200.1 V1 cos(t)+83.1 V1 sin(t)+40.2 V2 cos(t) 172.0 V1 cos(t)+131.8 V1 sin(t)20.1 V2 cos(t)+34.8 V2 sin(t) 172.0 V1 cos(t)131.8 V1 sin(t)20.1 V2 cos(t)+34.8 V2 sin(t) 172.0 V1 cos(t)+131.8 V1 sin(t)20.1 V2 cos(t)+34.8 V2 sin(t) 172.0 V1 cos(t)131.8 V1 sin(t)20.1 V2 cos(t)+34.8 V2 sin(t) 28.1 V1 cos(t)+214.9 V1 sin(t)20.1 V2 cos(t)34.8 V2 sin(t) 28.1 V1 cos(t)214.9 V1 sin(t)20.1 V2 cos(t)34.8 V2 sin(t) 28.1 V1 cos(t)+214.9 V1 sin(t)20.1 V2 cos(t)34.8 V2 sin(t) 28.1 V1 cos(t)214.9 V1 sin(t)20.1 V2 cos(t)34.8 V2 sin(t)
(2.50)
2.7 BDFM Rotor Mutual Inductance terms
39
Phase
Voltage (V)
Current (A)
Power (W)
I2R (W)
Diff. (W)
4 pole winding with parallel paths R = 1.77 R Y B 26.9 26.9 25.8 9 9 8.67 200 200 186 143 143 133
1
57 57 53
4 pole winding with series paths R = 3.4 R Y B 47.1 46.9 46.9 8.95 8.83 8.75 270 264 263
272 265 260
2 1 3
Table 2.1: Power dissipated as measured by a three phase power meter when the 4 pole winding of prototype machine described in appendix B.2 is excited with the 8 pole winding open circuit. The results compare the difference between series and parallel connection of the 4 pole winding. are circulating currents induced in the 8 pole winding by the 4 pole winding when the windings are implemented with parallel coil groups. Remark 2.16. The previous example leads to a general statement of caution: Theorem 2.12 gives rules for choosing non-coupling stator winding pole numbers. However non-coupling is only guaranteed when the coils groups are connected in series, furthermore it is easy to nd examples of parallel connected coil groups which lead to circulating currents being directly induced. Therefore it cannot be assumed a priori that parallel connection of coils groups will lead to a satisfactory winding for the BDFM.

Having considered mutual inductance parameters for realistic stator windings it is pertinent to consider the mutual inductance parameter of potential BDFM rotors. Although section 2.5.1 provides a method of calculating rotor-rotor and rotor-stator mutual inductance matrices it will be of benet to derive some analytical results. As BDFMs are not used commercially, there is not a standard BDFM rotor class, as there is for conventional induction machines. The most commonly adopted rotor design for the recent prototypes described in the literature (see table D.1) is the nested-loop rotor design, conceived by Broadway/Burbridge [17]. It is generally accepted that, at the current stage of BDFM development, the rotor design cannot
1 This experimental data was obtained by Xiaoyan Wang.
40
be considered nalised [109], however it will be argued in chapter 5 that all realistic BDFM rotors will share the same general form of the mutual inductance matrix. The general type of rotor design, which will be considered, falls within the following assumptions: 1. Each rotor circuit conductor distribution function must contain p 1 and p2 space-harmonic terms, so that it couples to the stator. 2. Rotors must have (or can be modelled as having) S = p 1 + p2 identical circuits spaced evenly around the circumference. As has generally been assumed so far, the circuits must be constructible from coil groups of the form illustrated in gures 2.2 and 2.3. 3. The rotor may have multiple sets of circuits as described in 2, each set may be different from the other sets, and each set of circuits may be positioned at any angular position around the rotor. For example if p1 = 2, p2 = 4 then S = 6, so the rotor must have 6 identical circuits evenly
spaced around the circumference. Further sets of 6 may be added as desired. The nested-loop rotor class of rotors is considerably broader.
design, in all its variants, clearly falls into this category as can be seen from gure 2.4, although this It should be noted that this class of rotor designs does not guarantee that the rotor will crosscouple and allow the machine to operate in synchronous mode. Chapter 5 presents results for a rotor design (rotor 3) within this rotor class which does not cross couple. Essentially the reason for this is that cross-coupling requires not only that the conductor distribution function contains the correct space harmonic terms, but also that the spatial distribution of rotor coils must be such that harmonic eld components of the right frequency are produced. This issue is discussed further in chapter 5.
2.7.1
Rotor-rotor mutual inductance matrices
Firstly we consider a rotor consisting of only the outer loops of each nest (see gure 2.4), that is a rotor with only 6 circuits. Clearly the self-inductance will be the same for every loop, and furthermore the mutual inductance between one loop and all other loops will always be the same, regardless which loop is considered. The self and mutual inductances can be calculated directly from (2.27), for appropriate parameter choices. If instead we consider a slightly more general rotor structure where there can be overlap between adjacent nests, then the mutual inductances will differ. The most parameters, where oor denotes round down. Without loss of generality it will be assumed that adjacent elements in the mutual inductance matrix correspond to parameters for adjacent nests. In general case possible for S = p1 + p2 circuits will have Q = oor(S/2) different mutual inductance
41
Type 1
Type 2
Rotor loops
Nest
Rotor bars
PSfrag replacements
Type 3
2 p1 + p2
End ring
Rotor core
Figure 2.4: Three variant nested loop rotor designs, the difference between the designs is essentially one of ease of fabrication [33, 115]
42
this case the rotor-rotor mutual inductance matrix for Mrr circuit is: Lr Mr1 Mr2 Mr Q1 Mr Q Mr Q1 Mr Lr Mr1 Mr2 Mr Q1 Mr Q 1 Mr2 Mr1 Lr Mr1 Mr2 Mr Q1 Mr2 Mr1 Lr Mr1 Mr2 .. .. .. .. .. .. ... . . . . . . Mr = M Mr2 Mr1 Lr Mr1 r Q Mr Q1 Mr Q1 Mr Q Mr Q1 Mr2 Mr1 Lr Mr Q1 Mr Q Mr Q1 Mr2 Mr1 M Mr Q1 Mr Q Mr Q1 Mr2 r2 Mr1 Mr2 Mr Q1 Mr Q Mr Q1 where all the mutual terms are equal.
Mr Q1 Mr Q Mr Q1 .. . Mr2 Mr1 Lr Mr1 Mr2
Mr2 Mr Q1 Mr Q .. . Mr2 Mr1 Lr Mr1
Mr1
and mutual inductance matrix of the outer loops of the rotor of gure 2.4, is a special case of Mrr
Mr2 Mr Q1 .. . (2.51) Mr Q1 Mr2 Mr1 Lr
If the rotor in question has N sets of S circuits, such as the rotor shown in gure 2.4, which has 3 sets, then the total mutual inductance matrix can always be written as: Mrr 1 Mrr 12 Mrr 1N Mrr T Mrr Mrr 23 12 2 Mrr = . . . . . . . . . . . . Mrr 1N T Mrr (N 1)N T Mrr N
(2.52)
where Mrr i R SS is the rotor-rotor mutual inductance matrix for the i th set of S circuits, and denition A.1 (2.51) is a symmetric circulant matrix. An example of a rotor-rotor mutual inductance rotor described in section 5.2.1. Mrr i j RS S is the mutual inductance between the i th and j th sets. It should be noted that from
matrix is given in equation (B.12), which is the inductance matrix for the prototype nested-loop
2.7.2
Rotor-Stator mutual inductance matrix
Firstly we consider the mutual inductance between a single stator phase and single rotor circuit. A phase of the stator winding will have a magnetic ux density of the form of (2.39). As the rotor circuit will be formed out of a series connection of coil groups, the conductor distribution function will be the summation of individual coil conductor distribution functions given by (2.26). The resulting mutual inductance can be calculated from the derived expression for mutual inductance, equation (2.4.2). From the Lemma 2.4, we know that the mutual inductance between a single phase and a single rotor circuit can only contain harmonics common to both stator and rotor coils. Therefore assuming the stator winding is a single or double layer winding, as previously described, then we know that the mutual inductance can contain no more harmonic terms than are present in the expression for the stator magnetic eld (2.39), that is odd multiples of stator pole-pairs.
2.8 Effect of Slotting on Mutual inductance terms
43
Hence the mutual inductance between one phase of a p 1 pole-pair stator and a single rotor circuit, where M is the coefcient for the n th harmonic, r is the rotor position, 1 is an offset angle, and from the next by 2/S radians, where S = p1 + p2 . Furthermore each stator phase is, of course,
2 3 p1
regardless of the particulars of the circuit itself, can be written M sr =
nQ
M(n) cos((r 1 )n),
Q = p1 (2k 1), k = {1, 2, 3, }. As discussed in the previous section each rotor circuit is offset radians for stator 1 and
2 3 p2
offset from the previous by
for stator 2. Hence we may write down
the rotor-stator mutual inductance matrix:

nQ 1 nQ 1 nQ 1
Msr =
M1 (n) cos((r
M1 (n) cos((r 1 )n)

2 3 p1 4 3 p1 2 3 p2 4 3 p2
nQ 1 nQ 1 nQ 1
M1 (n) cos((r
1 )n) 1 )n) 2 )n) 2 )n)
M1 (n) cos((r M1 (n) cos((r
M1 (n) cos((r
2 S 2 S
2 S

2 S
2 3 p1 4 3 p1 2 3 p2 4 3 p2
1 )n)
1 )n) 1 )n) 2 )n)
nQ 2 nQ 2 nQ 2
M2 (n) cos((r
M2 (n) cos((r
M2 (n) cos((r 2 )n)
nQ 2 nQ 2 nQ 2
where Q 1 = p1 (2k 1), Q 2 = p2 (2k 1), k = {1, 2, 3, }, S = p1 + p2 , p1 , p2 N are the harmonic of mutual inductance, and 1 , 2 R are offset angles between the rst rotor coil and the As with the rotor-rotor mutual inductance the rotor may contain additional sets of S identical, equally spaced, rotor circuits. In the case that the stator-rotor mutual inductance matrix has N such sets it becomes: Msr = Msr 1 by (2.53) for a suitable choice of parameters. Msr 2 Msr N
2 )n) M1 (n) cos((r (S1)2 1 )n) nQ 1 S 2 M1 (n) cos((r (S1)2 3 p1 1 )n) nQ 1 S (S1)2 4 3 p1 1 )n) nQ 1 M1 (n) cos((r S (2.53) M2 (n) cos((r (S1)2 2 )n) nQ 2 S (S1)2 2 3 p2 2 )n) nQ 2 M2 (n) cos((r S (S1)2 4 3 p2 2 )n) nQ 2 M2 (n) cos((r S
M2 (n) cos((r M2 (n) cos((r
M2 (n) cos((r
2 S 2 S
2 )n)
number stator winding 1 and 2 pole pairs, M1 (n), M2 (n) R are the coefcients for the n th absolute
rst phase of stator winding 1 and 2 respectively.
(2.54)
where Msr i is the mutual inductance of the i th set of S equally spaced circuits on the rotor and is given
2.8 Effect of Slotting on Mutual inductance terms

Hitherto all magnetic ux density and mutual inductance calculations have been derived assuming that the stator and rotor are smooth cylinders, i.e. without any slots. The BDFM will generally have a slotted stator and rotor to reduce leakage inductance.
44
The effect of slotting is to change the prole of the air gap from a uniform ring of constant width, to one where the width varies depending on whether or not the stator and rotor teeth align or not. It is expected, therefore, that the air gap magnetic ux density will change as a result of slotting. Under the stated assumptions the ux density was shown to be inversely proportional to the air gap width, so if the air gap width varies, then it is reasonable to expect the ux density to vary accordingly with a ripple pattern related to position of the slots. A detailed analysis of slotting effects is beyond the scope of this work, [43, chap. 6] includes a detailed discussion. However a signicant effect of slotting which cannot be ignored is that the effective width of the air gap is increased, on average. This will then scale the mutual inductance terms. The fractional increase in effective air gap width can be approximated by Carters Factor. Henceforth it will be assumed that the air gap width used is the effective air gap width. Appendix C.2 gives details of the calculation method with references. As noted in [43, sect. 6.11] a signicant effect of slotting, which has been ignored, is its effect on higher harmonic components of mutual inductance (differential leakage in the vocabulary of the reference).
2.9 BDFM Coupled-Circuit Model

The BDFM coupled circuit model was rst presented in [110] and [63]. The model will be presented here, using previously dened notation. In the BDFM it is convenient to partition v and i into stator 1, stator 2 and rotor quantities, noting that the rotor voltage will always be zero. Note that the subscripted elements are still, themselves, vectors: i s1 i s2 ir
vs1
vs1
it should be noted that to practically deliver supply voltages of v s1 and vs2 will, in general, require 4 wire (3 phases plus neutral) connections in each case. In partitioning the voltage and current it has been shown that for any well-designed BDFM: 1. The mutual inductance between stator 1 and stator 2 is zero by design, see section 2.6.2 for details. 2. The mutual inductance matrices, M12 = M21 T , are functions of the rotor angle, as described in section 2.5.1.
vs2 = vs2 , 0 vr
(2.55)
2.9 BDFM Coupled-Circuit Model
45
3. The mutual inductance matrices Ms1 , Ms2 , Mr are all constant as they link circuits which are not moving relative to one another, see section 2.5.1 for further details.
Therefore substituting (2.55) into (2.6): 0

T d Ms1r dr
vs1
Rs1
0 Rs2 0
0 Rr
0 0
T d Ms2r dr
vs2 = 0 0 0
0 + r 0
d Ms1r dr d Ms2r dr
i s1 Ms1 i s2 + 0 T ir Ms1r 0
0 Ms2
T Ms2r
di Ms2r dt Mr
Ms1r
(2.56)
The torque equation, from (2.12) and (2.56): i s1 T 0 i s2 0 ir T T d Ms1r d Ms2r ir 0 dr dr i s1 T d Ms2r T M T M irT dr i s1 d ds1r + i s2 d ds2r i s2 r r ir 0 0
d Ms1r dr d Ms2r dr
Te =
1 T i iT 2 s1 s2
= =
T 1 T d Ms1r i 2 r dr
T 1 T d Ms1r d MT T M T M ir dr i s1 + irT ds2r i s2 + i s1 d ds1r ir + i s2 d ds2r ir r r r 2 T M T M = i s1 d ds1r ir + i s2 d ds2r ir r r
T i s1
T i s2
d Ms1r dr d Ms2r dr
ir
(2.57)
At this stage it is convenient to note the mechanical differential equation:
dr = Te Tl dt
(2.58)
where J is the combined moment of inertia of the BDFM and load, and Tl is load torque exerted on the BDFM. Frictional forces are neglected. Stacking equations (2.58), (2.5) and (2.56) gives a complete state-space representation of the dynamics of the BDFM, with the currents in each individual circuit, the position and speed forming the state vector:
46
i s2 d i = dt r r r M 0 s1 0 Ms2 T T Ms1r Ms2r
i s1
1 Rs1 0 Ms2r M 0 Ms1r

r
0 Rs2 0
0 Rr
0 r 0 r
d Ms1r dr d Ms2r dr
0
T d Ms1r dr
0 0
T d Ms2r dr
d Ms1r dr d Ms2r dr
1 J
T i s1
T i s2
ir Tl
i s1 vs1 i s2 + vs2 i 0 0 r
(2.59)
note that the inverse is guaranteed to exist from Lemma 2.3. It is signicant to note that the BDFM is a non-linear parameter-varying system: while the electrical equations are linear in i, they depend on the parameters r and r , which are themselves states of the full system. The dependence on r is particularly problematic to the control engineer because normal machine operation necessitates a non-zero rotor speed, and therefore r varies rapidly. The torque equation is quadratic in i. Using equation (2.59), with parameters calculated as described earlier in the chapter a dynamic model was implemented in Simulink. Simulink is an environment for modelling systems of differential equations, solved by a variable time-stepping procedure, in this particular case a variable step version of the 4th order Runge-Kutta method was used, for details see [108]. For ease of implementation, and to maximise running speed, the position dependent parameters M s1r , Ms2r and their derivatives were computed off-line at 1 intervals and then values interpolated online. Figure 2.5 shows the implementation of the coupled circuit BDFM model of equation (2.59) implemented in Simulink. The block described at Mutual inductance matrix calculation contains a subsystem which implements the required interpolation of the rotor-stator mutual inductance terms, and their derivatives. The same implementation was used for all the BDFM designs considered in this dissertation, with different designs simply requiring different parameter values. As an example of the simulation speed, the simulation of a BDFM with a nested-loop design rotor (described in chapter 5 and appendix B as rotor 1) for 6s took 8.1s of CPU time on a 2.8GHz Pentium 4. The system had 26 states: 6 stator currents, 18 rotor currents and two mechanical states. The model derived, although relatively simple to implement in simulation, provides little insight into the behaviour of the machine, and there is therefore a strong motivation to transform the system into a simpler form. This work forms the basis of the next chapter.
2.10 Conclusions
47
current vector I
state vector 1 Vs1 2 Vs2
thetar_r omega_r
stator voltage
0 Rotor voltage
Inv R Resistance matrix

M
Matrix Multiply Matrix Product
thetar dMsrdtheta
Mutual inductance matrix interpolation
uT
Matrix Multiply
state derivative
1 s Integrator
1 State output
Matrix Multiply 3 Load Torque
uT
Matrix Multiply
1/J Moment of inertia
Figure 2.5: Block diagram showing the implementation of the coupled circuit BDFM model, equation (2.59), in Simulink
2.10 Conclusions
In this chapter a method of calculating machine inductance parameters which is sufciently general to cover any rotor or stator winding conguration, comprising of series connected groups of coils, has been presented. The method calculates the self and mutual inductance for each coil group, and then by choice of combination matrix, each series connected set of coils is combined into a machine parameter. Furthermore, Carters factor was used to approximate the effect of slotting on the mutual inductance parameters, and leakage inductances were estimated using standard methods applied on a coil by coil basis. The proposed method is straightforward to implement on a computer, and allows a general interface for different machine windings or rotor designs. It was argued that, because the assumptions made in the application of this modelling method are common-place, and have substantial experimental backing in different modelling approaches, the method proposed in this chapter is reliable and will give similar performance to other modelling techniques based on the same assumptions. Precise conditions were derived for the non-coupling of all harmonics of two stator windings, and a proof given for conditions to avoid unbalanced magnetic pull. It was shown by counter example, that it cannot be assumed that a winding with non-series connections of coil groups will meet the non-coupling requirements for BDFM operation. A broad class of BDFM rotors was dened, and the mutual inductance for the rotor-rotor and rotor-stator determined using the method previously described.
48
Using the derived parameters, a coupled-circuit model specic to the BDFM was derived from the general electrical machine equations. This model can accurately model the transient (and steadystate) performance of any BDFM. The model includes the effect of all space-harmonic components of inductance parameters, however no attempt is made to take any saturation effects into account.
Chapter 3
d-q Transformed Model

3.1 Introduction
In chapter 2 a model was derived for the BDFM, and a method of calculating the values of the mutual inductance parameters was presented. The derived coupled-circuit model has position varying parameters, which means that it is not possible to solve the equations analytically for any condition other than with the rotor at standstill. However, it will be shown that if the model is transformed into the rotor reference frame, that is a reference frame which rotates in synchronism with the rotor, then the position dependency vanishes. Such a transformation will be shown to be achievable by an invertible state transformation of the model equation (2.59). The transformation adopted is the so-called d-q-0, direct-quadrature-zero transformation. It will be shown that the zero sequence part of the transformation is of less interest, and thus the transformation is often referred to as simply d-q transformation. Li et. al [64, 63] rst showed that a d-q transformation could be applied to a BDFM. Their work was generalised to any pole pair conguration BDFM in Boger et. al. [11, 12]. However nowhere in the literature has there been a rigorous proof of why the transformation is effective. In fact, in both papers non-invertible state transformations are used without explanation (the use of a non-invertible state transformation implies a change to the model dynamics). It will be shown that this is of particular consequence when modelling nested-loop rotors. Kemp et al. presented a variation of the model found in [64, 63] for their prototype 6/2 conguration nested-loop design rotor BDFM which does, in fact, retain all observable dynamics [53]. We prove this fact as the model due to Kemp is a special case of that described in this chapter. Furthermore the generalised transformation given in [11, 12], starts from an idealised machine, rather than explicitly transforming a more general model, where the implicit assumptions become more explicit. In addition, all the d-q-0 BDFM models in the literature are derived for nested-loop rotor BDFMs.
49
50
In this chapter a new derivation, with full proofs, for the d-q-0 transformation is presented. The approach adopted is to transform the general coupled-circuit model derived in chapter 2, with all space-harmonic components retained in the stator-stator and rotor-rotor parameters. This approach allows a large class of BDFM congurations to be incorporated in this model. The issue of how to accurately model a nested-loop rotor will be addressed, using a reduced order model. All the assumptions described in chapter 2 are maintained, including the assumption of 2 double or single layer balanced three phase windings of appropriate pole numbers, as described in section 2.6, and a rotor as described in section 2.7. In order to improve readability, the majority of the Lemmas used to prove the results are to be found in appendix A, and referenced as appropriate from within the text. While some results in appendix A are standard results collected for convenience; others, particularly Lemmas A.14, A.15, A.16, A.17, A.18 and A.19 are believed to be original.
3.2 The d-q-0 state transformation matrix

We start by giving a physical interpretation of the d-q axis transformation. Figure 3.1, shows an idealised stator winding where the red, yellow and blue coils represent the three phases, each offset from the other by 120 . Two additional coils are shown in gure 3.1, representing ctitious coils on the d and q axes respectively. We now consider how the d and q axis coils relate to stator phases. q-axis, = iq ia
2
PSfrag replacements ic id d-axis, = 0
ib
Figure 3.1: A pictorial representation of the d-q transformation, showing three coils representing the phases of a stator winding, and two ctitious coils on the d and q axes by which the stator coils will be represented. The d and q axes have been aligned (arbitrarily) to = 0 and =
2
respectively.
As we have assumed that the machine windings are balanced, the coils representing the three
51
phases must be physically identical to one another, but positioned differently within the machine. To facilitate the explanation we assume that each stator coil produces a perfect 2-pole eld in the air gap (this assumption does not apply to the BDFM analysis following). Therefore if instantaneous currents, i a , i b and i c ow in the stator coils, then the magnetic ux density distributions due to each of the coils are Ba () = i a k cos(), Bb () = i b k cos( + the angular position around the air gap.
2 ), 3
Bc () = i c k cos( +
4 ), 3
where is
We rstly consider the ux due to the red phase coil. We want to nd a way of producing the same ux density using coils aligned to the d and q axes, as shown in gure 3.1. It turns out that if the d and q axis coils are physically the same as the individual phase coils (but differently aligned), then the same ux density, Ba (), can be achieved by exciting the d and q axis coils with currents cos()i a and sin()i a respectively. To show this we consider the magnetic ux density in the air gap due to the d and q coils: Bd () = i d k cos( + ) Bq () = i q k cos( + ) = i q k sin( + ) Now if i d = cos()i a and i q = sin()i a , then Bd ()+Bq () = cos()i a k cos(+)+sin()i a k sin(+ ), which may be combined using a cosine trigonometric identity to give B d ()+ Bq () = i a k cos(). Therefore the air gap ux density produced by current i a in the red phase, i b in the yellow, and i c
q axis current of i d = i a sin() + i b sin( 2 ) + i c sin( 4 ). Therefore the d and q axis currents 3 3 can be calculated by resolving the phase currents onto the d and q axes. This is the physical basis of the d-q transformation. , the angle of the d axis relative to the individual phase, is entirely arbitrary, and can be any differentiable function. It is this freedom to rotate the d-q axis reference frame to any angle that has the useful property that the position-varying mutual inductance terms found in electrical machines may be represented as constant inductance terms by suitable choice of reference frame (i.e. the value of ). Typically aligning the reference frame so that it rotates in synchronism with the rotor achieves the desired effect [56, ch. 3] However the transformation as is stands is not a 1 to 1 mapping. That is, there is an extra degree of freedom in the 3 phase currents, which does not affect the d and q axis currents. This is the zero sequence current. The zero (0) sequence current has the interpretation of resolving the currents which produce magnetic elds which do not move parallel to the air gap, that is non-rotating elds. See Krause [56, ch. 3] for further details. It can be shown that the zero sequence current is proportional to the sum of the instantaneous three phase currents. The full transformation from phase currents to d axis, q axis and 0 sequence currents may be to as the d-q-0 transformation. expressed as an invertible 3 3 matrix [56, ch. 3]. In the sequel this transformation will be referred
in the blue may be achieved by a d axis current of i d = i a cos() +i b cos( 2 ) +i c cos( 4 ) and 3 3
52
The d-q-0 transformation has been interpreted as an equivalent representation of the air gap magnetic ux density. However the same transformation may also be applied to the phase voltages, to give equivalent d and q axis and 0 sequence voltages. Additionally the d-q-0 transformation may be chosen so that it is orthogonal. This has the physical advantage that it becomes a power conserving transformation [52], in other words, the product v T i is the same whether v and i are expressed as d-q-0 or actual phase quantities. The d-q-0 transformation may be used to transform the system into a reference frame which rotates in synchronism with the rotor by aligning the d and q axes so that they track the rotor position. It is important to note that while the physical explanation of the d-q-0 transformation assumed coils giving a harmonic-free air gap eld, no such assumption is required to use the d-q-0 transformation. Harmonic effects will simply lead to different values of the d-q-0 currents. For a stator winding with p1 pole pairs [56] gives the d-q-0 transformation as: 2 4 cos( p1 r ) cos( p1 (r 3 p1 )) cos( p1 (r 3 p1 )) 2 sin( p1 r ) sin( p1 (r 2 )) sin( p1 (r 4 )) Cs1 = 3 p1 3 p1 3
1 2 1 2 1 2
(3.1)
for the second stator winding with p2 pole pairs, the transformation is: 4 2 cos( p2 r ) cos( p2 (r 3 p2 )) cos( p2 (r 3 p2 )) 2 sin( p2 r ) sin( p2 (r 2 )) sin( p2 (r 4 )) Cs2 = 3 p2 3 p2 3
1 2 1 2 1 2
(3.2)
It is noted at this point that some references (e.g. [12]) include a phase offset angle in the second
stator transformation matrix. This is because the orientation of the d-axis can be aligned with the stator eld without loss of generality (changing the zero position simply changes the physical position of r = 0). However the second stator eld cannot, in general, be assumed to be physically aligned with the stator slots). In this work it has been chosen not to modify the transformation matrices. This means that some mutual inductance parameters will contain a phase offset term. In order to transform the rotor into d and q axis components a different transformation matrix is needed. Boger et. al. [11] give the p1 pole pair rotor transformation matrix for a nested-loop design rotor with a single set of loops as: p 1 cos(0) cos( p2 pp2 ) cos( 22 p1 ) cos( 2( p1 ++2 1) p1 ) p1 + p 2 p1 p 2 1+ 2 sin(0) sin( 2 p1 ) sin( 22 p1 ) sin( 2( p1 + p2 1) p1 ) Cr 1 = p1 + p 2 p1 + p 2 p1 + p 2 p1 + p2 1 1 1 1 2 2 2 2 the stator 1 eld (as this would place a constraint on where the two windings are physically placed in
(3.3)
they also give a similar transformation for p2 pole pairs, however we will show that it is not necessary to use this transformation. In fact either the transformation for p 1 or p2 pole pairs can be used.
53
The transformation proposed by Boger, is a reasonable generalisation of the transformation matrix for the stator phases. The transformation matrix resolves each rotor loop into d and q axis components, and the bottom row is a zero sequence component. However unlike the stator transformation, the rotor transformation matrix is not square. This means that the transformation does not simply constitute a change of variables, rather it reduces the order of the model, and therefore will result in model which is a (possibly poor) simplication of the original model. In order to ensure that the model dynamics are unchanged, the transformation must be a similarity transformation, that is one which is invertible (and therefore square, and full rank) [131, p. 53]. We will therefore dene such a transformation for the rotor. span the orthogonal complement to the row space of C r 1 exists. We denote this matrix Cr . 1 rotor transformation matrix for a single set of loops is therefore: Cr = Cr 1 Cr 1 (3.4) From Lemma A.10 Cr 1 Cr 1 T = I . From Lemma A.12 a matrix whose rows are orthonormal and Using Cr we may augment original transformation matrix (3.3), to one which is invertible. The 1
note that not only is Cr full rank (and hence invertible), but is also orthogonal (as the rows of both Cr 1 and Cr are orthonormal). As for the stator transformation matrix orthogonality implies that the 1 transformation is power conserving. rotor, then the full rotor d-q-0 transformation becomes (assuming rotor state order in accordance with section 2.7.1): C r1 Cr = 0 0 Cr 1 0 .. 0 . 0 0 Cr N Cr N (3.5) When the rotor has N sets of S = p1 + p2 circuits, such as is the case with the nested-loop design
where each matrix on the diagonal is the same, the differing subscripts are simply to facilitate bookkeeping, that is Cr 1 = Cr 2 = = Cr N . Note that because (3.4) is orthogonal so is (3.5). later on, as follows: At this point we also dene Cr 1dq R2S , Cr 10 R1S , Crdq R2N N S , which will be useful Cr 1dq C r 10 Cr 1dq = 0 0 = Cr 1 0 .. . 0 0 Cr 1dq 0 (3.6)
Crdq
(3.7)
54
3.3 Transformation to d-q-0 axes

We will use the transformation described in the previous section to transform the coupled-circuit model derived in chapter 2, with stator windings fullling the requirements of section 2.6 and rotor design compliant with the assumptions outlined in section 2.7. One additional assumption will be required in the derivation: the stator-rotor (and rotor-stator) mutual inductance matrix, given by equation (2.54) can be adequately approximated by the rst nonzero term of the series of harmonic inductances. As the rotor must fall within the assumptions of
th th section 2.7 we need the p1 and p2 absolute harmonic for the stator 1-rotor and stator 2-rotor coupling
respectively. It is signicant that, in contrast to [11], we do not place any further constraint on the rotor-rotor and stator-stator mutual inductance terms. In [11] the stator-stator, and rotor-rotor mutual inductance are the fundamental (relative) harmonic terms from the Fourier series representation only. From section 2.9 and particularly equation (2.56), the BDFM electrical equations can be written as:
vs1
vs2 = 0 0 0
Rs1
0 Rs2 0
Ms1 i s1 d 0 i s2 + dt 0 T Ms1r ir Rr 0
0 Ms2
T Ms2r
i s1 Ms2r i s2 ir Mr Ms1r
(3.8)
where the terms are dened in section 2.9. From (3.1), (3.2), and (3.5) we may dene the overall d-q-0 transformation for the BDFM: Cs1 0 0 C = 0 Cs2 0 0 0 Cr note that as C s1 , Cs2 , Cr are all orthogonal, C is also orthogonal, and hence invertible. We also dene a non-square transformation, C ns R6+2N 6+N S . From (3.7): Cs1 0 0 = 0 Cs2 0 0 0 Crdq
(3.9)
Cns
(3.10)
where N is the number of sets of S = p1 + p2 rotor circuits.
Using (3.9) we may now dene the d-q-0 transformed currents, through an invertible state trans-
formation of the currents in machine variables. The transformed stator currents are denoted i dq0s1 , i dq0s2 . The transformed rotor currents we denote by i tr r , and we will later show that i tr r can be partitioned into d-q-0 components and other components:
55
Cs1 i dq0s1 i dq0 0 s2 i tr r 0 Cs1 i s1 i s2 = 0 0 ir
i s1 Cs2 0 i s2 0 Cr ir T i dq0s1 0 0 Cs2 0 i dq0s2 i tr r 0 Cr 0 0
(3.11)
(3.12)
Substituting (3.12) into (3.8) and pre-multiplying by C gives: Cs1 vs1 Cs1 0 0 0 Cs2 0 vs2 = 0 0 0 0 0 Cr Cs1 0 0 0 Cs2 0 d dt 0 0 Cr T i dq0s1 Cs1 0 0 Cs2 0 0 Rs2 0 0 Cs2 0 i dq0s2 + i tr r 0 0 Cr 0 0 Rr 0 Cr T Ms1 0 Ms1r Cs1 0 0 i dq0 s1 0 0 Cs2 0 i dq0 (3.13) Ms2 Ms2r s2 T T Ms1r Ms2r Mr i tr r 0 0 Cr 0 0 Rs1 0 0
Dening the d-q-0 transformed voltages in a similar manner gives: vdq0s1 vs1 Cs1 0 0 0 Cs2 0 vs2 vr 0 0 Cr
vdq0 s2 vtr r
(3.14)
Substituting (3.14) into (3.13) and simplifying gives:
i dq0s1 Cs1 Rs1 Cs1 T 0 0 i dq0 + vdq0 = 0 0 Cs2 Rs2 Cs2 T s2 s2 i tr r 0 0 Cr Rr Cr T vtr r Ms1 Cs1 T 0 Cs1 0 0 d 0 Cs2 0 0 Ms2 Cs2 T dt T T Ms1r Cs1 T Ms2r Cs2 T 0 0 Cr vdq0s1 stant:
i dq0s1 Ms2r Cr T i dq0s2 (3.15) T i tr r Mr Cr Ms1r Cr T
Further simplication and use of the chain rule gives, noting that C r , Ms1 , Ms2 , Mr are all con-
56
i dq0s1 Cs1 Rs1 Cs1 T 0 0 i dq0 + vdq0 = 0 Cs2 Rs2 Cs2 T 0 s2 s2 i tr r 0 0 Cr Rr Cr T vtr r d d Cs1 Ms1 dr (Cs1 T ) 0 Cs1 dr (Ms1r )Cr T i dq0s1 d d r 0 Cs2 Ms2 dr (Cs2 T ) Cs2 dr (Ms2r )Cr T i dq0s2 + d d T T 0 i tr r Cr dr (Ms1r Cs1 T ) Cr dr (Ms2r Cs2 T ) i dq0s1 Cs1 Ms1 Cs1 T 0 Cs1 Ms1r Cr T d 0 Cs2 Ms2 Cs2 T Cs2 Ms2r Cr T i dq0s2 (3.16) dt T T T T T i tr r Cr Ms1r Cs1 Cr Ms2r Cs2 Cr Mr Cr vdq0s1 Therefore from Lemmas A.15 and A.16: Cs1 Ms1 0 p1 (L 1 M1 ) 0
We have already established that Ms1 , Ms2 are symmetric circulant matrices (see remark 2.11).
where M1 , L 1 , M2 , L 2 are the mutual and self inductance parameters of the rst and second stator winding respectively as in remark 2.11. individual circuits. S is xed for a particular pole pair choice, but N may be chosen by the designer. Ms1r Ms2r Ms1r 1 Ms1r 2 Ms1r N As discussed in section 2.7, the class of rotors considered here have N sets of S = p 1 + p2
d (Cs1 T ) = p1 (L 1 M1 ) 0 0 = Q dq0s1 dr 0 0 0 0 p2 (L 2 M2 ) 0 d Cs2 Ms2 (Cs2 T ) = p2 (L 2 M2 ) 0 0 = Q dq0s2 dr 0 0 0 L 1 M1 0 0 = Mdq0 Cs1 Ms1 Cs1 T = 0 L 1 M1 0 s1 0 0 L 1 + 2M1 L 2 M2 0 0 = Mdq0 Cs2 Ms2 Cs2 T = 0 L 2 M2 0 s2 0 0 L 2 + 2M2
(3.17)
(3.18)
(3.19)
(3.20)
Consequently, from section 2.7, the rotor-rotor and stator-rotor mutual inductances can be written as: Ms2r 1 Ms2r 2 Mr1 Mr12 M T M r2 r Mr = .12 . . . . . Mr1N T = Ms2r N Mr1N . . . . .. . . . Mr N (3.21)
(3.22)
57
where Ms1r j , Ms2r j R3S , Mr ji R SS .
Now, it is immediate from the choice of rotor class that the diagonal block matrices of (3.22) are
symmetric circulant, as explained in section 2.7.1. However, after some consideration it can be seen that this implies that the off-diagonal block matrices (e.g. Mr12 ) are non-symmetric circulant. This fact is most easily seen by considering the mutual inductance between a single rotor circuit of one set of S circuits and another entire set of S circuits. The geometry of each circuit within a set of S circuits is the same, the only difference being the position around the circumference of each. Therefore it is reasonably straight forward to see that the mutual inductance between two different sets of S circuits is itself circulant. Considering, rst Cr Mr Cr T , from (3.5) and (3.22): C Mr1 C r1 r1 . . = . .. . Cr 1 Cr 1
T
.. . Cr N Cr N
.. . . .
Cr Mr Cr T
Mr N
Cr N Cr N
(3.23)
As all block matrices of (3.22) are circulant, with the diagonal blocks symmetric in addition, then with the chosen form of the transformation matrix, applying Lemma A.16 to (3.23) gives: L 0 0 0 dq r 1 0 L dq r 1 0 0 0 0 L 0r 1 0 0 0 0 11 M12a M12b 0 0 M M12a 0 0 12b = 0 0 M012 0 0 0 0 12 T . . . M 12a M 12b 0 0 M12b M12a 0 0 0 0 M012 0 0
0 0 12
Cr Mr Cr T
..
. . .
L dq r N 0 0 0
0 L dq r N 0 0
0 0 L 0r N 0 (3.24)
N N
0 0 0
sequence rotor self-inductance, and a matrix of other inductance terms for the j th set of rotor circuits. Mi j k , M0i j are the mutual inductance terms between different sets of S circuits, for d-q axes and zero sequence respectively. Explicit expressions for L dq r j , L 0r j , Mi j k , and M0i j can be found in the proof
where L dq r j , L 0r j R, i j R p1 + p2 3 p1 + p2 3 are the d and q axis rotor self-inductance, the 0
58
(typically) zero.
of Lemma A.16. We use to denote terms for which the exact value is not relevant, but that are not
We now consider the stator-rotor mutual inductance terms. As an example, consider C s1 Ms1r Cr T . From (3.21) and (3.5):
Cs1 Ms1r Cr T = Cs1 Ms1r 1
Cr 1 Cr 1
Cs1 Ms1r 2
Cr 1 Cr 1
Cs1 Ms1r N
Cr 1 Cr 1
(3.25)
the incrementing subscript on the transformation matrix has been dropped as it no longer serves any purpose. The general term, Ms1r j , corresponding to the stator-rotor mutual inductance between stator 1 and the j th set of rotor circuits, and is of the form of (2.53). The same is true for the Ms2r j terms. To proceed further we must invoke the additional assumption mentioned: we will assume that Ms1r j and Ms2r j may be well approximated by the rst non-zero term of their Fourier series representations. This is a physically reasonable assumption because the mutual inductance terms for a particular term of the Fourier series are derived from the product of the stator magnetic ux density and rotor conductor distribution function (or vice-versa). As the stator winding is specically (and quite effectively) designed to produce one pole number eld only, the harmonic content will be small. Therefore the higher terms in the Fourier series for the stator-rotor mutual inductance will typically be small, and thus the error in neglecting them slight. Nevertheless the omission of these harmonic terms will idealise the machine, in that it becomes impossible for any frequency other than the supply frequencies to appear in the current waveforms in the steady state, which is at odds with physical reality where the stator currents will contain harmonic components. With this assumption, the terms in (3.25) can be simplied by application of Lemma A.17. Furthermore similar terms in (3.15) may be simplied for the same reasons by Lemmas A.17 and A.18:
59
Cs1 Ms1r Cr T = M11 sin(11 ) 0
M11 cos(11 ) M11 sin(11 ) 0 M11 cos(11 ) 0 0
Cs2 Ms2r Cr
M2 cos(21 ) M21 sin(21 ) 0 1 = M21 sin(21 ) M21 cos(21 ) 0 03D 0 0 0 M2 cos(2 N ) M2 N sin(2 N ) 0 N M2 sin(2 ) M2 cos(2 ) 0 N N N N 0 0 0 M11 p1 sin(11 ) M11 p1 cos(11 ) 0 M11 p1 sin(11 ) 0 0 0 M1 N p1 sin(1 N ) 03D
M1 sin(1 ) N N 0
M1 N cos(1 N ) M1 N sin(1 N ) 0 M1 N cos(1 N ) 0 0
03D
03D
(3.26)
03D (3.27)
Cs1
d (Ms1r )Cr T = M11 p1 cos(11 ) d 0
M1 p1 cos(1 ) N N 0
M1 N p1 cos(1 N ) 0 M1 N p1 sin(1 N ) 0 0 0
03D
(3.28)
M21 p2 sin(21 ) M21 p2 cos(21 ) 0 d T Cs2 (Ms2r )Cr = M21 p2 cos(21 ) M21 p2 sin(21 ) 0 03D d 0 0 0 M2 N p2 sin(2 N ) M2 N p2 cos(2 N ) 0 M2 p2 cos(2 ) M2 p2 sin(2 ) 0 N N N N 0 0 0
03D (3.29)
where M1 j , M2 j are the mutual inductance between stator 1, stator 2 and the j th set of rotor circuits, and 1 j , 2 j are the electrical phase offsets between stator 1, stator 2 and the j th set of rotor circuits. For a precise denition see Lemma A.17. D = S 3, where S = p1 + p2 . The terms Cr Ms1r T Cs1 T , Cr Ms2r T Cs2 T may be induced by symmetry, and from Lemma A.17,
d dr
Cs1 Ms1r and C s2 Ms2r (and hence Ms1r T Cs1 T and Ms2r T Cs2 T ) are constant with r , therefore
d dr
Ms2r T Cs2 T = 0.
Ms1r T Cs1 T =
So far the resistance matrices have not been considered. The resistance matrices, R s1 , Rs2 , Rr
will be diagonal, as long as the current path of each circuit is independent. For the stator terms this
60
will always be the case under the present assumptions, however for the rotor terms it may not be the case. If the rotor were a squirrel cage then when modelling, it is more natural to take each circuit as a loop around two adjacent bars and the appropriate parts of the end ring. In this case neighbouring circuits will share common bars. It can be shown that this has the effect of introducing off-diagonal elements in the resistance matrix (see section B.7). However in any case the resistance matrix must be symmetric and circulant from denition A.1. For the present we will assume that all the resistance matrices are circulant, thus allowing for the possibility of modelling cage rotor designs. As all the resistance matrices are circulant, from Lemma A.16 it can be seen that the resulting transformed resistance matrix will be diagonal. Using (3.17-3.20), (3.24), (3.26-3.29) in (3.16) gives:
33 0 33 vdq0 = 0 s2 vtr r 0 0 vdq0s1
i dq0s1 0 i dq0s2 + i tr r Rrdq0 0

22 1 0
0 33 r 0 33 0 0 0
33
03D 0
3D
22 0 N 0 0 22 0 N 0 0
22 0 1
33 0 22 0 1 0 0 D3 0 . . . 22 0 N 0 0 D3 0
22 1 0
03D 0 0
3D
22 0 N 0 0 22 0 N 0 0
22
03D i dq0 s1 + i 3D dq0s2 0 i tr r 03D 3D 0 03D d dt 0 3D 0 DD
22 1 0
.. .
22 0 1 0 0 D3 0 . . . 22 0 N 0 0 D3 0
0 0 22 0 0 0
22
3D DD
0 0
i dq0s1 i dq0 (3.30) s2 i tr r
03D 0
22 0 0 0
where denotes a constant (typically) non-zero element, D = S 3. Where dimensions are not given for a particular element it is either scalar or the dimension can be deduced. If we now reorder i tr r to give i tr r so the zero portion of that C s1r Ms1r Cr T and similar elements are collected together. We dene the modied rotor resistance Rrtr , signicantly as Rrtr was diagonal, Rrtr must also be diagonal. Hence:
61
vdq0 = 0 s2 0 vtr r
vdq0s1
33
0 33 0
3 3
i dq0s1 0 i dq0s2 + i tr r Rrtr 0 0 33 0 0 33 0
r 0 0
22 22 02N D 1 N 0 0
22 22 02N D 1 N 0 0 0 0 0 0
33 0 22 1 . . . 22 N 0 N D2
22 22 02N D 1 N 22 22 02N D 1 N 0 0
i dq0 s1 i dq0s2 + i tr r d dt
0 . . . 0 0
0 . . . . . . 22 0 N N D2 0 0
22 1
0 .. . 22 22 02N D N 22 22 0 0 N DN D
22 1
22
2N D
i dq0s1 i dq0 (3.31) s2 i tr r
where D = S 2.
T
T By following through rotor terms it can be seen that we can now partition i tr r into i dq r T i dq r
Where i dq r = i dq r 1 T i dq r N T , with each i dq r j corresponding to the d and q axis currents from the j th set of rotor circuits. Note that the zero sequence circuit in each case has been included in i dq
r
along with terms corresponding to the orthogonal complement of the d-q transformation, Therefore with Rdq , Mdq R N DN D (3.31) may be written as:
r r
C r
(3.4).
62
v dq0s2 = vdq r vdq r
vdq0s1
Rdq 0
0 Rdq r
+ r
Q dq 0
0 0
i dq0s1 i Mdq dq0s2 + i dq 0 r i dq r
0 Mdq r
i dq0s1 d i dq0s2 dt i dq r i dq r
i dq0s1 d i dq0s2 = Mdq i dq dt 0 r i dq r =
0 Mdq r
Rdq + r Q dq 0 0 Q dq sr 1
0 Rdq r
(3.32) v i dq0s1 dq0s1 i dq0s2 vdq0s2 + i dq vdq r r dq vdq r i r 0 Mdq0s2 Mdq sr 2 T Mdq sr 2 Mdq r Mdq sr 1 (3.33) (3.34)
Q dq0s1 0 0
Q dq
Q dq0s1 0
Mdq0s1 , Mdq = 0 Q dq sr 2 Mdq0sr 1 T 0
only measure the other currents, it is quite impossible to determine the value of i dq r . i dq r is said to be unobservable from the output.
As the inductance matrix is block diagonal, its inverse will also be block diagonal from Lemma A.5, therefore it is clear from (3.33) that i dq is independent of the other currents. And therefore if we can
r
said to be uncontrollable from the input (the supply voltage).
Furthermore as there is no voltage supply to the rotor, the rotor voltage will always be zero. It is therefore not possible to affect, in any way, the value of i dq by changing the supply voltage. i dq is
r r
It is well-known to control engineers that removing states from a system which are uncontrollable and unobservable, but unstable will inevitably lead to unsatisfactory performance. Fortunately, with the BDFM, this is not the case. If we consider the rotor-rotor mutual inductance, M r of (3.8), then from Lemma 2.3 we know it to be positive denite, and furthermore we know it to be symmetric (some denitions of positive denite do not require symmetry), thus all the eigenvalues of M r must be greater than zero. As the only transformation applied to Mr is a constant, invertible state transformation its eigenvalues will remain unchanged. The same holds true for Rr , the rotor resistance. As Mr and Rr 1 Cr can be chosen so that Rdq and Mdq are diagonal, hence Mdq Rdq is negative denite, thus all 1
r r r r
eigenvalues are negative. Hence the uncontrollable and unobservable states highlighted are stable and can safely be removed. Therefore the full d-q transformed BDFM model can be written as:
63
where Mdq0s1 is given by (3.19), Mdq0s2 by (3.20), Q dq0s1 by (3.17), Q dq0s2 by (3.18). Mdq sr 1 is given by (3.26) with the appropriate elements removed, Mdq sr 2 by (3.27), Q dq sr 1 by (3.28), Q dq sr 2 by (3.29), and Mdq r by (3.24). The resistance matrix is diagonal as previously discussed. We now turn our attention to the torque equation. From (2.12): 1 dM Te = i T i 2 dr substituting with (3.11) and simplifying gives: T T Cs1 0 0 Cs1 0 0 i dq0s1 i dq0s1 dM 1 = i dq0s2 0 Cs2 0 dr 0 Cs2 0 i dq0s2 2 0 0 Cr 0 0 Cr i tr r i tr r T d 0 0 Cs1 dr (Ms1r )Cr T i dq0s1 i dq0s1 1 d = i dq0s2 0 0 Cs2 dr (Ms2r )Cr T i dq0s2 2 d d i tr r Cr dr (Ms1r T )Cs1 T Cr dr (Ms2r T )Cs2 T i tr r 0 = i dq0s1
T
i dq0s1 i dq0s1 vdq0 = Rdq + r Q dq i dq0 + Mdq d i dq0 s2 s2 s2 dt i dq r 0 i dq r vdq0s1 i dq0s1 i dq0s1 d i dq0 = M 1 Rdq r Q dq i dq0 + M 1 vdq0 dq dq s2 s2 s2 dt 0 i dq r i dq r Mdq0s1 0 Q dq0s1 0 Q dq sr 1 Q dq = 0 Mdq0s2 Q dq0s2 Q dq sr 2 , Mdq = 0 T Mdq sr 1 Mdq sr 2 T 0 0 0 Rdq0s1 0 0 Rdq = 0 Rdq0s2 0 0 0 Rdq r vdq0s1
Mdq sr 1 Mdq r
Mdq sr 2
(3.35)
(3.36)
(3.37)
(3.38)
i dq0s2
d Cs2 dr (Ms2r )Cr T
d Cs1 dr (Ms1r )Cr T
i tr r
(3.39)
by adopting the same partition of i tr r used in (3.33): Te = i dq0s1 i dq0s2

T
Q dq sr 1 Q dq sr 2
i dq r
(3.40)
We may now make a general statement about d-q axis transformations for this class of machines:
64
Remark 3.1. For the class of BDFMs with: single or double layer isolated, series connected stator windings, as discussed in section 2.6, with suitable choice of pole pairs p1 , p2 as given by theorem 2.12, and a rotor which fulls the requirements of section 2.7; and then if the mutual inductance between stator and rotor can be adequately approximated by the rst non-zero term of their Fourier series representation; then the BDFM equations may be transformed using the non square transformation given by (3.10) which is a power-conserving transformation and removes the states corresponding to i dq r . Furthermore, the states removed by this transformation are unobservable, uncontrollable and stable, and do not appear in the torque equation. In the special case of a nested-loop design rotor, the transformation derived is identical to one of the transformations of Boger [12, p. 25].
Remark 3.2. It is signicant that the BDFM, with a rotor within the class outlined, can have its rotor modelled using a single d-q pair for each set of S rotor circuits. This is somewhat surprising because the machine has two stator supplies, of different pole numbers, hence it would be expected that two d-q pairs would be required for each set of rotor circuits. Or in systems theory terminology one would expect matrix M, of Lemma A.17 to be of rank 4, when it is actually rank 2. This fact serves number. to reinforce the necessity for the rotor to have S = p1 + p2 circuits per set, rather than any other It it worth mentioning that the d-q transformation outlined may be extended for higher order harmonic couplings between stator and rotor. It can be shown that the stator parameters will remain the same for all higher harmonics, however the rotor parameters will change for some harmonic combinations. This is because the special condition which allowed a single d-q pair to represent a single set of S rotor circuits, as noted in remark 3.2 will only be met if (k1 p2 + k2 p2 )/( p1 + p2) is an integer where k 1 , k2 are the relative harmonic for each winding. This has the effect that the higher harmonics may require 2 d-q pairs for each set of S circuits, rather than 1, and consequently the rotor-rotor portion of Mdq will alter. For each harmonic the stator-rotor parameters will change.
3.3.1
Determination of d-q model rotor current from bar currents
The problem of determination of d-q model rotor current from bar currents will briey be considered. Clearly the d-q axis rotor current can be computed using (3.5), and then truncated to leave the nonzero portion, as discussed in the previous section. is controllable from the stator voltages. Therefore it is possible to calculate the d-q rotor currents for each set of S rotor circuits by measuring the currents in only two of the S circuits. For each of the N sets of S rotor circuits the transformation from bar currents to d-q currents is given by Cr N , where Cr N is of the form of 3.3. From the preceding discussion we know that: i dqr N = Cr N ir N (3.41) However, as we have shown, for every set of S = p 1 + p2 circuits on the rotor only 1 d-q pair
3.4 Model order reduction for Nested-loop rotor
65
where ir N R S , i dqr N R2 and Cr N R2S and it is full rank. We seek a transformation, Cr N R22 , such that: i dqr N = Cr N Pir N and is rank 2 (that is the 1s must be in different columns). For example P might be: P = I2 0 From 3.41 and 3.42 we may write: Cr N Pir N = Cr N ir N we have shown in the previous section that i r N = Cr N T i dqr N , therefore: Cr N PCr N T i dqr N = Cr N Cr N T i dqr N as Cr N Cr N T = I2 : Cr N PCr N T i dqr N = i dqr N as Cr N PCr N T is full rank: Cr N PCr N T = I2 (3.47)
T 1
(3.42)
where P R2S , and P is a matrix which has a single 1 in each row, with all other elements zero, (3.43)
(3.44)
(3.45)
(3.46)
Cr N = PCr N
(3.48)
Therefore for each of the N sets of S rotor circuits a 2 by 2 transformation, which determines the d-q 1 rotor currents from any two bar currents, is given by Cr N = PCr N T

The preceding derivation, culminating in remark 3.1, gives a full transformation for any BDFM rotor in a broad class. However, each of the N sets of S rotor circuits yields an additional d-q pair. Therefore the overall models the rotor, thus reducing the order to 8. model is of order 6 + 2N . It is desirable to reduce the order of the model so that a single d-q pair Boger et. al. proposed a 8 order model (which reduces to 6 if the stator zero-sequence states can be removed), [11]. However, their proposal is simply to sum the contributions from the N different sets of circuits. No comment is made as to whether or not this is a good or poor approximation (in fact they do not mention that it constitutes a model reduction at all).
66
1 Mdq Q dq 1 Mdq 1 Mdq Rdq
r i I i
Figure 3.2: Block diagram of d-q transformed BDFM model
We will show that this can lead to a poor approximation, sometimes little better than ignoring inner loops within a nest all together. The physical reason for this is that, in a nested-loop design rotor, the N sets are the different pitch coils within each nest. The difference in pitch means that different currents will be induced into each of the coils, and the mutual coupling between the coils within a nest will lead to circulating currents within each nest. These currents are eliminated if the inductance terms are simply summed, leading to erroneous results. Section B.3 details the mutual inductance matrices for the prototype nested-loop design rotor. Equation (B.13) shows the d-q mutual inductance matrix for this rotor design. Signicantly (B.13) conrms the coupling between the sets of rotor circuits by the presence of non-zero off-diagonal elements. There is, therefore, a need for an improved method of model reduction which reduces the approximation error in some sense.
3.4.1
Model Reduction Techniques
From the preceding analysis it is clear that the electrical equations for the BDFM are linear. That is, the currents (and voltages) appear linearly in the equations. Therefore the BDFM equations for the d-q transformed model, (3.35) may be represented by the block diagram in gure 3.2. It is important that the BDFM may be modelled as a linear system because linear systems are amenable to a rich set of techniques, particularly in terms of model reduction. Model reduction has received considerable attention in the eld of linear systems theory. It will be of use to redraw gure 3.2 separating out the connections between rotor and stator. To simplify the notation we will repartition the state vector, and corresponding matrices into rotor and stator quantities, rather then stator 1, stator 2, and rotor quantities as in equations (3.35):
67
PSfrag replacements Msi Rs i

i Msr T T
Msi r Rs Msi Q s Q s
T i Msr Rr
Msi Q sr r
i Msr T
r ir x
i Msr Q sr
ur
Mri Rr Figure 3.3: Block diagram of d-q transformed BDFM model with rotor and stator separate
Mdq = Q dq = Rdq =
Ms Msr T Qs 0 Rs 0
Msr Mr Q sr 0 0 Rr
(3.49) (3.50) (3.51)
and we partition the inverse of Mdq compatibly with Mdq as follows, using superscript i to denote
1 that the block matrix belongs to Mdq rather than Mdq . The analytical expression for these elements
can be determined from Lemma A.5. Note that Msi = Ms1 etc.:
1 Mdq =
Msi
i Msr
i Msr
Mri
(3.52) (3.53)
We start by reviewing some possible model reduction techniques
68
Linear, Time-invariant Model Reduction Techniques Linear, time-invariant (LTI) systems have had considerable attention devoted to their analysis, and model reduction is no exception. Two techniques of LTI model reduction are balanced truncation and optimal Hankel-norm approximation. Both these techniques offer automatic methods of model reduction for LTI systems, that is, they are mathematical procedures which give results that are guaranteed to be good in some sense. Before proceeding further is it necessary to introduce some of standard notation of linear systems theory: Denition 3.1. [131, p. 92] Any function, x(t) Rn may be said to be L2 (I ) space iff: x
2 I
|x(t) x(t)|dt
1/2
<
L2 (I ) if x(t) has a bounded 2-norm over the interval I . L2 [0, ) is denoted by the short hand L+ . 2 2-norm, P is dened as:
where I R is the interval, which will typically be [0, ). Thus a function x(t) is a member of
Denition 3.2. [131, ch. 4] Given a linear system, P, mapping u Rm to y Rn then the induced
i,2
i,2
sup
u(t)L+ 2
y u
2 2
where sup stands for supremum, or least upper bound. Hence the induced 2-norm of P is the worst-case gain for any input signal in L+ 2 Both balanced truncation and the optimal Hankel-norm approximation can guarantee that induced 2-norm of the error system will not be more than some upper bound. To explain the precise nature of this upper bound it is necessary to introduce some more systems theory notation: Denition 3.3. [131, ch. 3] A minimal realization of an LTI system, G, with input u(t) and output y(t) may be written as: x = Ax + Bu y = C x + Du or more compactly as: G = dimension of x is minimised. The controllability and observability gramians, P and Q respectively are given by the solutions to: A C B D . A minimal realization is a representation of G for which the
3.4 Model order reduction for Nested-loop rotor PSfrag replacements
69
G1 G2
G2
G1
Figure 3.4: Parallel and series connection of systems
A P + P A + B B = 0 A Q + Q A + C C = 0 where P = P 0, Q = Q 0, that is P and Q are Hermitian, positive semi-denite. The Hankel singular values of G are (P Q), where () denotes the eigenvalues. Furthermore there will always exist an invertible state transformation, T , such that x = T x, and the transformed 1 1 gramians become P = T P T = (T ) QT = Q = where is a diagonal matrix of the Hankel singular values of G in descending order. The realization of G which achieves P = Q = is called
the balanced realization.
Remark 3.3. Series and parallel connection of two systems, G 1 , G 2 may be conveniently represented by G 1 G 2 and G 1 + G 2 respectively. If G 1 , G 2 are linear then they can be represented by G 1 = A1 B1 A2 B2 , G2 = . The parallel connection can be represented by C 1 D1 C 2 D2 A1 0 B1 B2 D1 + D2
and the series connection by:
G 1 + G 2 = 0 A2 C1 C2
A1
B1
Diagrammatically the connection are illustrated by gure 3.4.
G 2 G 1 = B2 (C1 + D1 ) A2 D2 C 1 C2
0 D2 D1
Model reduction by balanced truncation reduces the order of the system by removing the states corresponding to the smaller Hankel singular values. The induced 2-norm of the error system, that is the error between the full-order and reduced order system for any input with unitary 2-norm, will be less than twice the sum of the truncated Hankel singular values [131, ch. 7].
Model reduction by the optimal Hankel norm approximation, seeks to nd a reduced order system, such that the Hankel norm of the error system G G is minimised. The Hankel norm of a system G,
70
is the maximum Hankel singular value of that system. It turns out that for LTI systems it will always be possible to nd G, and furthermore it will always be possible to nd a constant D 0 such that the 8],[40].
induced 2-norm of G G D0 is less than the sum of the truncated Hankel singular values [131, ch. Both these techniques could be applied to the d-q transformed model of the BDFM previously
derived, however the BDFM model, although linear, is not time-invariant. The model is dependent on a parameter, the rotational speed, r thus the techniques can only be applied for a xed speed. There are, however, generalisations of both balanced truncation and optimal Hankel-norm reduction for linear parameter-varying (LPV) systems, which will now be considered. Model Reduction Techniques for Linear Parameter Varying Systems The methods of model reduction for linear parameter varying systems, are rather less established than for the LTI case. Nevertheless methods have appeared in the literature which are applicable. In the BDFM model presented, the parameter variation appears in the A matrix. From gure 3.2 it can be that is, linear with an offset. seen that in the BDFM the A matrix becomes A 1 + r A2 . Hence the parameter dependence is afne Wood et. al. [118] and Wood [119] present generalisations of balanced truncation and Hankel norm approximation for LPV systems. Related work in discrete time, which will not be further discussed here, is presented in Lall and Beck [58], Beck and Doyle [5], Beck and Bendotti [4] and Beck, Doyle and Glover [6]. An LPV system is one where time-dependent parameters enter the state equations through one or more exogenous parameters: x = A()x + B()u where is the parameter vector which may be time-varying. y = C()x + D()u (3.54)
In [118] the notions of controllability and observability gramians, balanced realizations, and Hankel singular values is generalised to LPV systems. These are then used to obtain similar bounds for the worst case induced 2-norm error. However, unlike in the LTI case the results do not lead to a unique solution for each system. That is, suppose one wanted to reduce the order of a linear system from 10 to 6: In the LTI case both balanced truncation and the optimal Hankel norm approximation method lead to a unique upper bound on the 2-norm error for that particular system and reduction choice. In the LPV case however, although a bound can be found, there is no guarantee that it is the smallest bound possible. The reason for this will become apparent in the following discussion. From [118] the generalised gramians are given by any solutions for P and Q to: A()T Q + Q A() + C()T C() < 0, Q = Q T > 0, (t) F A()P + P A()T + B()B()T < 0, P = P T > 0, (t) F (3.55) (3.56)
71
where F is the set of all possible parameter trajectories. Having found such a P and Q the procedure is then the same as for the LTI case for balanced truncation. Although a Hankel-norm approach is also presented which achieves an analogous bound to the LTI case, the authors note that the additional complexity introduced limits the applicability of the result, and therefore it will not be considered further here. Note that, from the preceding discussion P and Q are not unique, because (3.55) and (3.56) are inequalities rather than equalities in the LTI case. Because the BDFM d-q transformed equations are afne in their only parameter, r (see gure 3.2), then from Lemma A.7 (3.56,3.55) will be satised
min max for all possible r if they are satised for the extreme values of r . To be precise, if r < r < r
then (3.56,3.55) are satised if:

max max (A1 + r A2 )T Q + Q(A1 + r A2 ) + C T C < 0 T
min min (A1 + r A2 ) Q + Q(A1 + r A2 ) + C T C < 0, Q = Q T > 0 T max max T (A1 + r A2 )P + P(A1 + r A2 ) + B B < 0 min min (A1 + r A2 )P + P(A1 + r A2 ) + B B T < 0, P = P T > 0 T
(3.57)
(3.58)
The problem of nding a P and Q satisfying (3.57,3.58) (and with any further linear constraint placed on the elements of P and Q) may be efciently solved. The problem is a linear matrix inequality (LMI) which is a convex optimization problem for which efcient algorithms exist [37]. As the upper bound on the error is the sum of the truncated singular values, the smaller the truncated singular values can be made, the tighter the bound that can be achieved. Unfortunately the constraint on P and Q to minimize the square-root of eigenvalues of P Q, or even the eigenvalues themselves, amounts to a non-linear constraint, and therefore the problem becomes non-convex, and hence impractical to solve. In [118] a procedure is proposed that will nd a locally optimal solution to the problem, via minimising the trace of a Q P pair, which therefore nds the minimum sum of the eigenvalues of the Q P pair. Their algorithm is as follows: 1. Find a Q 1 which satises (3.57) and minimises permissible r values.
k
trace(Q 1 P0 (rk )) where P0 (rk ) is the LTI
controllability gramian for r = rk , and rk is the k th value for r over some grid of range of 2. Solve for a P1 which satises equation (3.58) and minimises trace P1 Q 1 . 3. Solve for a Q 2 which satises equation (3.57) and minimises trace Q 2 P1 . 4. Repeat steps 2 and 3 until the decrease in the cost function satises some convergence criterion. Recall that the trace of a matrix is the sum of the eigenvalues, so this procedure has the effect of trying to nd a P and Q for which the sum of the eigenvalues of P Q are small. However the procedure will not necessarily nd the global minimum, and furthermore even if it did there may still
72
be a P and Q which lead to a smaller upper bound on the error between the original and reduced system. Even if the optimal solution were to be found, one potential shortfall of the previous approach is that the reduction process makes no assumption about the rate of change of the parameter, in our case r . This may lead to further conservatism in the results. Explicit bounds on the rate of variation of the parameter can be included, by making P and Q functions of the parameter, however the existence of a suitable transformation matrix to balance the system becomes potentially, hard to nd, see [119] for details. Therefore parameter dependent P and Q will not be considered further. It is worth mentioning that although the techniques described offer guaranteed bounds on the induced 2-norm of the error, it is possible to easily calculate the induced 2-norm of the error (or indeed of any system) in both the constant system and afne linear parameter varying cases. In the LTI case the induced 2-norm is simply the maximum (technically a supremum) singular value of the frequency response of the system over all frequencies [131, ch. 4]. In the LPV case it is a little more involved. Theorem 3.4. [7, Lemma 2.10] & Lemma A.9 Given the LPV system G = all possible parameter trajectories: where C() D() Rm is a time-varying parameter and R > 0. If there exists an X = X T Rnn such that for AT X + X A BT X C XB I D CT A() B()
then the system G is stable and there exists < such that ||G||i,2 . Using theorem 3.4 an upper bound on the induced 2-norm of a system can be computed by nding the solution X which minimises 2 . This problem is often known as the L2 performance problem. Fortunately theorem 3.4 presents this problem as an LMI with linear constraints, which can therefore be directly solved. However, in general, every parameter value must be checked, which constitutes an innite number of LMIs to be solved. The parameter space can, however, be gridded and the solution approximated with a nite number of LMIs (see, for example [119] for details). In the case that appears afnely in G then it can be shown that it is only necessary to check the extreme point of the parameter box. As the parameter dependence in the BDFM only appears in A, the sufciency of checking the extreme points can be seen from Lemma A.7 with Lemma A.8. Using Matlabs LMI Control tool box this bound can be computed using the quadperf function [37]. The L2 performance problem may also be represented as an integral quadratic constraint (IQC) [74], for which a tool box is available for calculation using Matlab [73].
DT < 0 I 2
73
Stability of the BDFM The LPV balanced truncation, and the new method of model reduction to be described, require the system to be quadratically stable. We will now prove that the electrical BDFM equations are always quadratically stable. Theorem 3.5. Any BDFM satisfying the present assumptions is quadratically stable for any bounded variation in shaft speed. Proof. It was established in gure 3.2, that the dq-transformed BDFM equations may be written as: di = A0 i + r A1 i + Bv dt for suitable choices of A 0 , A1 , B as given by gure 3.2. Or equivalently: G= A0 + r A1 I B 0
Quadratic stability can be ensured by nding a single Lyapunov function which holds for all parameter trajectories, that is nd a P = P T > 0 : A()T P + P A() < 0, F , where F is the set of permissible parameter trajectories and A() the system matrix ([7, Def 2.4]). We rst consider the eigenvalues of Rdq Q dq . Q dq and Rdq for the class of machines considered
are the union of the eigenvalues of: Q dq0s1 Rdq0s1 , Q dq0s2 Rdq0s2 and Rdqr . Q dq0s1 Rdq0s1 and Q dq0s2 Rdq0s2 have strictly positive real parts.
is given in (3.35). As Rdq Q dq is block triangular, from Lemma A.6, the eigenvalues of Rdq Q dq
As Rdq is diagonal and Rdq > 0, from the form of Q dq0s1 and Q dq0s2 then the eigenvalues of
values of Rdq0s1 Q dq0s1 , and similarly the eigenvalues of Rdq0s2 + Q dq0s2 are the same as those of are skew-symmetric with real coefcients, and the remainder of the matrices are zero. in the right half-plane, that is they have positive real parts. Rdq0s2 Q dq0s2 . Essentially the reason for this is that the upper left 2 2 blocks of Q dq0s1 and Q dq0s2
Furthermore it is easy to show that the eigenvalues of R dq0s1 + Q dq0s1 are the same as the eigen-
Therefore the eigenvalues of Rdq Q dq are the same regardless of the sign of Q dq and are strictly
1 From [32, Th. 3.1] Mdq (Rdq r Q dq ) must have all eigenvalues in the left half-plane (i.e. it is
Hurwitz), as Mdq is positive denite (from Lemma 2.3, and recalling that Mdq = T T M T , for some T ).
1 We have established that Mdq (Rdq r Q dq ) has the same eigenvalues regardless of the sign
of r Q dq . Therefore as the eigenvalues are the same both matrices must, or at least can, have the
same Jordan Canonical form (see [131, Theorem 2.4]). Therefore there exists a T Cnn such that in general, be a function of r .
1 1 Mdq (Rdq + r Q dq ) = T Mdq (Rdq r Q dq )T 1 , where n is the dimension of Mdq . Note that T will,
74
r < max . Because the BDFM equations are afnely parameter dependent, then from Lemma A.7, satisfy the intermediate points. if we can nd a single P which satises the extreme points, i.e. max and max then it will also The systems corresponding to the extreme points may be written as: G1 = G2 = A0 + max A1 I A0 max A1 I B 0 B 0
Now to prove quadratic stability we seek a single Lyapunov function which holds for all max <
but, from the previous discussion G 2 may also be written as: G2 = A0 + max A1 T 1 TB 0
and therefore both G 1 and G 2 have the same A matrices, thus any P which satises (A 0 T + max A1 T )P + P(A0 + max A1 ) < 0 proves quadratic stability for both G 1 and G 2 , and we know 2.2.1]). that such a P will always exist as A 0 + max A1 has strictly left half-plane eigenvalues ([47, Th Therefore we know that a P exists for both G 1 and G 2 , therefore the BDFM is quadratically stable for any speed variation in the range max < r < max . Then as max was chosen arbitrarily, quadratic stability is guaranteed for all r
Discussion of suitable model reduction methods for the BDFM with a nested loop rotor We return now to our aim, to reduce the order of the d-q transformed BDFM equations so that the rotor is represented by just two states. In other words we wish to nd a simplied model of the rotor which is equivalent in complexity to the model that one would derive from the special case when the rotor comprises of only one set of coils. Physically speaking, we are trying to nd an single set of rotor coils which reasonably approximate the performance of the true rotor which has multiple sets of coils. From the previous discussion, it is clear that for dynamic BDFM modelling the machine cannot be considered LTI, but rather LPV, as the rotor shaft speed will not be constant. Now we wish to reduce the order of only the rotor states, leaving the stator states untouched. To apply balanced truncation in this case it is necessary separate the rotor and stator systems, as shown in gure 3.3. If the rotor states are not separated then, although the model reduction techniques can still be used, the helpful physical distinction between rotor and stator states will be lost. The problem is therefore to approximate the rotor currents, i r , using only two states. From gure
75
3.3, it can be seen that the rotor currents are governed by the following differential equation: dir i T = Mri Rr + r Msr Q sr ir + u dt where u represents the exogenous inputs from the stator currents and input voltage. The rotor system is balanced by application of the appropriate state transformation, and then states truncated to reduce the order of the rotor to 2. However, there is no simple way to represent the balanced system in terms of mutual inductance, resistance, and Q terms, thus the physical interpretation that the component matrices hold is lost. The loss of physical meaning of the components doesnt necessarily have any bearing on the efcacy of the method, however it is signicant. Although a model is still of use as a black box model, in that its predictive power is unchanged, the designer can no longer easily glean insight from the form of the model. There is therefore, a need for a method of model reduction which retains some physical interpretation, yet is still an accurate approximation to the original system. We now present such a method. (3.59)
3.4.2
New BDFM Rotor State Reduction Technique
An alternative approach to the more automatic methods can be found through consideration of the BDFM differential equations. For reasons which will become apparent, it is necessary to diagonalize the rotor mutual inductance matrix, Mr , by means of an orthogonal (invertible) transformation - this will always be possible as Mr is symmetric (as shown in (3.24)). Furthermore the state order will be chosen such as to order the eigenvalues in decreasing order from the top left. This can be achieved with the following state transformation: I 0 is = ir 0 T is ir (3.60)
where T is a matrix of the eigenvectors of Mr in an appropriate order. With Mr diagonal we now partition the BDFM equations into three parts, stator, 2 rotor states, remainder of the rotor states. Then the equations may be written as: vs Ms 0 = MT sr 1 T 0 Msr
2
Msr 1 Mr1 0
Msr2
is Rs d ir + 0 1 0 dt Mr2 i r2 0
T Rr12
0 r1 R
note that the off-diagonal resistances will, in general, be non-zero.
0 Rr12 + Rr2 Q s Q sr 1 r 0 0 0 0
Q sr 2
is ir (3.61) 0 1 0 i r2
76
good approximation in all the cases tried.
could be removed with no error. It turns out that in the typical BDFM nested-loop rotors which were simulated, these terms were always small, and hence truncating the system by removing ir2 was a Physically, having a small value of Msr2 , (and hence Q sr 2 ) corresponds to having a (relatively)
Now, notice that if Msr2 , Rr12 and Q sr 2 were all zero, then the system would be reducible: ir2
large transformer turns ratio between the stator and that part of the rotor state. This means that smaller voltages are induced, and if the resistances are similar, then the current will be smaller. The effect
of diagonalization (and ordering) of Mr has been to prioritise the effect of the stator coupling on the various parts of the rotor, hence by keeping the states corresponding to the most signicant M r singular values (as Mr is symmetric the eigenvalues are the singular values), and removing the other states, the system will be reduced with little error. The reduction algorithm can be summarised as:
1. Compute T , a matrix of eigenvectors of Mr ordered so that the corresponding eigenvalues decrease from left to right. 2. Partition T into [ T1
T2
], where T1 is 2 columns wide.

I 0 0 T1 T
3. Apply the resulting non-square state transformation, i = order.
i, which reduces the state
Thus application of the above algorithm to (3.61) leads to: vs 0 Ms T Msr Msr 1 d i s + Mr1 dt ir1 Rs 0 0 Rr1 Qs 0 Q sr 1 0 is i r1
(3.62)
Signicantly, the structure of the component mutual inductance, resistance and Q matrices can be shown to be the same as that for a BDFM with a rotor comprising of a single set of coils (i.e. where each rotor nest has only 1 loop in it). This means that the method can be thought of as determining an effective single loop inductance and resistance for each nest. This fact will be used in chapter 5 to compare a nested-loop rotor design to a novel rotor design by comparing terms in the machine equations. This fact will now be proved. Equation (3.24) gives the general form of the rotor-rotor mutual inductance matrix, however in removing the unobservable and uncontrollable states, the rows and columns containing terms denoted by a and the zero sequence terms are deleted. In the case of the nested loop design rotor, each of the off-diagonal matrix blocks are themselves diagonal. In other
77
words, M12b of (3.24) is zero. Therefore the rotor-rotor inductance matrix has the form: M12a 0 L dq r 1 0 0 0 M12a L dq r 1 M12a 0 . .. . . . Mdq r = M12a 0 L dq r N 0 . . . 0 L dq r N xa = x1 0 x 3 0 x n1 0 ,
T
(3.63)
Let x be any eigenvector of Mdq r with corresponding eigenvalue, x . We now dene, x a and x b . xb = 0 x2 0 x4 0 0 xn
T
where xi is the i th element of the eigenvector, x. Because of the structure of M dq r then x a and x b are also eigenvectors of Mdq r with the same corresponding eigenvalue, x , as long as x a = 0 and x b = 0. that: From examination of the structure, as Mdq r xa = x xa and Mdq r xb = x xb then we can conclude xb = k 0 x 1 0 x 3 0 0 x n1 ,
T
where k R is arbitrary.
If x b is zero then the corresponding result for x a applies instead. Therefore as all the eigenvalues of Mdq r appear in pairs, and the corresponding eigenvectors are of the form of x a and x b , then the transformation matrix T1 will always be of the form: 1 0 0 1 2 0 T1 = 0 2 . . . . . . 0 n 0 n
k I , that is a scalar multiplied by the identity matrix. It is easy to show that the retained portion of the transformed resistance matrix, Rr1 , is also of the form k I . The stator-rotor portions of the mutual inductance and Q matrices may be shown to transform to matrices of the form: M1 cos(1 ) M1 sin(1 ) M sin( ) M cos( ) 1 1 1 1 Msr 1 = , M2 cos(2 ) M2 sin(2 ) 2 sin(2 ) M2 cos(2 ) M
Therefore, it is immediate that the transformed rotor-rotor mutual inductance, Mr1 , is of the form
Q sr 1
M1 p1 sin(1 ) M1 p1 cos(1 ) M p cos( ) M p sin( ) 1 1 1 1 1 1 = M2 p2 sin(2 ) M2 p2 cos(2 ) M2 p2 cos(2 ) M2 p2 sin(2 )
78
Therefore the new model reduction method reduces the order of the system in such a way that the reduced order model has the same structure as a machine with a rotor comprising of one loop per nest.
3.5 Simulation comparison of different BDFM model reduction techniques

In order to evaluate the new nested loop rotor BDFM model reduction technique, different BDFM congurations were tested. Typical results will be presented, for a range of reduction techniques: 1. The new reduction method, as described in section 3.4.2. 2. Biggest loop method: Model reduction by assuming that the rotor comprises of only the outer loop of each nest. This technique can be easily applied by simply truncating parts of the BDFM system matrices corresponding to the unwanted states. 3. Sum loops method: This is essentially the method proposed in [64, 63, 11, 12]. The model can be reduced by summing the inductance, resistance, etc components for each set of loops into a single lumped parameter set. 4. LPV balanced truncation method: the method described in section 3.4.1, due to [118, 119]. The reduction method is applied only to the rotor, as previously described. Once the balancing transformation has been found, the rotor system can be transformed, and then the component system matrices truncated along with the state. The overall reduced system is given by G = As Asr Cr Bs Br Ar s Ar Br Br . I 0 0
5. Full dq system: This is the full, untruncated system, as derived in section 3.3.
Figure 3.5 shows the upper bound on the induced 2-norm gain of the error between the full and reduced system in each case. Two lines are plotted for each case, the straight line is for arbitrary speed trajectories within the prescribed range, and is calculated using Theorem 3.4. The non-straight lines are for constant speed, and are calculated from the supremum of the maximum singular value of the frequency response for each speed. Clearly the new method provides signicantly better performance (lower error) than the other methods. Note that the summing method and biggest loop method performance are not that different, therefore suggesting that summing contributions is scarcely a better approximation than assuming a simpler rotor structure. The relatively poor performance of the LPV balanced truncation is probably due, partly to the non-convexity of the problem, and hence the solution is only locally optimal, but more signicantly, because of the requirement to apply the

10
0
79
L2 gain of error system
10
10
LPV bal. trun. Biggest loop Sum loops New method
PSfrag replacements
10 1000 800
600
400
200 0 200 Rotor speed (rpm)
400
600
800
1000
Figure 3.5: Maximum induced 2-norm gain of the error between the full and reduced d-q-0 transformed BDFM nested-loop rotor design models. Straight lines are upper bounds for arbitrary speed variations computed using Theorem 3.4. Curved lines are upper bounds for xed speeds, calculated from the supremum of the maximum singular value of the frequency response for each speed, the glitches in the red line are due to numerical problems. The upper bound on the induced 2-norm of the full system was 0.7382. The full-order system has 17-states (15 electrical states - of which 5 correspond to zero sequence states, and 2 mechanical states), and the reduced order systems are 8 states (6 electrical and 2 mechanical). technique only to the rotor states. If this limitation could be removed then the technique would be superior. By way of comparison, the LPV balancing procedure was applied to the entire system, and the results are similar or better than the new reduction method. Notwithstanding these comments, all the methods achieved some amount of effective model reduction. The upper bound on the induced 2-norm of full system was 0.7382, thus if the reduced systems all have similar 2-norm values then the worst case error is 1.4764 (twice 0.7382, from the triangle inequality), so even the LPV balanced truncation method achieved some level of error reduction. Figures 3.6, 3.7, 3.8, 3.9 and 3.10 show current and rotor shaft speed transients for the BDFM with a nested-loop rotor operating in synchronous mode. The machine was simulated with zero initial current, but with the shaft speed set to synchronous speed (700rpm in this case), the gures show the speed and current transient as the machine attains equilibrium. As expected, the error between the reduced order and full systems is least in the case of the new reduction method. The sum loops method actually appears to be slightly worse than the biggest loop method in the current plot, but
80
40 30 20 Red phase current (A) 10 0 10 20 30 40 0
New method Biggest loop Sum loops Full dq model
0.05
0.1
0.15
t (s)
0.2
0.25
0.3
0.35
Figure 3.6: Stator 1 phase current transient at startup for a BDFM with a nested loop design rotor in synchronous mode, for different BDFM model reduction methods. Full-order d-q-0 model: 17 states, reduced order models: 8 states
slightly better in the speed plot. The LPV balanced truncation method yielded very poor results in these transient tests, and so has been omitted. Again, these plots serve to illustrate the effectiveness of the new model reduction method. It is signicant in gures 3.7, 3.8 and 3.9 , that the new model reduction method achieves good estimation of the rotor bar currents, in contrast to the other method considered. Previously the only available BDFM model which could estimate the rotor currents with some degree of accuracy was the full order d-q model [53]. It is worth adding some condence to the claim made at the start of the chapter, regarding the additional assumptions made in the d-q-0 transformed BDFM model. It was claimed that the inductive coupling between stator and rotor can be adequately approximated by the rst non-zero spatial harmonic of the Fourier series representation of the mutual inductance matrix. The d-q-0 and original coupled circuit model of chapter 2 differ only in this respect. Figures 3.11 and 3.5 show currents and shaft speed transients at startup for the full order d-q-0 transformed model, and the original coupled-circuit model. The coupled circuit model for this BDFM is 26 states, of which 24 are electrical, and 2 mechanical. The full d-q-0 transformed model has 17 states, 15 electrical and 2 mechanical, however 5 of the electrical states correspond to zero-sequence states, and as the supply in this case was balanced 3-phase, then the zero sequence states will remain at zero.
81
800 600 400 Inner loop rotor bar current (A) 200 0 200 400 600 800 1000 0
0.05
0.1
0.15
t (s)
0.2
0.25
0.3
0.35
Figure 3.7: Inner loop rotor bar current transient at startup for a BDFM with a nested loop design rotor in synchronous mode, for different BDFM model reduction methods. Full-order d-q-0 model: 17 states, reduced order models: 8 states
1500
1000 Middle loop rotor bar current (A)
500
500
1000
1500 0
0.05
0.1
0.15
t (s)
0.2
0.25
0.3
0.35
Figure 3.8: Middle loop rotor bar current transient at startup for a BDFM with a nested loop design rotor in synchronous mode, for different BDFM model reduction methods. Full-order d-q-0 model: 17 states, reduced order models: 8 states
82
2000 1500 1000 500 0 500 1000 1500 2000 0
Outer loop rotor bar current (A)
0.05
0.1
0.15
t (s)
0.2
0.25
0.3
0.35
Figure 3.9: Outer loop rotor bar current transient at startup for a BDFM with a nested loop design rotor in synchronous mode, for different BDFM model reduction methods. Full-order d-q-0 model: 17 states, reduced order models: 8 states
850 New method Biggest loop Sum loops Full dq model
800
Speed (rpm)
750
700
650
600 0
0.1
0.2
0.3
0.4
t (s)
0.5
0.6
0.7
0.8
0.9
Figure 3.10: Rotor shaft speed transient at startup for a BDFM with a nested loop design rotor in synchronous mode, for different BDFM model reduction methods. Full-order d-q-0 model: 17 states, reduced order models: 8 states. The initial current was zero, the initial shaft speed was the synchronous speed, 700rpm.
3.6 Conclusion
83
The agreement between the two models for the conditions tested is very good, thus giving condence to the assumption made to derive the d-q-0 model. These results are typical of many operating points, and BDFM models tried.
40 30 20
Inner loop Rotor current (A) 500 400 300 200 100 0 100 200 300
Full dq model (17state) Full 26state coupledcircuit model
Red phase current (A)
10 0 10 20 30 40 0
400
0.05
0.1
0.15
0.2
t (s)
0.25
0.3
0.35
0.4
0.45
500 0
0.05
0.1
0.15
0.2
t (s)
0.25
0.3
0.35
0.4
0.45
(a) Stator 1 phase current

1500 2000 1500 1000 500 0 500 1000 1000 1500 2000 0
(b) Rotor inner loop bar current
1000 Middle loop Rotor current (A) Outer loop Rotor current (A) 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
500
500
1500 0
t (s)
0.05
0.1
0.15
0.2
t (s)
0.25
0.3
0.35
0.4
0.45
(c) Rotor middle loop bar current
(d) Rotor outer loop bar current
Figure 3.11: Current transients at startup for a BDFM with a nested-loop design rotor in synchronous mode for the original coupled-circuit BDFM model of chapter 2 and the full order d-q-0 transformed BDFM model of this chapter. Full-order d-q-0 model: 17 states (15 electrical, 2 mechanical), coupled-circuit model: 26 states (24 electrical, 2 mechanical)
3.6 Conclusion
In this chapter we present a rigorous derivation of the d-q transformed model for a wide class of BDFMs, including, but not limited to, the nested-loop rotor of Broadway and Burbridge [17]. The model was derived directly from the coupled circuit model presented in chapter 2. The derivation is
84
850
800
Speed (rpm)
750
700
650
600 0
0.1
0.2
0.3
0.4
t (s)
0.5
0.6
0.7
0.8
0.9
Figure 3.12: Rotor shaft speed transient at startup for a BDFM with a nested-loop design rotor in synchronous mode for the original coupled-circuit BDFM model of chapter 2 and the full order d-q-0 transformed BDFM model of this chapter. Full-order d-q-0 model: 17 states (15 electrical, 2 mechanical), coupled-circuit model: 26 states (24 electrical, 2 mechanical). The initial current was zero, the initial shaft speed was the synchronous speed, 700rpm.
3.6 Conclusion
85
consistent with that presented in [12], although the method is more general. Furthermore we show that the states removed from the d-q transformed model are unobservable and uncontrollable, but stable. In addition it was proved that all BDFMs in the analysed class are quadratically stable for all shaft speed trajectories. It has also been shown that the d-q transformation proposed in Boger et. al. [11] is a simplication, and can lead to signicant errors, both in dynamic and steady-state performance. We have derived a new method of reducing the order of the derived d-q model for nested-loop rotor design BDFMs. The new method yields a model of the same order model as that found in [11], but with greatly improved accuracy, in both transient performance and worst case L 2 gain.
86
Chapter 4
Equivalent Circuit Model and its Implication for BDFM Performance

4.1 Introduction
In chapter 3 the coupled circuit model derived for the BDFM was transformed into d-q axes. In this chapter we show how the d-q transformed model may be further transformed into complex symmetrical components. We then show how, in the steady state, this form of the BDFM model may be written succinctly as an equivalent circuit where the meaning of the parameters has a clear physical interpretation. The performance of the BDFM is then investigated using the equivalent circuit, and some measures of goodness of the BDFM are proposed, in terms of calculable machine parameters. While the development of the equivalent circuit in this dissertation is achieved by way of transformations applied to the d-q model, the author has also shown that it is possible to derive the equivalent circuit by considering the BDFM as two interconnected induction machines [86]. Li et al. used the d-q model for the 6/2 machine of [64], to derive an equivalent circuit representation, which was then used to investigate power ows within the machine [60]. A similar approach was adopted by Gorti et al. [41]. Other references propose equivalent circuits without derivation, [65, 59]. Yet in the literature the equivalent circuits presented are either for an idealised machine, or for a specic machine conguration. There is, therefore, a need for a generalised derivation of the equivalent circuit for a wide class of BDFM machines. Williamson et. al. analysed the synchronous mode of operation of all nested-loop rotor BDFMs [115] via harmonic analysis. However, the model is not developed into an equivalent circuit form, nor is it used to identify design parameters. Broadway and Burbridge [17], Hunt [48, 49], and Creedy [27] make qualitative comments regarding rotor design, however, nowhere in the current literature has any method been proposed to quantitatively evaluate the performance of BDFM machines with different rotor designs. In this chapter we consider quantitative methods of BDFM evaluation. Furthermore we briey consider the issue of magnetic loading for the BDFM, which is not
87
88
straightforward because there are two different pole number elds in the air gap during normal operation. During this chapter the modelling assumptions are exactly the same as for the d-q model, as described in chapter 3.
4.2 Conversion to Symmetrical Components

From [51, sect. 3.7], d-q-0 components may be transformed into positive, negative and zero sequence components using the following unitary (invertible) transformation matrix: i+ i = C 0 i0 where: 1 1 j C= 2 1 j the voltage sequence components are similarly dened. For brevity the zero sequence components will not be included, however they can be deduced from the d-q model, as the zero sequence are unaffected by the transformation (as the 0 of the d-q-0 components is the zero-sequence). model, or 3 copies of C along the diagonal for a reduced order model. N is the number of sets of is: The full transformation matrix will consist of N +2 copies of C along the diagonal for a full-order id 0 i q 1 i0
(4.1)
(4.2)
S = p1 + p2 circuits present on the rotor. For the full-order model the whole transformation matrix C1 0 0 C ... 2 = . . . .. ... . 0 0 0 . . . 0
Cfull
(4.3)
C N +2
where all Ci are equal, the subscripts are simply for book-keeping purposes. This transformation shall now be applied to the full-order BDFM d-q model of equation (3.35), with the zero-sequence states removed. Equation (3.35) is multiplied by (4.3) from the left and from
89
the right by the complex conjugate transpose of (4.3) which gives:
i cs1 i cs1 d vc = (Rc + r Q c ) i c + Mc i c s2 s2 dt s2 0 i cr i cr Q cs1 0 Q csr 1 Mcs1 Q c = C Q dq C = 0 Q cs1 Q csr 2 , Mc = C Mdq C = 0 0 0 0 Mcsr 1 T Rcs1 0 0 Rc = C Rdq C = 0 Rcs2 0 0 0 Rcr vcs1
(4.4) 0 Mcs2 Mcsr 2 T Mcsr 1 Mcr
Mcsr 2
where Q c , Mc , Rc C2(N +2)2(N +2) , and the elements of which can easily be shown to be:
Rc =
Rdq1 0 0 0 0 0 . . . 0 0
0 Rdq1 0 0 0 0 . . . 0 0
0 0 Rdq2 0 0 0 . . . 0 0
0 0 0 Rdq2 0 0 . . . 0 0
0 0 0 0 Rdqr 1 0 . . . 0 0
0 0 0 0 0
0 0 0 0 0
Rdqr 1 0 .. .. .. . . . 0 Rdqr N 0 0
Rdqr N
0 0 0 0 0 . . . 0
(4.5)
The vertical and horizontal lines demark rotor and stator portions of the matrix. Q c and Mc are given by:
90
Mc =
0 (L 1 M1 ) 0 0 0 M11 exp( j11 ) 0 0 (L 2 M2 ) 0 0 M21 exp( j21 ) 0 0 0 (L 2 M2 ) M21 exp( j21 ) 0 M11 exp( j11 ) 0 0 M21 exp( j21 ) L dq r 1 0 M12r exp( j12r ) 0 0 M11 exp( j11 ) M21 exp( j21 ) 0 0 L dq r 1 0 M12r exp( j12r ) . . . . . . . . . M12r exp( j12r ) 0 L dq r 2 0 . . . . . . . . . . . . . . . 0 M12r exp( j12r ) 0 L dq r 2 . . . . . . . . . . . . . . . . . . . . . . . . M exp( j ) 0 0 M2 N exp( j2 N ) M1Nr exp( j1Nr ) 0 M2Nr exp( j2Nr ) 0 1N 1N 0 M1 N exp( j1 N ) M2 N exp( j2 N ) 0 0 M1Nr exp( j1Nr ) 0 M2Nr exp( j2Nr )
(L 1 M1 )
M11 exp( j11 )
M1 N exp( j1 N ) 0 M2 N exp( j2 N ) M2 N exp( j2 N ) 0 M1Nr exp( j1Nr ) 0 0 M1Nr exp( j1Nr ) M2Nr exp( j2Nr ) 0 0 M2Nr exp( j2Nr ) . . .. . . . . . L dq r N 0 0 L dq r N 0
M1 N exp( j1 N )
j p1 (L 1 M1 ) 0 0 0 0 0 . . . . . . . . . 0 0
0 j p1 (L 1 M1 ) 0 0 0 0 . . . . . . . . . 0 0
0 0 j p2 (L 2 M2 ) 0 0 0 . . . . . . . . . 0 0
0 0 0 0
j p1 M11 exp( j11 ) 0 0 0 0 . . . . . . . . . 0 0
j p2 (L 2 M2 ) j p2 M21 exp( j21 ) 0 . . . . . . . . . 0 0
j p2 M21 exp( j21 ) 0 0 0 . . . . . . . . . 0 0
j p1 M11 exp( j11 )
j p1 M1 N exp( j1 N ) 0
Qc =
j p1 M1 N exp( j1 N ) 0 j p2 M2 N exp( j2 N ) j p2 M2 N exp( j2 N ) 0 0 0 0 0 . . . . . . . . . . . . . . . . . . 0 0 0 0
(4.6)
91
where the symbols used in Mc and Q c are the same as those used in (3.17-3.20), (3.24) and (3.263.29), and M1ir = The torque equation was given by (3.38) for the d-q model. Repartitioning the currents into stator
2 2 M1ia + M1ib , 1ir = arctan M1ib /M1ia .
and rotor quantities, and noting that as all components in (3.38) are real, the transpose may be replaced by complex conjugate transpose. (3.38) may therefore be written as: 1 i dqs Te = 2 i dqr
0 Q sr dq
Q dqsr 0
i dqs i dqr
(4.7)
Applying the complex transformation to the currents in this equation gives: 1 i cs Te = 2 i cr
0 Q sr c
Q csr 0
i cs i cr
(4.8)
where i cs1 , i cs2 , i cr are the symmetrical components of the stator 1, stator 2 and rotor currents. Q csr denotes the stator-rotor portion of Q c as dened in (4.6). Q csr is therefore given by: Q csr = j p1 M11 exp( j11 ) 0 0 j p2 M21 exp( j21 ) 0 j p1 M11 exp( j11 ) 0 j p1 M1 N exp( j1 N ) 0 0
j p2 M21 exp( j21 )
j p2 M 2 N
j p1 M1 N exp( j1 N ) 0 j p2 M2 N exp( j2 N ) exp( j2 N ) 0
(4.9)
If we now consider a re-ordering of the states of (4.4) using a permutation matrix (hence orthogoThis can be achieved with the permutation matrix, 1 1 1 . . . . . . . . . . . . . . . P = 1 1 . . . . . . . . . . . . . . . where a single denotes a zero.
+ + nal). Particularly we re-order the states so that the new order becomes: i s1 , i s2 , ir+ , , ir+ , i s1 , i s2 , ir , , ir . 1 1 N N
P:
1 . . . . . . . . . . . . . . . 1 1 . . . . . . . . . . . . . . . 1
(4.10)
92
to see that the transformed matrices, P Mc P T , P Rc P T , P Q c P T are block diagonal. Therefore it is immediate that, as expected [81], the different sequence components are orthogonal, that is a positive sequence does not affect negative sequence components and vice-versa:
If (4.4) is multiplied from the left by P and the substitution i c = P T i c p made, then it is easy
vc p1 vc p2
= P Rc P T + r P Q c P T
+ Mc
i c p1 i c p2
+ P Mc P T
d i c p1 dt i c p2
(4.11) (4.12) (4.13) (4.14)
P Mc P T = P Qc P T = P Rc P T =
0 Q+ c 0
+ Rc
0 + Mc 0 c Q+ 0 c R+
where i c p1 , i c p2 are upper and lower partitions of i c , and similarly for vc p1 , vc p2 . Also:
+ Mc =
0 (L 2 M2 ) M21 exp( j21 ) M11 exp( j11 ) M21 exp( j21 ) L dq r 1 M12r exp( j12r ) . . . . . . M12r exp( j12r ) L dq r 2 . . . . . . . . . . . . M1 N exp( j1 N ) M2 N exp( j2 N ) M1Nr exp( j1Nr ) M2Nr exp( j2Nr ) Q+ = c
(L 1 M1 )
M11 exp( j11 )
M2 N exp( j2 N ) M1Nr exp( j1Nr ) M2Nr exp( j2Nr ) . .. . . . L dq r N (4.15)
M1 N exp( j1 N )
j p1 (L 1 M1 ) 0 j p1 M11 exp( j11 ) 0 j p2 (L 2 M2 ) j p2 M21 exp( j21 ) 0 0 0 . . . . . . . . . . . . . . . . . . 0 0 0
j p2 M2 N exp( j2 N ) 0 . . . . . . 0
j p1 M1 N exp( j1 N )
(4.16)

+ Rc =
93
Rdq1 0 0 . . . . . . 0
0 Rdq2 0 . . . . . . 0
0 0
0 0 0 . . . 0
Rdqr 1 0 .. . 0 . .. . . . 0
.. . .. .
0 Rdqr N
Rdq1 0 = 0 Rdq2 0 0 where:
0 Rdqr Rdqr 1 0 . . . 0 0 .. .. . . .. . .. 0 .
(4.17)
Rdqr
Rdqr N
0 . . . 0
In the BDFM, however, the special rotor design cross-couples positive and negative sequence components from the different stator elds. Recall that in chapter 3 it was argued that rotor states could be transformed into d-q axes using the stator 1 pole number transformation matrix, rather then using two transformation matrices and then recombining. The effect of doing this is to, in effect, reverse the phase sequence of stator 2. If we had used the stator 2 pole number transformation matrix then the stator 1 phase sequence would have been reversed. This is not solely a mathematical phenomenon, it corresponds to the physical reality that if the phase sequences of the stator supplies are the same, then the frequencies of the stator elds viewed in the rotor reference frame are in opposite directions, as shown in section 1.2. Considering now, the torque equation. After transformation by P, the torque equation becomes: 0 Q+ csr 0 Q+ csr 0 0 c Q+ sr 0 c Q+ sr 0
1 i c p1 Te = 2 i c p2
i c p1 i c p2
(4.18)
where: Q+ = csr j p1 M11 exp( j11 ) j p1 M1 N exp( j1 N ) j p2 M21 exp( j21 ) (4.19)
j p2 M2 N exp( j2 N )
and i c p1 and i c p2 are the top and bottom partition of the symmetrical component current vector, i c under the transformation P. We now consider the supply voltage in the transformed state. Recall that the voltage must be rst transformed into d-q-0 axes, then into symmetrical components, removing the zero sequence state, and nally re-ordered by P. The supply voltage is assumed to be a balanced three phase set, which
94
can be represented by: V cos ((t))

2 3 4 3
vs = V cos (t) V cos (t)
(4.20)
where (t) is some arbitrary function of time, in steady state it will generally be t, and note that V may also be a function of time. This may be transformed to d-q-0 axes using (3.1) or (3.2) if this is the stator 2 supply. For this example we use (3.1): V cos ( p1 r (t))
vdqs =
3 V sin ( p1 r (t)) 2 0
(4.21)
now remove the zero sequence voltage, and converting to sequence components using (4.2) gives:
+ vs vs
v cs =
3 V exp ( j ( p1 r (t))) 2 V exp ( j ( p1 r (t)))
(4.22)
is composed of two separate, yet similar, equations, one supplied with v c p1 the other supplied by vc p2 = vc p1 . Therefore from (4.12) it is clear that i c p2 = i c p1 . Hence the torque can be written as:
+ Therefore it is clear that vs = vs . It is signicant that the equation (4.12), shows that the BDFM
1 i c p1 Te = 2 i c p2
0 1 = i c p1 2 Q+ csr =
0 Q+ csr 0 c Q+ sr 0
Q+ csr 0 c Q+ 0
sr
Q+ csr
0 1 i c p1 + i c p1 c 2 Q+ sr
i c p1 1 0 = i c p1 2 Q+ csr ic
p2
Q+ csr 0
0 1 i c p1 + i c p2 c 2 Q+ sr 0 Q+ csr Q+ csr 0
c Q+ sr 0
i c p2
c Q+ sr 0
i c p1 =
i c p1
i c p1
i csp1 Q + i cr p1 csr
i csp1 Q + i cr p1 + i cr p1 Q + i csp1 = csr csr
c i csp1 Q + i cr p1 + i csp1 Q + i cr p1 = 2 csr sr
(4.23)
Hence the BDFM may be fully described, in balanced three phase conditions, using:
+ + vc p1 = (Rc + r Q + )i c p1 + Mc c
di c p1 dt
(4.24)
with the torque given by (4.23). Note that this is a full dynamic model, which is mathematically equivalent to the full d-q model under balanced conditions.
4.3 Steady-state Equivalent Circuit Representation
95

We will now show that under steady-state conditions the equations lead to a symmetric impedance matrix, which means that a simple equivalent circuit comprised of passive components exists. Table 4.1 compares the rms values and frequencies of the voltages under the various modelling stages presented. An immediate consequence of table 4.1 is that if the positive sequence stator 1 1 = p2 r + 2 , which implies the familiar condition on the rotor speed for BDFM synchronous 1 + 2 p1 + p2 Sequence quantity rms freq. (rad/s) current is to be equal in frequency to that of the negative sequence stator 2 current, then p 1 r
operation:
r =
(4.25)
Phase quantity rms freq. (rad/s) Stator 1 V1 Stator 2 V2 2 1
d/q quantity rms freq. (rad/s)
3V1
p1 r 1
3 V 2 1
p1 r 1
3V2
p2 r 2
3 V 2 2
p2 r 2
Table 4.1: Comparison of rms values and frequencies for the different models presented under a 3 phase balanced supply in steady-state: coupled-circuit model, d-q axis model, complex sequence components model, the latter two models being in the rotor reference frame Therefore from (4.24), when (4.25) holds all the non-zero voltage sources are at a frequency of by splitting (4.24) into real and imaginary components and taking the frequency response (Fourier transform). This, of course, makes physical sense, as the d-q (and hence symmetrical components) reference frame chosen is in synchronism with the rotor. The frequency of the currents in the rotor reference frame under steady state conditions we denote by s : s = p1 r 1 = p2 r + 2 In steady-state conditions, with (4.25) in effect, (4.24) becomes: 3 3 + + Vc = (Rc + r Q + + js Mc ) Ic c 2 2 where
3 V 2 c
1 p1 r . Therefore the currents, i c p1 must also be at the same frequency, this can be most easily seen
(4.26)
(4.27)
and
3 I 2 c
of current and voltage. N is number of sets of S = p1 + p2 rotor coils in the particular machine. We
(Vc , Ic C N +2 ) are vectors of appropriate steady-state sequence components
96
now dene: Mcs1r = M11 exp( j11 ) M1 N exp( j1 N ) Mcs2r = M21 exp( j21 ) M2 N exp( j2 N ) L dq r 1 M12r exp( j12r ) M1Nr exp( j1Nr ) M12 exp( j12 ) L dq r 2 M2Nr exp( j2Nr ) r r Mcr = . . . .. . . . . . . . M1Nr exp( j1Nr ) M2Nr exp( j2Nr ) L dq r N
+ and note that from (4.15) and (4.16) r Q + + js Mc is given by: c
(4.28) (4.29)
(4.30)
+ r Q + + js Mc = c
j (L 1 M1 )(s r p1 ) 0
js Mcs1r
0 j (L 2 M2 )(s + r p2 )
js Mcs2r
j (s + r p2 )Mcs2r js Mcr
j (s r p1 )Mcs1r
(4.31)
We now dene the slip between stator 1 and the rotor, and stator 2 and the rotor as follows: s1 s2 1 / p1 r 1 p1 r s = = 1 / p 1 1 1 2 p2 r s 2 / p2 r = = 2 / p 2 2 2 (4.32) (4.33) (4.34)
2 s2 = s1 1
The notion of slip will be familiar for readers with a machines background. Slip is so-named because it is the ratio of the difference in rotational speed between the stator magnetic eld and the physical rotor speed, normalised by the rotational speed of the stator magnetic eld. Therefore the slip is zero when the machine rotates at the stator eld speed, and unity at standstill. Substituting (4.32), (4.33) and (4.31) into (4.27) gives (subject to 1 = 0 and s = 0) : Rdq1 0 0 (L 1 M1 ) 0 Mcs1r I1 s1 s1 V2 = j1 Rdq2 (L 2 M2 ) s2 Mcs2r + 0 0 I2 (4.35) 0 s2 s1 Mcs1r s1 Mcs2r s1 Mcr 0 0 Rdqr 0 Ir I1 V1 (L 1 M1 ) 0 Mcs1r Rdq 0 0 s 1 s I2 2 V2 = j1 + 0 2 s 0 (L 2 M2 ) Mcs2r R 0 s1 dq2 1 Ir 0 0 Rdqr /s1 0 Mcs1r Mcs2r Mcr V1 where
V1 V2 0
(4.36)
= Vc exp( js ), and similarly for I1 , I2 , Ir , that is V1 , V2 are the complex phasors.
97
In this new notation from (4.23) the torque equation becomes: Te = 2 =2 =2
Adopting a change of notation (4.36) becomes: Rc L cs1 0 Mcs1r V1 1 s 2 V2 = j1 0 L cs2 Mcs2r + 0 s1 PSfrag replacements Mcs1r Mcs2r Mcr 0 0
0
s2 s1
R c2 0
I1 0 I 2 Ir Rcr /s1 0
(4.37)
exp( js )I1 j p1 Mcs1r Ir exp( js ) + exp( js )I2 j p2 Mcs2r Ir exp( js ) j I1 p1 Mcs1r Ir + j I2 p2 Mcs2r Ir I1 p1 Mcs1r Ir I2 p2 Mcs2r Ir
(4.38)
Now (4.37) is in a form which admits an equivalent circuit representation. Figure 4.1 shows an equivalent circuit representation. The terms M cs1r 1 , Mcs1r 2 , , Mcs1r N are the elements of Mcs1r . Mcs2r 1 , Mcs2r 2 , , Mcs2r N are the elements of Mcs2r . Mcsr 11 , Mcsr 12 , , Mcsr N N are the elements of Mcr . And Rcr 1 , Rcr 2 , , Rcr N
R1
I1
Mcs1r 1 j1 L cr 1 Mcs1r 2
Ir1 Rcr 1 s1
V1
j1 L cs 1 Mcs1r N Mcr 12 j1 L cr 2 Mcr 1N I2 Mcs2r 1 Mcr 2N . . . Ir N j1 L cr N Mcs2r N Rcr N s1 Ir2 Rcr 2 s1
R 2 s2 s1
V2
s2 s1
j1 L cs 2
Mcs2r 2
Figure 4.1: Coupled coils Equivalent Circuit for BDFM model with N sets of rotor circuits Note that as there will, in general, be phase offsets between stator and rotor, and between the different rotor circuits, the mutual inductance terms are complex, rather than real. This means that the equivalent circuit is unusual in that there will be a phase offset at the voltage terminals. This is made explicit in gure 4.2, following the dot convention for the polarity of the mutual inductance terms (see e.g. [30] for details) between currents forming the input and output of each coupled coil. Physically this corresponds to the physical phase offset between coil sets in the machine. It is worth noting that in the case of a nested-loop design rotor, the matrix M cr is purely real, as the off-diagonal S S
98
sub-matrices in the coupled circuit rotor-rotor mutual inductance matrix will be symmetric circulant matrices, rather than just circulant matrices.
PSfrag replacements I1 e j M I2
V1 e j
jL 1
jL 2
V2
V1 e j = jL 1 I1 e j + jM I2 V1 V2 L1 Me
j
V2 = jL 2 I2 + jMe j I1 = j Me j L2 I1 I2
Figure 4.2: Circuit equivalent of coupled coils with complex mutual inductance terms
4.4 Equivalent circuit for the BDFM with a single set of rotor coils
We now turn our attention to a sub-class of BDFMs having either a single set of rotor coils, or equivalently a nested-loop design rotor to which the model reduction method described in section 3.4.2 has been applied. While equivalent circuits have been presented for BDFMs with a single set of rotor coils, they have been for specic machines. Both Li et al. [60] and Gorti et al [41] give an equivalent circuit for a 6/2 conguration nested loop machine, however it is not clear how the model order has been reduced. It is most likely that the method of summing the contributions from each loop given in [64, 63] was used (see section 3.5 for details). We will show that the circuit derived for a specic BDFM conguration in [60, 41] is in fact the correct form for any BDFM of this sub-class, that is with a single set of rotor circuits. Furthermore we will synthesize the equivalent circuit in such as way as to preserve the physical meaning of components, as far as possible. A BDFM with a single set of rotor circuits is a special case of the general case previously presented in section 2.7. In this case the rotor-rotor coupling term, Mcr is a real scalar. From gure 4.2 and (4.37) the steady-state equations may be written as:
V1 e j Mcs1r
s 2 V2 e j Mcs2r = j1 0 L cs2 |Mcs2r | + 0 s1 0 0 |Mcs1r | |Mcs2r | Mcr
L cs1
|Mcs1r |
R c1
s2 s1
0 R c2 0
0 0
I2 e j Mcs2r Rcr /s1 Ir
I1 e j Mcs1r
99
taking the complex conjugate of each side (recalling that now the mutual inductance and resistance terms are real) gives: V1 e j Mcs1r Rc L cs1 0 |Mcs1r | s 1 j Mcs2r = j 2 V2 e L cs2 |Mcs2r | + 0 s1 1 0 |Mcs1r | |Mcs2r | Mcr 0 0 0
s2 s1
R c2 0
I1 e j Mcs1r 0 I2 e j Mcs2r r I Rcr /s1 0
(4.39)
It will now be convenient to change notation into a more concise form, dening: V1 V2 I1 I2 Ir V1 e j Mcs1r V2 e j Mcs2r I1 e j Mcs1r I2 e j Mcs2r Ir (4.40)
Hitherto the steady-state BDFM equations presented have been functions of sequence phasors. From table 4.1 it is clear that the only difference in magnitude between a per-phase quantity and a sequence quantity is a scalar multiple. Furthermore, as the steady-state equations are in terms of phasors, the equations hold regardless of the reference frame chosen. A change of reference frame simply requires a different phase on V and I . Therefore the electrical equations may be considered per-phase equations without modication. Each current in the torque equation, however, must be multiplied by the factor
3 2
as given in table 4.1, if per-phase quantities are used.
Using the simplied notation the per-phase steady-state BDFM equations become: V1 L cs1 0 |Mcs1r | I1 R c1 0 0 s s2 2 V2 = j1 0 0 I2 L cs2 |Mcs2r | + 0 s1 Rc2 s1 0 0 0 Rcr /s1 |Mcs1r | |Mcs2r | Mcr Ir and multiplying the current phasors in (4.38) by phase equivalent circuit is (noting that for a, b C, Te = 3 a per-phase representation.
3 2
(4.41)
the torque equation corresponding to the per{ab} = a b ): (4.42)
I1 p1 |Mcs1r |Ir + I2 p2 |Mcs2r |Ir
It it worth noting that equation (4.42) is identical to eq. (7) of [41], subject to the factor of 3 as (4.42) Some useful alternative forms of (4.42) may now be derived. From the the rst row of (4.41):
V1 I1 = j1 L cs1 I1 I1 + j1 I1 |Mcs1r | Ir + Rc1 I1 I1 j1 I1 |Mcs1r | Ir
(4.43) (4.44)
V1 I1 Rc1 | I1 |2 =
100
and from the second row of (4.42): s2 s2 V2 I2 = j1 L cs2 I2 I2 + j1 I2 |Mcs2r | Ir + Rc2 I2 I2 s1 s1 j1 I2 |Mcs2r | Ir Rc2 | I2 |2 = (4.45) (4.46)
s2 s1
V2 I2
noting that (4.42) can be written: Te = 3

j I1 p1 |Mcs1r |Ir j I2 p2 |Mcs2r |Ir
(4.47)
hence, substituting (4.43) and (4.46) in (4.47): Te =

PSfrag replacements
3 p1 1
3 p 2 s2 V1 I1 Rc1 | I1 |2 1 s 1
V2 I2 Rc2 | I2 |2 (4.48)
We may write the per-phase BDFM equations (4.41) as an equivalent circuit as shown in gure 4.3. This form of the equivalent circuit is the same as that found in [60, 41].
Ir R c1 I1 |Mcs1r | Ir
V1
j1 L cs1
R c2 s2 s1
I2
j1 Mcr
Rcr s1
s2 V2 s1
j1 L cs2
|Mcs2r |
Figure 4.3: Coupled Coils Per-Phase Equivalent Circuit for Single Rotor Circuit BDFM
4.4.1
Physical interpretation of parameters in the per-phase equivalent circuit model
The parameters used in (4.41) are calculated, in the rst instance, from the coupled-circuit method described in section 2.5.1, using in particular equation (2.22). These parameter values then undergo transformation to d-q axes, model reduction (if appropriate) and then conversion to symmetrical components as described in sections 3.3, 3.4.2, and 4.2. However, as noted in section 2.4.3, it is also possible to calculate machine parameters from the Fourier series representation of mutual inductance using (2.27) instead of (2.22). For each harmonic a separate inductance matrix will be given, and an innite sum of the harmonic terms will, of course, give the same result as the direct calculation method. As the transformation to d-q axes, model
101
reduction and transformation to symmetrical components are linear operations, the transformations may be directly applied to each harmonic component. Note that the matrix used in model reduction will be the one derived for the full system, and this model reduction technique then applied to each harmonic in turn. Furthermore the leakage inductance terms, as described in section 2.5.2 may be retained separately and converted to symmetrical components via the same procedure. We may therefore split the inductance terms into their constituent parts: Mcr = L r1 + L r2 + L rl + L rh L cs1 = L 1 f + L 1h + L 1l L cs2 = L 2 f + L 2h + L 2l These terms correspond to inductance due to leakage, and inductances due to particular space harmonic components. Inductance due to leakage is self-inductance arising from ux which does not cross the air gap, the leakage terms are L rl , L 1l , L 2l . Of the other terms, we know from section 2.6 that for the class of BDFMs considered, the rst non-zero component of the harmonic Fourier series of the inductance matrices will be p1 for stator 1 and p2 for stator 2. These harmonic terms, when suitably transformed, become L 1 f and L 2 f respective, the fundamental space harmonics for the two stator windings. On the rotor, in general there will be a rich set of harmonics. From Lemma 2.9 we know that the rotor-rotor mutual inductance term can only comprise of harmonics of p 1 and p2 , the fundamental stator eld, as the rotor is not directly energised, however in general the magnitude of these harmonic terms is signicant. L r1 and L r2 are the p1 , p2 pole number harmonic components of the rotor selfinductance, that is, the harmonics which link to the stator 1 and 2 respectively. The L rh , L 1h , and L 2h terms are higher order space harmonic self-inductance terms, which are not included elsewhere, as given by (4.49). Using the relationships in (4.49), gure 4.3 may be re-written as gure 4.4. Using standard results, gure 4.4 may be transformed into a T equivalent form using the relationships shown in gures 4.5 and 4.6 (see, for example [20]). Note that in a network sense, the choice of N is entirely arbitrary. However we will choose the turns ratio in each case so that it is the effective turns ratio for each coupling. The effective turns ratio, is the square root of the ratio of the fundamental space harmonic self-inductance terms for the rotor and the stator. Equivalently, it is the ratio of the amplitudes of the fundamental air gap mmf waveforms produced by unit current in the stator and rotor [100, p. 381]. Therefore the effective turns ratio for stator 1 - rotor is: L1 f L r1
(4.49)
N1 =
(4.50)
102
I1 R1

j1 L 1l j1 L 1h |Mcs1r | Ir Rr s1 j1 L rh
V1
j1 L 1 f
j1 L r1
s2 I 2 R 2 s1
j1 L 2l
j1 L 2h
|Mcs2r |
V2
s2 s1
j1 L 2 f
j1 L r2
j1 L rl
Figure 4.4: Coupled coils Equivalent Circuit Separating Mcr , L cs1 , L cs2 and for stator 2 - rotor: N2 = L2 f L r2 (4.51)
Because L 1 f , L r1 and L 2 f , L r2 represent couplings between same pole number elds with no parasitic effects, the coupling in each case is perfect, therefore: |Mcs1r | = and |Mcs2r | =
M
L 1 f L r1
(4.52)
PSfrag replacements
L 2 f L r2
(4.53)
I1
I2
V1
jL 1
jL 2
V2
Figure 4.5: Coupled coils Therefore converting the coupled coils arrangements of gure 4.4 using gure 4.6 gives the equivalent circuit as shown in gure 4.7. Note that from (4.52) and (4.53) the series terms in gure 4.6 are zero. It is helpful, for physical insight, to note that gure 4.7 is equivalent to gure 4.8. This can be seen by noting that from (4.32) and (4.33) s = s1 1 . Then as the sign of E 1 and Er 1 can be
PSfrag replacements
N :1 I1 PSfrag replacements V1 j(L 1 M N ) I2 /N jM N E1 E1 = E2 N E2 I2
103
j(L 2
M ) N
V2
Figure 4.6: Transformer equivalent of coupled coils
I1
R1
j1 (L 1l + L 1h ) j1 |Mcs1r |N1
Ir N1
N1 : 1
Ir
Rr /s1
V1
E1 Ir N2 E 1 = E r1 N1 N2 : 1
E r1
j1 L rh
I2 R 2 s2 s1
j1 (L 2l + L 2h ) j1 |Mcs2r |N2
Ir
j1 L rl
V2
s2 s1
E2 E 2 = E r2 N2
E r2
Figure 4.7: BDFM Equivalent circuit with separate inductance components
arbitrarily assigned, gure 4.8 becomes equivalent to gure 4.7 if the direction of positive V1 , I1 is reversed. Figure 4.8 may be recognised as two normal induction machine equivalent circuits with rotors connected together, which is conceptually how the BDFM operates.
Figure 4.7 may be referred to stator 1 or stator 2, as is customary for the standard induction machine. We will refer all quantities to stator 1, although they could easily be referred to stator 2. The use of the modier denotes that the quantity is referred. The relationships between the
104
I1 R1

j1 L 1 Ir /N1 N1 : 1 Ir js L r1 Rr
V1
j1 L m 1 E1 E 1 = E r1 N1 s1 Ir js L r2 E r1
I2
R2
j2 L 2
Ir /N2
N2 : 1
V2 j1 |Mcs1r |N1
j2 L m 2 E2 E 2 = E r2 N2 s2 E r2
j1 |Mcs2r |N2
Figure 4.8: BDFM Equivalent represented as two induction machine connected together. Equivalent (L 2l + L 2h ), L r1 + L r2 = L rh + L rl with gure 4.7 is achieved with L m 1 = |Mcs1r |N1 , L m 2 = |Mcs2r |N2 , L 1 = (L 1l + L 1h ), L 2 =
parameter of gure 4.7 and gure 4.9 are: L 1 = (L 1l + L 1h ) L 2 = (L 2l + L 2h )

2 N1 2 N2
2 N1 2 N2
2 L r = (L rl + L rh )N1
R2 = R2
2 Rr = R2 N1
(4.54) from (4.50) and (4.52) N2 L2 f 1 2 N2 from (4.51) and (4.53)
L m 1 = |Mcs1r |N1 = L 1 f L m2 = V2 = N2 (|Mcs2r |N1 ) 1 2 N2 =
N1 V2 N2 N2 I2 = I2 N1
4.4.2
Development of Torque Equations from the Equivalent Circuit
It is interesting to now consider how the conversion of electrical to mechanical energy is represented in the equivalent circuit. This leads to development of expressions for the electrical torque produced by the machine. Such expressions will be developed in terms of supply currents, voltage, and a combination of the two. During these derivations two voltages, Vr1 and Vr2 are introduced, and are dened
Ir
I1 R1 j1 L 1 j1 L r Rr /s1 Ir j1 L 2 R 2 s2 I s1 2 s2 V s1 2
105
V1
j1 L m 1
Vr1
Zr
Vr2
j1 L m 2
I2
V2
N1 : 1
1 : N2
Figure 4.9: BDFM Referred Per-Phase Equivalent Circuit in gure 4.9. Working in terms of these voltages leads to considerable algebraic simplications, yet prevents the needs for any simplication to the overall model, as analysis can be performed using these intermediate variables, and then the terminal quantities backed-out. Vr1 and Vr2 are physically meaningful in that they represent the voltages across the magnetizing inductances of the two windings. Thus a limit on the magnetization of the laminations may be written in terms of a limit on Vr1 and Vr2 . This point will be discussed further in section 4.7. By applying the principle of power conservation to gure 4.9 we get:
3 {V1 I1 } + 3
R s2 s2 2 {V2 I2 } = 3|I1 |R1 + 3|Ir 2 | r + 3|I2 2 |R2 s1 s1 s1
(4.55) (4.56)
2 3 {V1 I1 } + 3 {V2 I2 } = 3|I1 |R1 + 3|Ir 2 |Rr + 3|I2 2 |R2 + r T
where T is the torque generated by the machine, with positive values for motoring, and the 3s arise because gure 4.9 is a per-phase equivalent circuit. From gure 4.9 we may dene Pr1 , Pr2 , the power delivered at the rotor terminals from stator 1 and 2 respectively: Pr1 Pr2
2 3 {Vr1 I1 } = 3 {V1 I1 } 3|I1 |R1 s1 3 {Vr2 I2 } = 3 {V2 I2 } 3|I2 2 |R2 s2
(4.57) (4.58)
Subtracting (4.55) from (4.56) gives the fraction of power not dissipated in the rotor and stator resistances, i.e. the power which is converted to mechanical power: 3 {V2 I2 } 1 s2 s1 = 3|Ir 2 |Rr 1 1 s1 + 3|I2 2 |R2 1 s2 s1 + r T (4.59)
106
Rearranging, and substituting (4.58) for the I2 terms gives the torque as: T = Pr2 r 1 s2 s1 3 2 1 |I |R 1 r r r s1 (4.60)
Alternatively, substituting (4.57) and (4.58) into (4.55) and (4.56) gives: Pr1 + s2 R Pr2 = 3|Ir 2 | r s1 s1 (4.61) (4.62)
Pr1 + Pr2 = 3|Ir 2 |Rr + r T the torque: T = p1 p 2 s2 Pr1 Pr 1 1 s 1 2
Combining (4.61) and (4.62) and with (4.32), (4.33), (4.25), (1.2) gives the following expression for
(4.63)
which, using (4.57) and (4.58) and some sustained manipulation may be written as: T = 3 p1 3 p 2 s2 V1 I1 R1 |I1 |2 + V2 I2 + R2 |I2 |2 1 1 s 1 3 p 2 s2 3 p1 s1 = Vr1 I1 + Vr2 I2 1 1 s 1 s2 3( p1 + p2 )s2 Rr p1 |Ir |2 + =3 R2 |I2 |2 V2 I2 1 s 1 1 s 1 3( p1 + p2 ) Rr p1 |Ir |2 + =3 R2 |I2 |2 V2 I2 1 s 1 1 s c Rr ( p1 + p2 ) V1 I1 R1 |I1 |2 3 p2 |Ir |2 =3 1 s 1 1 2 1 ( p1 + p2 )r s1 = = s2 1 1 (4.64) (4.65) (4.66) (4.67) (4.68)
where: sc = expected. It is convenient as this stage to recall the natural speed, n for the BDFM, given in (1.2). n = 1 p1 + p2 (4.69)
Note that (4.64) is the same as (4.48), the torque equation derived from the coupled circuit model, as
Referring back to the denition of BDFM synchronous speed (4.25), and slips (4.32), (4.33), it is straightforward to see that the rotor speed, r can be written in terms of a deviation from natural speed: r = 1 = s1 s2 s1 n s2
s2 1 n s1
(4.70)
107
Using (4.70), (4.32) and (4.58) the expression for the torque, (4.60), may be written as: T = or more generally because T = 3 p1 1 Vr1 Ir + 3 {Vr2 I2 } p1 + 3|Ir 2 |Rr n 1 s 1 Vr1 Ir and Vr2 I2 = (4.71) Vr2 Ir , from (4.65): (4.72)
Vr1 I1 =
3 p2 1
Vr2 Ir
writing Ir in terms of Vr1 and Vr2 : Vr Vr Vr Vr 3 p2 3 p1 2 1 2 1 Vr1 Vr2 + = 1 Zr 1 Zr 2 2 Vr1 Vr Vr2 Vr |Vr | 3 p1 |Vr | 3 p2 2 1 2 + = + 1 + (4.73) 1 Zr Zr 1 Zr Zr 3 p1 |Vr Vr | |Vr |2 |Vr1 |2 3 p2 |Vr1 Vr2 | = 1 2 cos( + ) + 2 cos() + cos() + cos( ) 1 |Z r | |Z r | 1 |Z r | |Z r | (4.74) = 3 |Vr1 Vr2 | 1 |Z r |
2 2 p1 + p2 2 p1 p2 cos 2 sin arctan
3 1
( p1 p2 ) cos + ( p1 + p2 ) sin p1 |Vr1 |2 p2 |Vr2 |2 cos() |Z r | |Z r |
(4.75)
where = Z r and = Vr1 Vr2 , the phase angle between Vr1 and Vr2 .
Furthermore note that from (4.75) for any given torque, there are either 0, 1, or 2 values of , for a maximum and minimum torque envelopes, one solution when the torque is precisely on these torque
Hence the maximum (and minimum) torques occurs when ( p 2 p1 ) tan = ( p1 + p2 ) tan .
given set of |Vr1 , |Vr2 |, 1 , Z r , p1 and p2 . There will be zero solutions when the torque is outside the
if one these values of corresponds to a stable operating point, the other typically corresponds to an unstable operating point. This fact is explored further in chapter 7.
envelopes, and two solutions in all other cases. The two solutions will be 0 and 0 . Physically,
A further variant of the torque expression is obtained by re-writing the rst term of equation (4.71) in terms of Vr1 and Vr2 . As no real power is dissipated in L m 2 , from gure 4.9 it is straightforward Vr Vr1 therefore, from to see that Vr2 I2 = Vr2 Ir . From gure 4.9 Ir may be written as 2 Zr 4.71: T = Vr2 Vr Vr2 Vr 3 p1 2 1 + 3|Ir 2 |Rr n Zr 1 s 1 2 |Vr2 | p1 3 |Vr1 Vr2 | cos( ) cos() + 3|Ir 2 |Rr = n |Z r | |Z r | 1 s 1
(4.76)
where is the angle between Vr1 and Vr2 and = Z r , as before.
108
PSfrag replacements
Vr1
Z r Ir
Ir Vr2
Figure 4.10: BDFM Rotor Circuit Phasor Diagram
4.4.3
Phasor diagram for rotor branch circuit
Figure 4.10 shows the phasor diagram for the rotor branch of the circuit shown in gure 4.9. Figure 4.10 can be used to determine the power and VAR ow in and out of the rotor terminals on gure 4.9. We therefore dene: Q r1 Q r2 3 3 Vr2 Ir
Vr1 Ir
=3
V2 I2
V1 I1 3|I1 |2 L 1 1 3
s1 =3 s2
|Vr1 |2 L m 1 1 |Vr |2 s1 s1 + 3|I2 |2 L 2 1 + 3 2 s2 L m 2 1 s 2
(4.77) (4.78)
Hence if Vr2 , Vr1 , and Ir are known then the VARs produced/consumed at the machine terminals terminal VARs will also be minimised. Therefore specications on the power factor at the terminals can be translated into specications at the rotor branch, which is completely described by the phasor diagram in gure 4.10. can be calculated. For example if Q r1 and Q r2 are minimised for a given |Ir |, |Vr1 | and |Vr2 |, then the
4.5 Equivalent Circuit Numerical Simulation

Figures 4.11, 4.12 and 4.13 show the maximum torque envelope for the prototype BDFM previously described with the nested loop rotor. The machine is excited with a constant 240V, at 50Hz on the and (Vr2 ) is chosen such that the torque is maximum or minimum respectively. The base value power (4 pole) winding, and the 8 pole winding excitation is chosen such that: |Vr2 | is held constant,
chosen is such that at 0rpm |V2 | = 240V , and from (4.25), 2 = 2 50. Maintaining a constant the impedance of Z r increases and therefore |Vr1 | changes, thus the 4 pole ux changes. This explains
|Vr2 | corresponds to maintaining constant ux on the 8 pole winding. However, as the s 1 decreases
the slight change in torque envelope. Because V1 is constant it is not possible to compensate for this change directly, rather is must be achieved indirectly by modifying V2 . Section 4.7 discusses the issues surrounding maintaining a constant magnetic loading for different relative amplitudes of 4 and 8 pole elds.
4.6 Equivalent Circuit Analysis
109
Figure 4.12 shows how the envelope changes with a 10% decrease in L rh , the rotor excess inducin changing the value of L rh this will have changed the 8 pole ux as |Vr1 | will have changed. In fact the change is a small (3%) decrease for positive torque and a 1% decrease for negative torque, thus an even greater change in torque envelope would have resulted if the ux level had been maintained. Figure 4.13, is similar to gure 4.12 except it shows the effect of a voltage limit on the inverter of (from (4.75). This gives a unique value Vr2 , however it will not necessarily put a bound on |Vr2 |,
200
tance due to spatial harmonics. The 4 pole ux (i.e. |Vr2 |) was kept the same in both cases. Note that
240V. It has been assumed that |V2 | is constant and (Vr2 ) is chosen so that the torque is maximized
which would be required to ensure that the magnetic loading is not exceeded.
150 Total Sync. Stator 1 Async. Stator 2 Async.
100 Torque (Nm)
50
50
100
150 0
500
Rotor Speed (rpm)
1000
1500
Figure 4.11: The synchronous and asynchronous components of torque on the maximum torque envelope, for constant |Vr2 | and |V1 |

The preceding analysis leads to a number of general points regarding BDFM operation. 1. The presented equivalent circuit holds for all BDFM operating modes where the shaft speed is constrained by equation (4.25). If this condition does not hold then the machine may be modelled as the sum of two equivalent circuits at different frequencies, one supplied from stator 1, the other from stator 2, and with the unsupplied stator short-circuited. This mode of operation is known as double cascade induction mode:
110
200
150
100 Torque (Nm)
50
Total Total, reduced Zr Lower Async. Upper Async.
50
100
150 0
500
Rotor Speed (rpm)
1000
1500
Figure 4.12: Torque envelope (blue) showing the effect of a 10% decrease in L rh (red) The preceding derivation requires (4.25) to hold. However, if this is not the case then by the superposition theorem, the machine may be modelled as a machine with stator 1 supplied and stator 2 short-circuited, and a machine with stator 2 supplied and stator 1 short-circuited. The true current in each branch is the sum of the contribution from each circuit. The average torque may be calculated using any of the per-phase torque equations derived, and the contribution from each circuit added. However the true torque will have a signicant ripple component, which cannot be calculated using the steady-state equivalent circuit. In this case the torque should be calculated from the full complex currents of the two frequencies concerned, rather than just phasors. 2. The BDFM may operate either as a p1 or p2 pole pair induction machine when only stator 1 or stator 2 is supplied and the unsupplied stator is open-circuit. This mode of operation is known as simple induction mode: It is immediate from gure 4.9 that if I2 = 0, then the machine reduces to a standard induction case of supply from stator 1. Clearly analogous results hold when supplied from stator 2. also noteworthy that a good BDFM design will inevitably lead to poor simple induction mode torque. This is because the second stator magnetizing inductance appears in series with the rotor leakage inductance, and typically in a well-designed machine the magnetizing inductance
motor equivalent circuit, but with the rotor leakage inductance equal to L r + L m 2 , in the
The torque reduces to the well-know expression for a standard induction machine (4.66). It is

200
111
150
100 Torque (Nm)
50
Total Total, reduced Zr Lower Async. Upper Async.
50
100
150 0
500
Rotor Speed (rpm)
1000
1500
Figure 4.13: Torque envelope (blue) showing the effect of a 10% decrease in L rh (red) with inverter voltage limit will be large in comparison to the leakage inductance. 3. The BDFM, when operating according to (4.25), is comprised of synchronous torques and induction machine-like asynchronous torques. This can be seen most clearly from equation (4.75). The last two terms are xed for a given operating speed, and rotor voltage magnitudes, as Z r is xed. These terms are induction machine like torque terms, in that they are xed for a given operating speed, regardless of the phase of the supply voltage. In contrast, the rst term is dependent on the sine of , the angle between the two rotor supply voltages. This term is reminiscent of the torque of a standard synchronous machine. 4. The asynchronous torques in the BDFM will always be of opposite signs and sum to zero if either, = /2, which means the rotor is purely inductive, or p 1 |Vr1 | = p2 |Vr2 |: These points come directly from equation (4.75). 5. Unlike the conventional induction machine, the BDFM would produce torque if the rotor resistance were zero. From (4.75), if Rr = 0 cos() = 0, therefore the only non-zero term in the torque expression is the rst term. It can be shown that, as = (4.76).
, 2
the term reduces to the rst term of
112
6. All torque in the BDFM falls to zero at the synchronous speed of stator 1: From gure 4.8, when s1 and s2 are both zero, then Ir is zero; hence from (4.47) the torque will be zero. From (4.32) s1 is zero at stator 1 synchronous speed, and therefore, from (4.26), s (the frequency of the rotor currents), must be zero. As stator 2 is supplied in such a way that the frequencies of the rotor currents induced match those induced by stator 1, then s 2 must also be zero. This can be seen from (4.32). A corollary to this point is that for a given BDFM with stator 1 and stator 2 (of course of different pole numbers) a wider speed envelope for BDFM operation is obtained if the lower pole number eld is connected as stator 1. 7. When stator 2 is short-circuited the BDFM operates in cascade induction mode, which is a special case of BDFM synchronous mode (that is, the stator 2 currents are at frequency 2 , given by (4.25)) similar to that observed when two induction machines are connected in cascade [49]: Equation (4.26) gives the frequencies of the rotor currents for the BDFM. As the rotor current frequency in the BDFM is the same for currents induced from both stator supplies, then even if the second stator is short-circuited, the stator 2 current frequency will be at 2 . Therefore cascade induction mode is a special case of BDFM synchronous mode, where the voltage Vr2 of equation (4.76) although non-zero is not an independent variable as it is in synchronous mode. stator 1 induction mode torque characteristic, the other may be thought of as the induction mode Equation (4.67) shows that, when V2 = 0 there are two terms in the torque equation, one is the characteristic of the cascade combination of the (effectively) two machines. (4.67) introduces the notion of cascade slip, sc which has a similar interpretation to standard slip, and for which the synchronous speed is n , (1.2), the natural speed, corresponding to the case that 2 = 0. 8. The cascade torque of a BDFM is inherent, and will be present on all functional real-world BDFMs. The absence of a cascade torque implies no BDFM synchronous torque, or that R 2 , the second stator resistance is zero. Point 7 illustrates the close relationship between cascade torque and synchronous torque, and therefore that its absence implies that there is no synchronous torque. Equation (4.67) shows that if R2 is zero then there can be no cascade torque. In practice R 2 cannot be zero, so the absence of cascade torque will generally imply that there will be zero synchronous torque. 9. The size of the available BDFM synchronous torque envelope for a machine in practical use excess harmonic reactance and resistance terms. is inversely proportional to the rotor impedance, |Z r | which comprises of leakage reactance, Figure 4.11 shows the three components of torque arising from the three terms of equations (4.75). Notice that the two asynchronous torque terms (marked as Stator 2 Async. and Stator 1 Async. on the diagram) are (and in general will always be) opposite in signs (immediate from
113
(4.75)), and furthermore will generally be small in comparison with the maximum synchronous torque for relatively large values of s1 . From gure 4.12 the total asynchronous torque is less than 10% of the maximum synchronous torque for s1 > machine with s1 >
1 , 3 1 . 3
We assume that we operate the
which is likely as it is desirable to operate the machine around natural

p2 . p1 + p 2
speed, as natural speed will correspond to s1 = of (4.76): Tmax
Under this condition the maximum
torque available is dominated by the rst term of equation (4.75) which reduces to the rst term 3 |Vr1 Vr2 | n |Z r |
(4.79)
ux densities of p1 and p2 pole number elds will be xed. It is not immediately obvious what rated ux density should be with two different pole number elds sharing the same iron laminations, however certainly there will be some upper limit, determined by what is considered a
tances for the machine (see gure 4.9). Therefore for any given values of |Vr1 | and |Vr2 | the
Now, signicantly, the voltages |Vr1 | and |Vr2 | are the voltages across the magnetizing reac-
reasonable degree of saturation. This point will be discussed further in section 4.7. Therefore the maximum torque available from a machine is inversely proportional to |Z r |. We may therefore conclude that the maximum rotor impedance for a given maximum output torque is: |Z rmax | = 3 |Vr1 Vr2 | n Tmax (4.80)
Now if we assume that, for any reasonable machine design, in normal operation Vr1 and Vr2 are close to the supply voltages (i.e. stator drops are small), then equation (4.80) can be used as a design guide. It should be noted, that because of the direction of power ow, the estimate will be an overestimate in some regions, and an underestimate in others, due to the stator drops. Figures 4.12 and 4.13 serve to illustrate the effect of the series rotor impedance, (Z r of gure 4.9). 10. The power factor of the stator 1 winding can be controlled by the magnitude of Vr2 in syntive the rotor, the greater the current required: chronous operation, and can always be made to lead, however for a given |Z r |, the more resis-
From gure 4.10, it is immediate that the phase of Ir can be controlled by varying the magnitude of Vr2 for a given . Furthermore, as 0 < <
, 2
for speeds less than the stator 1 synchronous
speed, it will always be possible to make negative. If is negative then for a sufciently large current it will be possible to make the power factor of stator 1 leading. It is clear that the closer rotor is purely resistive). is to 0, the larger the current required for a unity or leading power factor ( = 0 implies the
114
11. If the machine is to operate at a xed power factor on stator 1, then the smaller |Z r |, the larger the available torque, for a well-designed machine. From gure 4.10 a xed stator 1 power factor will imply a xed value of , if it can be assumed that the stator 1 series impedance is relatively small, which is likely to be the case in a well designed machine. If is xed then for any operating speed, the phasor diagram is fully detorque envelope increases, as the rate of decrease of |Z r | must always be greater than the corresponding rate of increase of |Vr2 |. Therefore the available torque increases, as we assume that the induction torques are small because is close to 90 . scribed, as Z r is constant. Therefore from equation (4.75), as |Z r | decreases, the synchronous
12. In the nested-loop design BDFM rotor the components of rotor inductance L rh and L rl are both signicant, and therefore directly impact the torque envelope: Recall that L rh is the additional inductance which is due to unwanted space harmonic components in the rotor. L rl is the additional inductance due to leakage effects - that is ux that does not cross the air gap as discussed in appendix C.1. Table 4.2 lists the calculated components of Z r for the prototype nested-loop design rotor. X rh ( ) X rl ( ) Rr ( )
Nested-loop design rotor 307 332 74
Table 4.2: Components of Z r for the prototype nested-loop design rotor at 50Hz From table 4.2 when the slip, s1 , is less than 0.11 then |Z r | is less than 1.41 times its value at
s1 = 1. Therefore for the majority of the available operating regime |Z r | is determined by the inductive components. Indeed, both L rh and L rl are signicant in determining |Z r |.
4.7 Magnetic Loading for the BDFM

As previously noted, to address the question of what rated magnetic ux density is for a BDFM is not a straightforward question because the iron circuit carries two different frequency magnetic ux density waves. So the question that arises is what, if any, combination of amplitudes of the two ux density waveforms constitute a constant magnetic loading? A conservative answer to this question would be the sum of the amplitudes of the ux density waveforms, as inevitably this peak value will be attained. However intuition suggests that this may lead to conservative results, as the peak value occurs relatively infrequently.
4.7 Magnetic Loading for the BDFM
115
The issue of magnetic loading for the BDFM has received little attention. It is believed that the only signicant contribution in this area is due to Broadway [16], however Broadway only discussed the issue for a specic example, rather than presenting any general conclusions. In conventional induction machines designers use a specic magnetic loading, dependent on the electrical steel chosen, which achieves the right balance between under use of the iron (and therefore a reduced output torque from the machine), and causing excess magnetizing current and therefore a poor power factor by driving the iron into saturation, see, for example, [92] for details. In the conventional induction machine, the specic magnetic loading is traditionally dened as the mean absolute ux per pole in the air gap of a machine. In a BDFM this denition becomes unworkable because of the double feed. We therefore propose the following generalisation of the denition: 1 B = lim T T
T 0
1 2
2 0
|B()|ddt
(4.81)
where B() is the ux density in the air gap (assumed uniform along the axis of the machine). It is clear that this denition is consistent with the concept of mean ux per pole, and the well known result of B = 2 2 Brms may be veried.
From similar reasoning to [115] ignoring harmonic elds, the magnetic elds in the air gap of the
BDFM may be written as: B() = B1 cos(1 t p1 ) + B2 cos(2 t p2 + ) These elds may be referred to the rotor reference frame as follows: = r t + from (4.26) we may write: Br () = B1 cos(s t + p1 ) + B2 cos(s t + p2 + ) Therefore, from (4.81) the specic magnetic load for the BDFM is: 1 B = lim T T s = 2
0 T 0
(4.82)
(4.83)
(4.84)
1 2
0
2 0 2
|B1 cos(s t + p1 ) + B2 cos(s t + p2 + )|ddt
(4.85) (4.86)
2 s
1 2
|B1 cos(s t + p1 ) + B2 cos(s t + p2 + )|ddt
This integral may be evaluated either analytically or numerically. It is useful to note the standard deviation of the time-varying spatial mean signal, as it gives an indication of the degree of ripple to be expected in the specic magnetic loading in a BDFM. It may easily be shown that when B 2 = 0 2 then B = 2 B1 and when B2 = B1 then B = 2 2 B1 , regardless of the value of p1 , p2 , s , .

Figure 4.14 shows how (4.86) varies with B2 , for unity B1 . Although a simple proof has not
been forthcoming, the same relationship (the green line) was found regardless of the pole number
116
combinations or offset angle chosen. 70 pole number combinations were tried under the standard one standard deviation away from the mean for the case of p 1 = 2, p2 = 4. Note that the standard deviation is strongly dependent on the pole number combination chosen, thus suggesting that specic magnetic loading for the BDFM may not be the best measure to determine the magnetic loading for all pole number combinations. Figure 4.14 also includes a plot of the magnetic loading if it is assumed that the total magnetic loading is the sum of loading for each individual eld, that is B = B1 + B2 and, for comparison, an alternative denition of the combined magnetic loading as B =
2 2 .
restrictions on BDFM pole numbers, p1 = p2 and | p1 p2 | > 1. The dashed green lines show the
This alternative can be thought of as the root-mean-square of the rms values of the air gap ux density
2 2 B1 + B2 .
waves, scaled by
1.2
New generalised B 2 2 B = B1 + B2
B = B1 + B2
1.1
PSfrag replacements
0.8
B
0.9 0.7 0.6 0
0.1
0.2
0.3
0.4
0.5
B2
0.6
0.7
0.8
0.9
2 Figure 4.14: B, the specic magnetic loading a 4/8 BDFM for B1 = 1 (hence B1 = ) as B2 varies between 0 and 1 (hence B2 varies from 0 to 2 ) . Three variations in the denition are shown, the
summation of the B for each eld, the new generalised and averaged B and B = analytical alternative
2 2 B1 + B2 , as an
From gure 4.14 it is clear that the generalised magnetic loading is considerably less conservative than assuming that the magnetic loadings for each individual eld should be summed. Furthermore it is signicant that the variance of the generalised magnetic loading about its mean was small for certain pole number combinations, thus indicating that the average eld is well distributed around the circumference of the machine under normal operating conditions. This serves to increase condence that the generalised magnetic loading is a good measure of iron utilisation. However the alternative
4.8 Conclusion
117
tage that it has a simple closed-form representation, which lends itself to analytical calculations. The convenience of this alternative representation is that a specication on the magnetic loading can be translated into a specication in terms of Vr1 and Vr2 . It should be noted, however, that to use the new generalised magnetic loading, or indeed the analytical alternative, as a drop-in replacement for its counterpart in the conventional induction machine may lead to higher iron losses and poorer power factor as the peak elds present are higher than in a conventional machine. Nevertheless to assume that the true magnetic loading is the sum of each contribution will inevitably lead to a conservative design.
denition, B =
2 2 B1 + B2 is close in value to the newly proposed generalisation, but has the advan-
4.8 Conclusion
In this chapter we have transformed the d-q equations of chapter 3 into a complex dynamic model using sequence components using a similarity transformation. It was shown that the complex form states, where N is the number of set of rotor coils. of the BDFM equations can be further reduced, without error, to a system containing 2 + N complex This complex form of the equivalent circuit was shown to admit an equivalent circuit representation. Various implementations of the equivalent circuit were derived for a BDFM with a single set of rotor coils. Particular care was taken to maintain the physical meaning of the inductance terms, especially the rotor leakage inductance. The intermediate variables Vr1 and Vr2 were used to derive expressions for the torque, and invesand |Vr2 | can take were considered via their relation to the magnetic ux density of each eld, where a generalisation of the notion of specic magnetic loading was proposed. The derived equivalent circuit was then used to explore operating modes and expected behaviour of BDFM machines. This led to the observation that the absence of signicant cascade induction mode torque typically implies poor BDFM synchronous torque. Furthermore it was shown in points 8 and 12 that the available torque envelope in BDFM synis to operate at a particular power factor, then a larger operating torque is achieved for a smaller |Z r |. chronous mode is inversely proportional to |Z r |, the rotor impedance, or similarly, that if the machine tigate the effect that stator 2 can exert on the power factor of stator 1. Bounds on the values that |Vr1 |
impedance due to L rh , the spatial harmonic inductance term. Therefore the component L rh is a crucial parameter for a BDFM. This makes physical sense as the rotor must cross-couple between two different pole number elds and therefore it is likely that there will be some harmonic penalty. We gure of merit for a BDFM rotor in terms of its ability to cross couple between the two pole number elds. This has a direct impact on the available torque envelope for a particular machine. propose, therefore, that the ratio of L rh to the total rotor self inductance (L rh + L rl + L r1 + L r2 ) is a
It was shown that in an example nested loop rotor design, a signicant part of |Z r | comprises of the
118
In the next chapter we will investigate the proposed gure of merit for a range of BDFM rotor designs.
Chapter 5
Possible Rotor Designs and Evaluation

5.1 Introduction
In chapter 4 an equivalent circuit model was derived for the class of BDFM machines considered, as described in chapter 2. This equivalent circuit model was used to investigate the characteristics and performance of the BDFM, and it was proposed in section 4.6 that |Z r |, the rotor impedance is a key parameter in determining the maximum torque envelope for the BDFM. The composition of Z r was discussed, and three potentially signicant components were identied: L rh , the excess harmonic series rotor inductance, L rl , the rotor leakage inductance, and Rr , the rotor resistance. Furthermore, it was proposed that the absence of cascade induction torque in a potential BDFM was an indicator that the machine will not generate synchronous torque due to the absence of cross-coupling. In this chapter these claims will be investigated by considering different possible BDFM rotor congurations. In total, seven different rotor designs will be considered. In order to verify the efcacy of measuring the cascade torque as an indicator of potential BDFM performance, 5 different rotors are considered: 3 potential BDFM rotors, and 2 rotors which do not cross-couple. Then, to illustrate the how Z r can vary with different rotor designs, 4 different rotor designs are considered and the different components of Z r are discussed. The chapter starts with a description of the rotor designs, and motivation for the choice of each design. All rotors considered are tested for the same stator conguration, a D180 frame size machine, originally a 22 kW (143 N m at 1463 rpm) Marelli Motori induction machine congured with 4 and 8 pole stator windings for operation as a BDFM (see appendix B). The 7 designs considered are as follows: 1. nested-loop design rotor: originally proposed and investigated by Broadway and Burbridge [17]. The design of the nested-loop rotor was considered in Wallace et al. [109], in which the effect of the span of loops within each nest was investigated. It was found that loops within the centre of a nest which subtended a small angle contributed little to the torque. However there does not appear to be any other papers in the literature which address the design of the rotor in
119
120
any detail. A prototype nested-loop rotor was available for experimental testing. 2. A new design of BDFM rotor requiring two layers comprising of single set of p 1 + p2 circuits, with the coil pitch chosen in such a way as to couple effectively with both pole number elds. is no evidence that either author ever designed or built such a rotor as they sought a single layer design. A rotor of this design was manufactured and experimental tests from it are presented. 3. A rotor design consisting of 18 isolated loops, each pitched over 7/36 of the circumference of the rotor, and each loop offset from the next by an equal amount (1/18 of the circumference). This rotor will be used to illustrate that linking both p1 and p2 (in this case 2 and 4) pole pair elds is not sufcient to ensure cross-coupling. A rotor of this design was manufactured and experimental tests from it are presented. 4. A rotor identical to rotor 3, but with every third coil removed. This rotor illustrates that by case of rotor 3) cross coupling is achieved. The prototype of rotor 3 was modied to provide an experimental rotor of this design. 5. A cage rotor with only p1 + p2 bars. This rotor design was the conceptual start for the nestedloop design of rotor, however it was dismissed as it was perceived to have a ...very high referred rotor slot-leakage [17]. This rotor design is considered by simulation. 6. A wound rotor consisting of two double layer windings (therefore 4 layers in total), one of p 1 pole pairs, the other of p2 pole pairs. The two windings phases are connected together and the star points are connected together. This rotor design is essentially the same as that used by Lydall [97] (although Lydall did not hard-wire a connection between the two windings), which was dismissed by Hunt as being wasteful of copper [48]. The rotor design is included because, it is likely to have low harmonic content. This rotor design is considered by simulation. 7. A standard squirrel cage induction machine rotor with 40 bars. This rotor is considered in order to illustrate the lack of cross-coupling present in a standard squirrel cage machine. The rotor is the standard rotor supplied with the induction machine which was converted into a BDFM. For each rotor design appendix B gives details of the mutual inductance matrices, including leakage inductances. Section C.1 gives the slot geometries used in each rotor and the formulae used to estimate leakage inductance in each case enforcing a p1 + p2 order of rotational symmetry on the design (as opposed to 18 order in the The design is developed from comments made in Broadway and Burbridge [17], although there
5.2 Rotor Designs
121
5.2 Rotor Designs

5.2.1 Rotor 1 - the nested-loop design rotor
The prototype nested-loop rotor is of type 2 design - see gure 2.4 - that is, the rotor comprises of nested loop terminated with a common end ring at one end only. Figure 5.1 shows the non-end ring end of the prototype rotor. As noted in [115] the difference between types 1, 2 and 3 is essentially one of ease of fabrication. N = p1 + p2 = 6 nests. Each nest is allocated 6 slots. Therefore three loops can be housed within The rotor consists of 36 slots, and as the prototype machine has p 1 = 2 and p2 = 4, the rotor has
each nest sub-tending 5/36, 3/36, and 1/36 of the circumference respectively. Figure B.3 displays this information diagrammatically. It can be seen from gure 5.1 that the bars are insulated from the laminations. It was chosen to insulated the bars as it was established that BDFM performance is signicantly degraded unless there is signicant bar-to-lamination impedance [117]. The insulation used was a high temperature paper, such that it would not be damaged by the brazing process needed to join the bars. Section B.3 gives full details of the design.
Figure 5.1: Prototype nested-loop design rotor, rotor 1
122
5.2.2
Rotor 2 - the new double layer design rotor
The design of rotors for the BDFM, when undertaken by Broadway and Burbridge, had the objective in mind of designing a rotor which was similar to a cage rotor - that is single layer and manufacturable by conventional casting methods [17]. This approach led to the nested-loop design of rotor. However whether or not the nested-loop rotor is the best design possible, in some sense, has never been established either mathematically or empirically. It is therefore worth investigating whether or not improved performance can be obtained with a different design of rotor. It is clear from simulations of nested-loop design rotors, and noted in [109] and [53], that loops within a nest, which only subtend a small fraction of the pitch between nests, will have relatively small currents owing in them, and therefore make little contribution to the output torque of the machine. The reason for the relatively small currents can be seen from considering the ux linked by a coil of pitch , recalling that the ux linked is proportional to the induced voltage as the magnetic ux density waveforms from the two pole number elds are moving at the same speed in the rotor reference frame at BDFM synchronous speed. Consider an n pole-pair winding. The peak ux linked by a single turn of pitch in sinusoidal magnetic ux density of amplitude B is: max sin(n /2) B cos(n)d = 2 B n /2
/2
In the BDFM there are two elds of p1 and p2 pole pairs respectively, having peak ux densities of B1 and B2 . Therefore the peak ux linked is: sin( p1 /2) + 2 B2 sin( p2 /2) max 2 B1 p1 p2 (5.1)
which is the peak linkage from each of the two ux density waves. The peak value may be obtained as the phase between the ux density waves can be varied. max = 0: Equation (5.1) may be maximised by such that max 1 [ B1 cos( p1 /2) + B2 cos( p2 /2)] 2 hence the maximum is achieved when B1 cos( p1 /2) = B2 cos( p2 /2). (5.2)
Thus the pitch which can give the maximum ux linkage is dependent on the relative magnitudes of the two pole number elds. However in the case that B1 = B2 this is achieved when: = 2 p1 + p2 (5.3)
In the prototype machine the windings were wound to give nominally equal ux densities. Therefore if the coil pitch is chosen such that (5.3) holds then, for p1 = 2, p2 = 4, the pitch is 60 . It should be noted that the preceding discussion it not a complete picture, as no explicit con-
sideration has been made of the requirement to link both pole number elds, so in practice it is an approximation.
5.2 Rotor Designs
123
The design problem then is how to maintain a suitable coil pitch to ensure good coupling to both pole number elds, whilst still achieving cross coupling. During the development of the nested-loop phases [17]. This fact was implicitly proved in chapter 3 by the particular choice of the class of rotors analysed. This design of rotor will be adopted, however it is necessary to choose both the pitch of the coils making up each phase, and the number of progressive coils in each phase. Figure 5.2 shows calculations, performed using the models techniques developed in chapters 2, 3 and 4, for the harmonic leakage and rotor 8 pole self-inductance. The values are expressed as percentages, of the total rotor self-inductance in the case of the harmonic leakage, and as a percentage of the 4 pole plus 8 pole harmonic inductance components for the 8 pole self-inductance. The values are calculated for 3 different coil pitches, and between 3 and 6 such coils connected in series each one rotor slot advanced on the previous. The rotor has 36 slots.
25
3.5 % harmonic leakage inductance 3 2.5 2 1.5 1 0.5 0 3 4 5 No. coils in phase 6
design rotor it was noted that cross coupling occurs when the rotor contains p 1 + p2 equally spaced
% 8pole selfinductance
20
70o coil pitch 60o coil pitch o 50 coil pitch
70o coil pitch o 60 coil pitch 50o coil pitch
15
10
4 5 No. coils in phase
(a) Rotor 8 pole self-inductance as a percentage of 8 pole + 4 pole terms
(b) Rotor harmonic leakage inductance as a percentage of total rotor self-inductance
Figure 5.2: Simulated rotor inductance parameter variations for different possible design for Rotor 2. The pitch of the coils making up each phase were varied (different colours on plots), and the number of progressive loops in each phase were varied (horizontal axis) Figure 5.2(b) shows that having 4 or 5 coils in each phase generally yields a minimum value for the harmonic leakage inductance percentage. However gure 5.2(a), shows that as the number of coils in each phase increases, the 8 pole self-inductance percentage decreases. This is undesirable as it implies that the effective turns ratio from the 8 pole stator to the rotor will be signicantly higher than that from the 4 pole stator to the rotor. Therefore the referred rotor impedance to the 8 pole side will be high, thus limiting the current that can be induced in the rotor from the 8 pole stator. This point was rst made in Broadway and Burbridge, where it is noted that as the angle between phases will
124
always exceed half a spatial period of the higher pole number eld, the effectiveness of the phase decreases [17, sect. 4.2]. Therefore 4 coils will be used per phase and it only remains to choose the coil pitch. There is little to choose between the options considered in terms of the harmonic leakage, however from gure 5.2(a) decreasing the coil pitch improves the proportion of 8 pole self-inductance. A 60 coil pitch was chosen as it gives a higher overall current than the 50 pitched coils, because the prototype machine stator produces equal amplitude 4 and 8 pole elds. However, if the stator were rewound, the 50 pitched coil would likely to be preferable as the 8 pole coupling is increased, albeit at the expense of the 4 pole coupling. Figure 5.3 shows a single phase of the new rotor design, with the chosen 60 coil pitch. Figure 5.4 shows a photograph of this rotor design after manufacture. Section B.4 has further details of the rotor including a full winding diagram.
single coil pitch
rotor core
etc
9 10 11 12 13 14 15 16 17 18
etc
Slot no.
rotor phase
Figure 5.3: Prototype new double-layer design rotor, rotor 2
5.2.3
Rotor 3 - isolated loop rotor
In the discussion of the design of rotor 2, section 5.2.2, the pitch of the coils was discussed. However it was noted that the pitch of the coils is not the only important factor to consider in the design of a rotor for a BDFM which is required to cross-couple. The isolated loop rotor design consists of 18 coils pitched over 70 , and each coil offset from the next by 20 . Thus the loop pitch is such that it will couple with both the 4 and 8 pole elds. However, if the current distribution due to each eld is considered then it can be seen, that the 4 pole eld induces a current distribution which is predominantly 4 pole, and is certainly without an 8 pole
5.2 Rotor Designs
125
Figure 5.4: Prototype new double-layer design rotor, rotor 2
component; and similarly the 8 pole eld induces a predominantly 8 pole current distribution without any 4 pole component. Therefore cross-coupling between the elds cannot take place. Section B.5 details the design of the rotor, and notes that the rotor is in fact a member of the class of rotors considered in chapter 3. Included in this section is a winding diagram for the machine, and the mutual inductance matrices. The lack of cross-coupling is evident from the rotor-rotor diagonalised machine mutual inductance matrix after the d-q transformation, equation (B.27). From equation (B.27) it is immediate that the dynamics associated with different stator supplies are entirely decoupled. This is consistent with a lack of cross-coupling, which by denition requires coupling between the two stator supplies. Figure 5.5 shows a photograph of the manufactured prototype rotor.
126
Figure 5.5: Prototype isolated loop design rotor, rotor 3
5.2.4
Rotor 4 - isolated loop rotor with coils removed
As described in the introduction rotor 4 is identical to rotor 3 but with every third loop removed, leaving a total of 12 isolated loops, in 6 groups of 2, where each group is spaced by 40 and each coil within each group spaced by 20 as before. 6. In doing so the rotor is made to cross-couple. This may be understood by considering a cage rotor elds [17, sect. 4.1], and although rotor 4 is not a cage rotor, a similar argument holds. The removal of one in every three loops is to enforce an order of rotational symmetry of p 1 + p2 =
with p1 + p2 bars. It is well known that such a rotor will cross-couple between p 1 and p2 pole pair Full details of the rotor design are given in section B.6. Included in this section are the mutual
inductance matrices. The effect of the removal of one in three loops from rotor 3 becomes apparent from the rotor-rotor diagonalised machine mutual inductance matrix after the d-q transformation. Equation (B.27) shows the original rotor with all 18 loops intact. Equation (B.28) shows the mutual inductance matrix for rotor 4. Equation (B.28) clearly shows cross-coupling between the stator supplies, albeit rather weakly in comparison to the direct coupling. The prototype rotor 4 was made by cutting one in every three bars of rotor 3 with a hacksaw. The
5.2 Rotor Designs
127
unmodied rotor is shown in gure 5.5.
5.2.5
Rotor 5 - 6 bar squirrel cage rotor design
Rotor 5 is a standard 6 bar squirrel cage rotor. It well known that such a rotor will function as a BDFM rotor, and was the conceptual starting point for Broadway and Burbridge when developing the nested-loop design of rotor [17]. The rotor uses 6 bars of cross-sectional area 4 times that of the bars used in the nested-loop design rotor. Unconventionally, the rotor is assumed to have open slots, of trapezoidal geometry. It is shown in section B.7 that for the purposes of mutual inductance calculation a cage rotor may be considered to be the same as a set of independent loops spanning the bar pitches. The resistance and leakage inductance matrices do alter and become non-diagonal, although they are still symmetric circulant. Therefore the inductance parameters for rotor 5 may be calculated using the method described in chapter 2 as the rotor is a member of the class of rotors considered.
5.2.6
Rotor 6 - wound rotor design
Rotor 6 consists of identical windings to those on the stator, but short-circuited together with the neutral points connected. As mentioned in the introduction, this rotor is included to illustrate its low harmonic content, but poor copper utilisation, as noted by Hunt [48].
PSfrag replacements c
c b 4 pole 8 pole b
Figure 5.6: Connection diagram for Rotor 6 The rotor does not form part of the class of rotors considered in this dissertation, however it is a simple case to analyse. Figure 5.6 shows the connection of the two 3 phase windings, therefore it is easy to verify that this connection may be represented by the combination matrix (B.35). This combination matrix may be used with the method described in section 2.5.1. It is easily shown that the rotor-rotor mutual inductance and resistance are given by the sum of the individual components and the stator-rotor mutual inductance is the difference between the mutual inductance matrices of the individual rotor windings. Hence it is immediate that rotor 6 will cross-couple both stator elds.
128
Section B.8 contains full details of the rotor-rotor and stator-rotor inductance terms including the leakage inductance calculation.
5.2.7
Rotor 7 - standard squirrel cage rotor
Rotor 7 is the standard cage rotor used in the induction machine frame used for the prototype BDFM. The rotor is made of cast aluminium, has 40 bars with Boucherot type slots (to give a higher starting resistance due to the skin effect). The rotor slots are closed as can be seen in gure 5.7. The closed nature of the slots complicates the estimation of the leakage inductance, as discussed in section C.1. This rotor will couple both stator elds, however there will be no cross-coupling for similar reasons to those discussed for rotor 3. Section B.9 has complete details of the mutual inductance matrices, and the d-q transformation used. The geometry of the rotor does not fall into the class of rotors analysed therefore special treatment is required. It can be shown that the d-q transformation used only removes unobservable/uncontrollable rotor states. Therefore the resulting mutual inductance parameters are a complete description of the machine. The d-q mutual inductance matrix, (B.45), clearly shows the lack of cross-coupling, as the dynamics due to the stator 1 supply are not coupled to the dynamics from the stator 2 supply.
Figure 5.7: Standard squirrel cage rotor 7
5.3 Experimental and Calculated Torque-Speed Curve Results
129
5.3 Experimental and Calculated Torque-Speed Curve Results

Of the 5 rotor designs that were manufactured, rotors 1, 2, 3, 4 and 7 were tested using the experimental rig described in appendix E. As discussed in section 4.6, it is to be expected that simple induction mode and cascade induction mode give a good indication of possible BDFM performance. High torques in simple induction mode suggest poor BDFM performance. The absence of cascade mode torques imply poor BDFM performance. Therefore by determining the simple induction mode and cascade induction mode the rotors may be qualitatively evaluated for BDFM performance, that is presence of signicant BDFM synchronous mode torque. The experiments were done at nominally 90Vrms (phase). This voltage was supplied from a 25A Variac variable transformer, and was therefore at nominally 50 Hz. The supply, therefore, has some impedance, although the current was generally low enough so that the supply dip was less than 1%. The operating point was set in each case by driving the DC load motor on the rig to the required speed, and then measuring the torque. The experimental data is overlaid with the same test done in simulation calculated using the equivalent circuit 4.3, with the parameter values calculated as described. In each case the resistance parameters were increased by 40% from their measured DC values (and calculated values in the case of the rotor) to allow for heating effects. Figures 5.8 and 5.9 show 4 and 8 pole simple induction mode (second stator open circuit) torquespeed curves respectively. Figures 5.10 and 5.11 show 4 and 8 pole cascade induction mode (second stator short-circuited) torque-speed curves.
5.4 Calculated Rotor Parameters

Using the tools developed in chapters 2, 3, and 4 the parameters for the machines which showed good BDFM performance were calculated. The parameter values correspond to gure 4.3. The intermediate states in the transformations are shown in appendix B. Table 5.1 shows the calculated values. Table 5.2 compares the parameter values of table 5.1 in a number of ways. As discussed in section 4.6 the parasitic rotor impedance terms, X rh , X rl , Rr /s1 directly affect the available synchronous torque envelope for a BDFM. We therefore dene the following dimensionless measures of BDFM performance: |X rh | , |X r1 + X r2 | |X rl | , |X r1 + X r2 | |Rr /s1 | |X r1 + X r2 | (5.4)
term of the rotor impedance. It is this term which couples, without leakage, to the stator, and the changes in value of this term simply reect different effective turns ratios between rotor and stator.
It is chosen to normalise the measures by |X r1 + X r2 | as this term represents the total non-parasitic
130
0 Torque (Nm)
10
15 0
Rotor 1 Rotor 2 Rotor 3 Rotor 4 Rotor 7 200 400 600 800 1000 1200 Rotor Speed (rpm) 1400 1600 1800
Figure 5.8: Torque-Speed plot in simple induction mode with nominally 90Vrms phase supply voltage to 4 pole winding, 8 pole winding open circuit, rotors as described in section 5.1. Experimental results with calculated results superimposed.
4 3 2 1 Torque (Nm) 0 1 2 3 4 5 6 0 Rotor 1 Rotor 2 Rotor 3 Rotor 4 Rotor 7 200 400 600 800 1000 1200 Rotor Speed (rpm) 1400 1600 1800
Figure 5.9: Torque-Speed plot in simple induction mode with nominally 90Vrms phase supply voltage to 8 pole winding, 4 pole winding open circuit, rotors as described in section 5.1. Experimental results with calculated results superimposed.
5.4 Calculated Rotor Parameters
131
10 5 0 Torque (Nm) 5 10 15 20 0 200 400 600 Rotor 1 Rotor 2 Rotor 3 Rotor 4 Rotor 7 800 1000 1200 Rotor Speed (rpm) 1400 1600 1800
Figure 5.10: Torque-Speed plot in cascade induction mode with nominally 90Vrms phase supply voltage to 4 pole winding, 8 pole winding in short circuit, rotors as described in section 5.1. Experimental results with calculated results superimposed.
8 6 4 2 Torque (Nm) 0 2 4 6 8 10 12 0 200 400 600 800 1000 1200 Rotor Speed (rpm) 1400 1600 1800
Rotor 1 Rotor 2 Rotor 3 Rotor 4 Rotor 7
Figure 5.11: Torque-Speed plot in cascade induction mode with nominally 90Vrms phase supply voltage to 8 pole winding, 4 pole winding in short circuit, rotors as described in section 5.1. Experimental results with calculated results superimposed.
132 R1 ( ) L 1l (mH) L1 f (mH) |Mcs1r | (mH) L r1 (H) L rh (H) L rl (H) Rr (m )
Possible Rotor Designs and Evaluation L r2 (H) |Mcs2r | (mH) L2 f (mH) L 2l (mH) R2 ( )
Rotor 1: nested-loop design 2.7 3.54 307 2.7 24.1 0.98 1.06 0.074 13 2 314 8 4.4
Rotor 2: new double-layer rotor design 2.7 3.63 313 9.6 293 5.3 8.82 1.9 44 3.7 321 8.4 4.4
Rotor 5: 6 bar squirrel cage rotor design 2.7 3.08 339 2.8 23 3.5 0.888 0.028 5.7 1.4 347 6.2 4.4
Rotor 6: Wound rotor design 2.7 3.32 287 2.9 28.6 0.46 1.02 35 29 2.9 294 7.1 4.4
Table 5.1: Calculated parameter values for different BDFM rotor using coupled circuit approach described in chapters 2, 3, and 4, corresponding to gure 4.4 The nal column of table 5.2 shows the total cross-section conductor area in the machine stack (therefore end windings are not included). For example, Rotor 1 has 133mm2 bar area and 36 bars in total hence the total cross-sectional area of copper is 4774mm 2
5.5 Discussion of Results

5.5.1 Experimental Torque-Speed Curves
ment. In gure 5.8 rotors 1-4 show good agreement. In gure 5.9 rotors 1-4 show reasonable agreement. The errors, although proportionally large, are less than 1 N m in all except rotor 3. This is well within the measurement error of the torque transducer, which drifts by around 0.5 N m over the course of an experiment, and the frictional forces on the BDFM, which are up to 0.8 N m. Figures 5.10 and 5.11 also show good agreement for rotors 1-4. The exceptions being rotor 2 at low speeds in 4 pole cascade mode, and 8 pole cascade mode at higher speeds; and rotor 3 at higher speeds in 8 pole cascade. The most probable cause for these errors are inaccurate resistance values. Rotor 7 gave consistently less good agreement, particularly in 8 pole mode. The poor agreement at large slips (positive and negative) is due to the Boucherot slots, which are deep slots which rely on the skin effect to increase the effective resistance at large slips where the frequencies of the currents in the rotor are higher. It is therefore expected that the simulated results give
The simulated torque-speed curves give generally good agreement with those found by experi-
5.5 Discussion of Results
133
|X r1 + X r2 | (m )
|X r1 | (%)
|X r2 | (%)
|X rl | (%)
|X rh | (%)
|X rl + X rh | (%)
|Rr /s1 | (%)
|Z r p | (%)
ACu (mm2 )
Rotor 1: nested-loop design 11.6 65.5 34.5 2.9 2.7 5.5 1.9 5.9 4774
Rotor 2: new double-layer rotor design 106 86.9 13.1 2.6 1.6 4.2 5.4 6.8 1235
Rotor 5: 6 bar squirrel cage rotor design 9.03 80.1 19.9 3.1 12 15 0.92 15 3182
Rotor 6: Wound rotor design 18.1 49.6 50.4 1.8 0.8 2.6 585 585 6514
Table 5.2: A comparison of calculated parameter values of table 5.1 expressed as a percentage of |X r1 + X r2 |, the rotor fundamental reactances at 50 Hz. The slip, s 1 , is assumed to be 0.33, which (Rr /s1 )2 +(X rl +X rh )2 corresponds to 1000rpm when stator 1 (4 pole) is grid-connected. |Z r p | = |X r +X r |
1 2
lower (closer to zero) torques at large slips. However the 4 pole results at small slips are still fairly inconsistent. This is due to the crude estimate of rotor leakage inductance used. The simulated results could be made to coincide much more closely by adjusting the value of leakage inductance used. Simple induction mode: All the rotors exhibited some simple induction mode torque, in both 4 and 8 pole modes. For 4 pole simple induction torque: Rotors 7 and 3 gave the largest peak values, rotor 4 a little less, rotor 2 even less and rotor 1 the least. This is to be expected, as rotors 7 and 3 fail to cross-couple but both couple 4 pole elds strongly. Rotor 4 gives somewhat less torque than rotor 3 which is consistent with it producing weak cross-coupling due to the removal of one set of windings. Rotors 1 and 2 exhibit strong cross coupling hence the smaller torques. For 8 pole simple induction torque: Rotors 7 and 3 gave the largest peak values, rotor 4 a little less, rotor 1 even less and rotor 2 the least. For similar reasons to the 4 pole test the results for rotors 7, 3 and 4 are as expected. The very small torque produced by rotor 2 is a reection of its relatively poor 8 pole coupling combined with its relatively strong cross-coupling. In contrast, the performance of rotor 1 is similar in both 4 pole and 8 pole simple induction mode.
134
The differing magnitudes of the 4 and 8 pole simple mode results for rotor 7 reect the differing stator impedances for the two supply windings. Cascade induction mode: Rotors 3 and 7 showed no noticeable cascade torque. The torque-speed curves are essentially the same as those obtained in simple induction mode. This conrms the absence of cross coupling in these rotor designs. Rotors 1 and 2 showed strong cascade torque and rotor 4 showed weak cascade torque. Rotors 1 and 2 both exhibit signicant cross-coupling, therefore it is expected that the cascade mode torque is relatively large. Rotor 4 only produced weak cascade action because its cross-coupling is also weak. As discussed in section 4.6 the cascade torque-speed curve consists of an induction machinelike torque-speed characteristic at 500rpm (cascade synchronous speed for p 1 = 2, p2 = 4 at 50 Hz) superimposed on top of the standard machine torque-speed curve. The amplitude of the cascade mode torque-speed characteristic is similar in magnitude to the standard torque-speed characteristic at the machine synchronous speed. This is to be expected as cascade mode in effect lowers the impedance of the rotor as the second stator resistance and leakage inductance appear in parallel with the second stator magnetizing inductance, see gure 4.9.
5.5.2
Calculated machine parameters from tables 5.1 and 5.2
In consideration of the rotor resistance term, it is desirable to minimise this term as far as possible as it directly impacts the efciency of the machine. It is immediately apparent that rotor 6 suffers from its large relative rotor resistance, which is particularly bad as this rotor has the largest amount of copper of all the rotor designs. This conrms the point made by Hunt that having a rotor design comprising of two three phase windings is wasteful of copper [48]. Rotor 2 has a relatively large resistance, however it suffers from having a relatively low amount of copper in the design, as compared to 1. This was simply a consequence of manufacturing a prototype rotor with a standard slot shape, and could therefore easily be overcome. If rotor 2 had the same cross-sectional area of copper as that of rotor 1, then the resistance would drop to 1.4%. Rotor 5 has the least resistance percentage, and rotor 1 somewhat higher, but still quite acceptable. The values of percentage leakage reactance on the four rotors are similar. The slightly lower value for Rotor 6 is not signicant.
5.6 Conclusion
135
In contrast to the leakage reactance, the values of excessive rotor harmonic reactance differs greatly between the 4 rotor designs. Rotor 6 has the least percentage value of excess harmonic reactance of the four rotors. This is expected as this rotor has the lowest realistic value for any machine of this frame size, as the rotor is comprised to two double layer windings which therefore have a low harmonic content. Rotor 5 has 12% excess harmonic reactance which is over 4 times the next highest value of 2.7% in rotor 1. This relatively high value of excess harmonic reactance conrms the prediction in Broadway and Burbridge that a 6 bar cage rotor suffers from excessive harmonic inductance [17, sect. 4.1]. Rotor 2 achieves a signicantly lower value than rotor 1, demonstrating that the better choice of coil pitch has achieved the desired effect of reducing the harmonic content. It is desirable to minimise the value of the excess leakage reactance percentage, not only to increase the available torque but also because it represents some of the generated harmonic magnetic ux densities in the air gap.
5.6 Conclusion
In this chapter seven rotor designs have been considered including a new potential BDFM rotor design. Five of these rotor designs were manufactured and torque-speed curves were measured for 4 different operating modes. From these tests it is clear which rotors will cross-couple and, therefore, are potential BDFM rotors, as predicted from analysis of the equivalent circuit in section 4.6. The use of simple and cascade mode tests to qualitatively test potential BDFM rotors is therefore conrmed. The experiments were complemented by simulation results obtained using the equivalent circuit model and method of parameter calculation previously derived. The agreement between measured and predicted results was within expected experimental errors and the limitations of the model. Therefore it is possible to have condence in the modelling and parameter calculation methods described in chapter 2, 3, and 4. The four rotors which show strong cross-coupling, rotors 1, 2, 5 and 6 were analysed by calculation of their equivalent circuit parameters. Rotors 1 and 2 are both plausible BDFM rotors if the cross-sectional area of copper in each slot is increased in rotor 2. Rotor 2 should lead to a rotor with a larger torque envelope, and greater efciency than rotor 1 which has twice the percentage excess harmonic reactance. Rotor 5 might possibly be considered as a BDFM rotor as it will have low copper losses, however its excessive harmonic reactance is very high, as expected, and this will not only limit the available torque but also introduce unwanted magnetic ux density harmonics. Rotor 6 will not make a good BDFM rotor due to its excessive resistance. The rotors were evaluated by three measures. Of these measures, the ones providing the most differentiation between rotors were in each rotor. In terms of their effect on |Z r p | the excess harmonic reactance percentage is the more
|X rh | |X r1 +X r2 |
and
|Rr /s1 | , |X r1 +X r2 |
taking into consideration the area of copper
136
Therefore both the excess harmonic reactance and leakage reactance are crucial parameters in the
signicant, as in a typical good rotor design (e.g. rotor 1) |Z r p | is dominated by the reactive parts. design of BDFM rotors. The slot shape is the main controlling factor for the leakage reactance, so this should be designed to ensure minimum leakage. The design of the rotor geometry (which
determines the rotor harmonic content) may, therefore, be evaluated using the percentage excess harmonic reactance. However the reduction of the excess harmonic reactance must be weighed up against any penalty in terms of rotor resistance. Rotor 6 illustrates the lowest realistic percentage harmonic leakage reactance, however the penalty of increased rotor resistance is too great to make the design viable. Rotor 2 achieves a signicant reduction in percentage harmonic leakage reactance (over rotor 1), with no such resistance penalty. Therefore, as rotor designs other than the nested-loop design achieve an improved BDFM performance, the issue of rotor design is worth further investigation. The measures proposed may be used by the designer to evaluate proposed rotor designs. There are, however, limitations to the proposed methods of rotor evaluation. The most signicant are restrictions imposed by the design of the iron circuit in the machine. Designs aimed at the reduction of leakage reactance may not be feasible as they will lead to either excess iron loss, poor power factor or both. Furthermore the skin effect has been entirely neglected in this analysis. The skin effect will increase the rotor resistance. This will not only have a direct effect on the efciency of the machine, but will also impact the value of |Z r p |.
Chapter 6
BDFM Parameter Identication

6.1 Introduction
In chapter 2 a method of calculating machine parameters was presented. In chapters 2, 3, and 4 different models were developed for the BDFM, and the transformations between these models derived, thus allowing parameter values for these models to be computed using the method described in chapter 2. These models included different equivalent circuit representations for the BDFM. Chapter 5 showed experimental torque-speed curves for ve rotor designs using calculated parameter values. Although the agreement was reasonable, and well within the limitations of the instrumentation and the parameter calculation methods, there is a need for a method of experimentally determining parameters for the machine which leads to models giving closer agreement with experimental data. Only two previous works are known which consider the problem of parameter estimation for the BDFM, [80, 79]. These works use a frequency domain method to nd d-q parameters for a prototype nested-loop design machine. The methods are presented in some detail in simulation, and some (but not all) parameters are estimated from a prototype machine in [79]. Although the method proposed in [80, 79] appears to generalise to a wider class of BDFM machines, a signicant amount of equipment is needed to implement such a scheme: 10 kHz multichannel isolated A/D, closed-loop external speed control for the test rig, variable frequency current source supply for the BDFM [79]. A simple method of parameter estimation is therefore proposed. The method is based on curve tting torque-speed curves from both simple and cascade induction mode tests. This is a natural generalisation of the standard no-load and locked rotor tests commonly used to estimate parameters in standard induction machines. The standard no-load test can be used to determine the stator resistance and magnetizing reactance, and if the series stator resistance is measured with DC, then the core loss resistance term can be estimated. The standard locked-rotor can estimate the rotor impedance; it works by assuming that the magnetizing reactance is very much larger than the referred series rotor impedance, which appears
137
138
in parallel with the magnetizing reactance (and the core loss resistance). As discussed in section 4.6, the BDFM can operate as a normal induction machine, in simple induction mode, and the no-load test can be applied to get information for both stator supplies. However, as discussed in section 4.6, and conrmed in section 5.5, good BDFM designs will generally have very high referred rotor impedance in simple induction mode, therefore the locked rotor test will not be directly applicable. Although it would be possible to perform the locked rotor test, taking into account the magnetizing reactance, the actual rotor leakage reactances would be prone to large errors as they appear in series with the second stator magnetizing reactance. Even with the normal induction machine, the standard locked-rotor and no load tests are known to give parameters which lead to models which have signicant errors away from the operating points at which the parameters were determined [111]. This problem motivated a number of papers where output error minimization methods were successfully used [29, 66, 78]. As the BDFM is similar to the standard induction machine, approaches of this nature should lead to acceptable results. It should be noted that all the methods considered are off-line methods, which require a special experiment to be performed on the machine in order to determine the parameter values. This is somewhat distinct from on-line methods where no explicit experiment is performed, and parameters are estimated during normal machine operation. There have been efforts to address this (rather more difcult) problem for the induction machine, a recent example is [23]. We begin by describing a new optimization method.
6.2 Parameter Extraction Optimization Method

6.2.1 General Optimization Problem
The optimisation problem to be solved is a non-linear least squares problem, and can be stated as: Given Yi Rk , P Rn , U Rkm , Fi : P, U Rk , i l
l
Find
P:
PW B+P0
min
i=1
Si Yi Fi (P, U )
(6.1)
where Fi is a function which estimates the i th measured output for a particular parameter vector, P. U is a matrix of input data points where each row is an individual data point. Y i is a vector of the measured i th output data due to U . There are k input/output data pairs and m different measured inputs (i.e. U has k rows and m columns). Si is a scalar weight applied to the i th output error. So there are n parameters, m inputs, k data points for each input, and l outputs. Furthermore, Fi is dened as: f i (P, u 1 ) . , . Fi (P, U ) = . f i (P, u k ) u1 . U =. . uk
(6.2)
139
where f i : P, u j R1 .
is a hypercube), centred on P0 with corners at P0 diag(W ). Clearly the value of P that gives the minimum cost function value will, in general, not be unique. As stated this is a non-convex optimization problem, and therefore difcult to solve [123]. In order to solve this optimisation problem we use a simple (or crude) random search (also known as a Monte Carlo method). For practical and theoretical details of the algorithm see, for example, [123]. The principal advantage of a simple random search, for our purposes, is its strong resilience to measurement noise and modelling error while guaranteeing convergence to the global minimum. Furthermore, because it requires no gradient information, there is no need for any a priori information about the particular cost function. It can be shown that without any such information, a simple random search is as efcient as any other method, in reaching the global minimum [44]. The simplicity of the algorithm means that implementation is straightforward and, because the method does not depend on the cost function chosen, the algorithm can be applied to different cost functions without modication. Fi (q, U ) The search is performed across W B + P0 to nd Q = {q1 , , q N } where Q :
2 m i=1
W B + P0 represents the set of possible parameter values chosen from a rectangular set (as B
guesses for candidate qk s using the Mersenne Twister [71], a fast uniform random number generator with a long repetition period. Having found Q we then nd the, q k corresponding to the minimum cost, K min . This qk , called qopt henceforth is an estimate of the global minimizer and K min an estimation of the global minimum. Furthermore the standard deviation of Q gives an indication of how much condence one should have in taking q opt as the global minimum. Although the efciency could be increased by using a multi-start, (see [123] for details) or probabilistic branch and bound algorithm (such as [106]), the additional complexity of the algorithm, and requirements to have some appreciation of the modelling error and noise, make these algorithms less appealing than they might rst appear. Furthermore there is a tradeoff between increased complexity in the algorithm versus the decreased number of iterations required.
, where > 0 is an acceptable value for the cost function to take. We generate
Si Yi
6.2.2
Application to BDFM
Notice that thus far it has not been necessary to specify what the measured outputs Y are, or either what the estimation functions F are. We desire to choose the inputs and output such that they are easy to measure terminal quantities. Therefore we are restricted to the stator voltages, currents, the torque and the rotor speed. Furthermore due to the substantial difculty of trying to measure the phase difference between stator 1 and stator 2 quantities we restrict ourselves to magnitude measurements of stator 2 quantities. Having decided what quantities to measure, it is necessary to consider what parameters are sought, and determine whether or not nding the parameters uniquely is feasible. The problem of whether or nor the parameters may be determined uniquely from any set of experiments is known as identia-
140
bility. Recognising that the equivalent circuit can predict the behaviour of the machine in both simple induction and cascade modes, we propose then an approach based on the extraction of parameters from measured torque-speed characteristics of the kind shown in gures 5.8-5.11. This approach has the advantage that a single frequency supply may be used to perform the tests. This is in contrast to the only previously published works on BDFM parameter extraction, [80, 79], where a method Nevertheless, not all the inductance terms in the equivalent circuit shown in gure 4.9 can be unambiguously determined from external measurements. Figure 6.1 shows equivalent forms of the
XT and networks, details can be derived using, for example, [20]. m X1 Xm Xm X2 Xm X1 X1 X2 1 X1 2 Xm X2
2 Xm X1 2 Xm X1
PSfrag replacements Xm
requiring operation at a range of different supply frequencies is proposed.
1: Xm X1 1: Xm X1
PSfrag replacements
Figure 6.1: Equivalent forms of T and networks

Rrn /s1 R2n s2 /s1 I2 n
I1
R1
j1 L rn
Irn
V1 , 1
j1 L m 1n
j1 L m 2n
s2 V2 s1 n
I2
V2
Nb : Na
N1 : 1
1 : N2
Figure 6.2: Simplied Referred Per-Phase Equivalent Circuit Using gure 6.1, gure 4.9 may be transformed into an alternative form of the equivalent circuit as shown in g. 6.2 for which all component values can be determined from the estimation procedure, as the number parameter has reduced. A similar modication was proposed by Slemon in the case
141
of parameter determination for the equivalent circuit for the standard induction motor [98]. For the circuit to be electrically equivalent to that of g. 4.9 the parameters, and turns ratios, will assume slightly different values: L m2 L m1 Nb = L 1 + L m1 L 2 + L m2 R Rrn = r2 L m 1n = L m 1 + L 1 Na N2 Nb R 2n = R 2 b L m 2n = L m 2 2 2 Na Na Lm L + Lm L rn = r 2 1 L m 1n + 22 L m 2n Na Na Na =
(6.3)
However, in a BDFM the referred series rotor inductance is likely to be larger than the series stator inductance so differences in inductance values between the two circuits will be small. Direct measurements of rotor bar currents, for example using techniques described in [1, 83], would enable all parameters in g. 4.9 to be determined. In simple induction mode operation there are 2 inputs, 3 measured outputs, 4 parameters, and N s data points. The subscript simple/1 is used to indicate simple induction mode referred to stator 1. Y is chosen as follows: T1 . . , =. Tk {I11 } . = . . {I1k } r 1 V11 . . . = . . . r k V1k
Ysimple/11
Ysimple/12
(6.4)
Ysimple/13
{I11 } . = . , . {I1k } R1
Usimple/1
(6.5)
Psimple/1
Fsimple/1i can be derived from the referred equivalent circuit, gure 6.2, taking I2 = 0: T = f simple/11 (Psimple/1 , u simple/1 ) {I1 } = f simple/12 (Psimple/1 , u simple/1 ) {I1 } = f simple/13 (Psimple/1 , u simple/1 ) (6.7) (6.8) (6.9)
R r = Lr + L m 2 L m1
(6.6)
142
If Usimple/11 is chosen so that the BDFM is exercised over a range of speeds such that the slip (4.32) varies substantially, say from 1 to -0.5, it can be shown from the equivalent circuit that each f simple/1i depends strongly on each element of Psimple/1 , which suggests that there is a unique Psimple/1 which solves the optimization problem. However due to the presence of measurement noise and modelling error this will not necessarily be the case. Solving this optimization problem as outlined in section 6.2.1 gives the parameters indicated in (6.6).
N1 with V2 = V2 N2 (from gure 4.9):
However we have not used |V2 | as one of the measurable outputs. It can be seen from gure 6.2,
|V2 | and the ratio of the turns ratios, of

N2 N1 N1 N2
N1 = |Ir 1 |L m 2 N2
(6.10)
is unknown. Therefore all that can be achieved is the determination
L m 2 . It is proposed therefore that, having determined the optimal parameter set, linear leastN2 N1
squares is used to determine
L m2 : |V21 | . = (A T A)1 A T . , . |V2k | |Ir 1 11 | . A= . . |Ir k 1k |
N2 Lm N1 2
and |Ir j | are calculated from gure 6.2 using the optimal parameter set.
In cascade induction mode operation there are 2 inputs, 4 measured outputs, 6 parameters, and
Nc data points. The subscript cascade/1 is used to indicate simple induction mode referred to stator 1. Y is chosen as follows: T1 . . , =. Tk {I11 } . = . , . {I1k } {I11 } . = . . {I1k } |I21 | . = . . |I2k | R r L = r L m1 R 2
N1 N2
Ycascade/11
Ycascade/12
(6.11)
Ycascade/13
Ycascade/14
(6.12)
Ucascade/1
r 1 V11 . . . , = . . . r k V1k
R1
(6.13)
Pcascade/1
6.3 Parameter Estimation Results
143
Fsimple/1i can be derived from the referred equivalent circuit, gure 6.2, taking V2 = 0: T = f simple/11 (Psimple/1 , u simple/1 ) {I1 } = f simple/12 (Psimple/1 , u simple/1 ) {I1 } = f simple/13 (Psimple/1 , u simple/1 ) |I2 | = f simple/14 (Psimple/1 , u simple/1 ) (6.14) (6.15) (6.16) (6.17)
As for the simple mode Ucascade/11 is chosen so that the BDFM is exercised over a range of speeds such that the slip (4.32) varies substantially. It can be shown from the equivalent circuit that each f cascade/1i depends strongly on each element of Pcascade/1 . However f cascade/1i also depends on L 2 , and it is not possible to determine this from the measured outputs. Therefore the value extracted from the simple/2 is used, and is explicitly included in f cascade/1i . In these circumstances there is a unique Pcascade/1 which solves the optimization problem, with the same caveat regarding noise and measurement error as in the simple mode procedure. Again the optimization problem is solved as outlined in section 6.2.1 giving the parameters indicated in (6.13). Although the original problem contained Si , a weight, it is set to unity without exception for the applications described. The choice of weight can improve the algorithm by compensating for different scaling of the measured outputs, for example if |V2 | had been included as a measured output then it would have been appropriate to scale down the effect of the error on this measurement as it is typically 2 orders of magnitude larger than the other measured outputs.

The method was applied to data similar to that shown in gures 5.8-5.11. Each data set had between 10 and 20 data points. When solving the optimization problem, in order to minimise the execution time of each random test point the mathematical functions governing the behaviour of the machine were simplied as far as possible using the symbolic maths package Maple, and then hard coded in a C function along with the random number generation code. This function was then called from Matlab. With this implementation it was possible to achieve 4 10 5 tests per second on a 2.8GHz Pentium 4 in cascade mode, and slightly more in simple mode. As an indication of the time taken to solve the optimization to satisfactory accuracy, it took approximately 9 108 tests, taking around 40 minutes, to generate 10 candidate solutions of suitable from cascade induction mode. The tables show the results of the optimization, along with the standard deviation taken over the best 20 results. The standard deviations are generally considerably smaller than the estimated parameter values, which indicates that the noise level in the measurements is sufciently low to have condence in the results, and that the parameters can be uniquely determined. As accuracy for rotor 1. Tables 6.1, 6.2 show data from simple induction mode and tables 6.3 and 6.4
144
R1 ( )
Rrn ( )
L rn + L m 2n (mH)
L m 1n (mH)
L m 2n N2 N1 (mH)
Rotor 1: nested-loop rotor design Opt: Std. Dev.: 4.02 0.0243 1.2 0.00769 120 0.204 240 0.0507 137 0.00879
Rotor 2: new double-layer rotor design Opt: Std. Dev.: 3.52 0.0384 3.1 0.0306 65.5 0.246 260 4.44 105 0.547
Rotor 3: isolated loop rotor design Opt: Std. Dev.: 3.28 0.0548 1.36 0.00627 26.1 0.199 252 5.19 5.46 0.00855
Rotor 4: isolated loop rotor design with removed loops Opt: Std. Dev.: 3.75 0.112 2.12 0.0553 48.2 0.509 240 11.9 41.1 0.276
Rotor 7: standard cage rotor design Opt: Std. Dev.: 3.54 0.049 0.486 0.00705 22.8 0.202 253 17.7 0 0
Table 6.1: Parameter Extraction in simple induction mode with 4 pole supply an additional measure of effectiveness, the algorithm returns the value of the minimised error which, if suitably normalised, gives a measure of t quality. This value could be normalised against the experimental data if required.
6.3.1
Comparison of data obtained using estimated parameter values with experimental data
In order to verify the machine parameter values estimated by the technique described, the parameter values tabulated in tables 6.1 to 6.4 were used in the BDFM equivalent circuit model given in gure 6.2 to generate torque-speed data. The generated torque-speed data was produced at a constant 90Vrms (phase) stator 1 supply, with stator 2 short and open circuit, to give simple and cascade induction modes. The data was calculated for the 4 pole winding being stator 1 and then for the 8 pole winding as stator 1. Figure 6.4 shows the same experimental data used in gures 5.8 to 5.11 overlayed with the calcu-
145
R1 ( )
Rrn ( )
L rn + L m 2n (mH)
L m 1n (mH)
L m 2n N2 N1 (mH)
Rotor 1: nested-loop rotor design Opt: Std. Dev.: 5.58 0.089 2.52 0.0273 507 1.92 267 0.692 300 11.4
Rotor 2: new double-layer rotor design Opt: Std. Dev.: 6.21 0.0502 30.8 0.314 1460 7.42 246 0.27 491 2.53
Rotor 3: isolated loop rotor design Opt: Std. Dev.: 5.46 0.0419 14 0.0624 203 0.464 272 0.625 6.6 0.082
Rotor 4: isolated loop rotor design with removed loops Opt: Std. Dev.: 5.99 0.0847 32.4 0.411 501 3.46 268 0.81 91.4 0.313
Rotor 7: standard cage rotor design Opt: Std. Dev.: 5.08 0.0895 2.42 0.0306 78.9 0.374 266 5.06 0 0
Table 6.2: Parameter Extraction in simple induction mode with 8 pole supply
146
R1 ( )
Rrn ( )
L rn (mH)
L m 1n (mH)
R 2n ( )
L m 2n (mH)
N1 N2
Rotor 1: nested-loop rotor design Opt: Std. Dev.: 3.63 0.106 1.26 0.038 35.1 0.379 277 27.1 2.46 0.0628 101 3.83 0.685 0.0131
Rotor 2: new double-layer rotor design Opt: Std. Dev.: 3.6 0.125 2.92 0.0673 27.4 0.386 299 32.8 0.716 0.0197 40.8 1.55 0.408 0.00788
Rotor 4: isolated loop rotor design with loops removed Opt: Std. Dev.: 3.61 0.176 2 0.0699 39.8 0.986 266 26.1 0.188 0.0223 7.6 0.681 0.169 0.00718
Table 6.3: Parameter Extraction in cascade induction mode with 4 pole supply
R1 ( )
Rrn ( )
L rn (mH)
L m 1n (mH)
R 2n ( )
L m 2n (mH)
N1 N2
Rotor 1: nested-loop rotor design Opt: Std. Dev.: 5.79 0.213 2.82 0.0951 72.9 0.919 288 19.5 7.25 0.28 391 25.5 1.41 0.0447
Rotor 2: new double-layer rotor design Opt: Std. Dev.: 5.39 0.391 23.5 0.377 189 1.27 281 5.52 22.9 0.487 1640 39.4 2.51 0.0301
Rotor 4: isolated loop rotor design with loops removed Opt: Std. Dev.: 7.61 0.297 23.1 1.02 337 5.77 264 3.15 16.7 0.815 607 30.1 1.59 0.0385
Table 6.4: Parameter Extraction in cascade induction mode with 8 pole supply
147
lated torque-speed data. The gures show a generally good agreement between predicted and measured data, and show that the use of parameter values determined by the estimation technique, yield more accurate results than those derived by calculations in some cases. This is, of course, to be expected. To illustrate the applicability of the parameter extraction method beyond the prediction of solely simple and cascade induction mode torques, the parameters estimated from 4 pole cascade mode, given in table 6.3, were used to predict the maximum and minimum torque envelope for the prototype machine with rotor 1. The parameters were used in the equivalent circuit model, and the model solved to nd the peak torque values for given supply voltages, over a range of shaft speeds. The stator 2 phase voltage magnitude was set as max{220
f2 f1
, 20.7}, that is linearly proportional to the
supply frequency with a boost at lower frequencies. The stator 1 supply phase voltage was xed at 220Vrms , and 50 Hz. It was chosen to measure the maximum and minimum values of the torque, thus each data point was well dened for given supply voltages and frequencies. Each data point was measured by increasing the load torque (or generating torque) until pullout occurred and recording the nal torque value immediately prior to pullout. For some operating points the machine dynamics were very lightly damped, or unstable, therefore it was necessary to use the feedback control scheme proposed in section 7.7 to restore stability. The process was repeated for a range of shaft speeds. The experimental apparatus used is described in appendix E. Figure 6.3 shows the experimental results (blue crosses), with the simulation results overlaid in red. The agreement between the experimental and simulated results is very good. The experimental torque values are less than those predicted from calculation, however this is to be expected as the peak torque value is practically unattainable (although one should be able to get arbitrarily close). The structure in the plot at around natural speed (500rpm) could have been eliminated by making the voltage boost suitably asymmetric about natural speed. A comparison of calculated and estimated parameter values will now be made.
6.3.2
Comparison of estimated to calculated parameter values
Table 6.5 shows calculated parameter values for the prototype machine with rotors 1, 2, 3 and 7 in the form of the equivalent circuit shown in gure 6.2. These parameter values were calculated using the conversion formulae, (6.3), from the parameters presented in table 5.1 and, in addition, parameter values are given for rotors 3 and 7, which were calculated in the same way. However, as discussed in chapter 5, rotors 3 and 7 do not cross-couple, therefore parameter values are only given for simple induction mode. Rotors 1, 2 and 4 cascade induction mode parameters will be compared, as the cascade mode parameters are a superset of the simple mode parameters. Simple mode parameters will be compared for rotors 3 and 7. We rst consider the self-consistency of the estimated parameters. If the machine considered is
148
200
100
0 Torque (Nm)
100
200
300
400 0
Experimental Modelled const V/f w/boost 100 200 300 400 500 600 Rotor Speed (rpm) 700 800 900 1000
Figure 6.3: Experimental verication of the parameter extraction procedure: maximum synchronous torque envelope with varying rotor speed for rotor 1 (nested-loop design) with the 4 pole winding supplied with nominally 220V (phase), the 8 pole is supplied with a const. V / f law with boost, i.e.: are those of table 6.3. Where the red line is dashed, it was not possible to experimentally determine the pull-out torque due to insufcient dissipative power in the experimental test rig. V8 = max{220
f2 f1
, 20.7} (solid lines, with experimental points marked). The parameters values used
precisely modelled by the equivalent circuit form given in 6.2, then parameter values estimated for the machine supplied from stator 1 or stator 2 should give identical results. Furthermore, these results are related by the square of the turns ratio, N 1 /N 2. The results of parameter estimation when the 4 pole winding was supplied should be equal to the results when the 8 pole winding was supplied divided by the square of the turns ratio. For rotor 1 the errors ranged from 0.5% to 45% across the different parameters, for rotor 2 the errors ranged from 0.6% to 20% and for rotor 4 the errors went up to 1500%. These results are expected as rotor 2 has only one set of 6 rotor coils and should therefore be precisely modelled. Rotor 1 has 3 sets of 6 rotor coils, the inner, middle and outer loops of each nest and although it was shown in chapter 3 that using the proposed model order reduction technique leads to relative small modelling errors, errors are nonetheless present, and these are reected in the inconsistency within the parameter values. Rotor 4, has 2 sets of 6 rotor coils, however the coils are not arranged in a manner which leads to good BDFM behaviour, as can be seen from the cascade mode torque-speed curves. It is therefore unsurprising that the machine with this rotor is not well modelled by the circuit of gure 6.2.
149
We now compare the cascade induction mode estimated parameters for rotors 1 and 2 with those found by calculation. The magnetizing inductance, L m 1n , is between 5% and 10% higher in the calculated parameter values, this is most likely due to the effective air gap being larger than that estimated. The ratio N1 /N2 is in close agreement for both rotors 1 and 2, with the error ranging from 1% to 8%. The estimated leakage inductance terms, L rn , are within 20% of their calculated values for rotor 2, and within 8% for rotor 1. In both rotors 1 and 2, the value of L m 2n is subject to the greatest percentage error, suggesting that the parameter estimation method is relatively insensitive to changes in this parameter value. The resistance values are universally higher for the estimated parameter values, which is expected as the calculated resistance values were calculated at room temperature. The estimated values are consistent with an approximately 80 C rise in temperature, which is plausible. From the simple induction mode estimated parameter values we may note the following: The estimated magnetizing inductance, L m 1n is generally lower in simple mode than cascade mode. This suggests that in simple induction mode, the cost function chosen is relatively less sensitive to the value of L m 1n . The error is particularly noticeable for rotor 7, the standard squirrel cage rotor, which suggests that the effective air gap for the cage rotor may be signicantly greater than estimated. This could be the result of localised saturation in the rotor slot bridges. The leakage inductance, L rn , and rotor resistance estimates for rotors 4 and 7 are up to 100% in error. For rotor 7 the Boucherot slots will inevitably lead to increased error as the resistance changes substantially (by design) with shaft speed. However, the errors for rotor 4 are likely to be due to poor quality data, as the results from rotors 1 and 2 are somewhat better in simple induction mode.
6.3.3
Comparison of estimated parameter values to manufacturers parameter values
In the case of rotor 7, the standard squirrel cage rotor, it was possible to compare estimated parameter values against those given by the manufacturer. The details of this comparison can be found in [87], where the conclusion was that there was a close agreement, thus corroborating the parameter estimation method.
150
4
5
3 2
0 Torque (Nm)
1 Torque (Nm) 0 1 2 3
10
15 0
4 5 6 0
(a) Simple induction mode, 4 pole winding supplied
(b) Simple induction mode, 8 pole winding supplied

8
10 5 0 Torque (Nm) 5 10 15 20 0 200 400 600 Rotor 1 Rotor 2 Rotor 4 800 1000 1200 Rotor Speed (rpm) 1400 1600 1800 Torque (Nm)
6 4 2 0 2 4 6 8 10 12 0 200 400 600 800 1000 1200 Rotor Speed (rpm) 1400 Rotor 1 Rotor 2 Rotor 4 1600 1800
(c) Cascade induction mode, 4 pole winding supplied
(d) Cascade induction mode, 8 pole winding supplied
Figure 6.4: Torque-Speed plot in simple induction mode with nominally 90Vrms phase supply voltage. Rotors as described in section 5.1. Experimental results with calculated results using estimated parameter values superimposed.
6.4 Relationship of extracted parameters to the d-q axis model
151
R1 ( ) Rotor 1: Referred to 4 pole side Rotor 1: Referred to 8 pole side Rotor 2: Referred to 4 pole side Rotor 2: Referred to 8 pole side Rotor 3: Referred to 4 pole side Rotor 3: Referred to 8 pole side Rotor 7: Referred to 4 pole side Rotor 7: Referred to 8 pole side 2.7 4.37 2.7 4.37 2.7 4.37 2.7 4.37
Rrn ( ) 0.969 1.96 2.08 14.7 1.01 9.2 0.522 1.54
L rn (mH) 37.3 75.4 22 155 22.5 213 35.4 105
L m 1n (mH) 310 322 317 329 315 328 364 377
R 2n ( ) 2.16 5.45 0.62 19 -
L m 2n (mH) 159 627 46.8 2230 -
N1 N2
0.703 1.42 0.377 2.65 -
Table 6.5: Calculated parameters for the equivalent circuit representation given in gure 6.2
PSfrag replacements
6.4 Relationship of extracted parameters to the d-q axis model

Before concluding this chapter, it will be useful to relate the estimated parameters back to those used in the d-q model. The rst stage in this process is to convert the equivalent circuit of gure 6.2 to the coupled coil equivalent circuit representation of gure 4.3. T equivalent form, but with the left hand inductance zero. The T equivalent form is given by gure shown in gure 6.5.
I1 R1 s2 V2 The left hand side of gure 6.2 may be transformed into a coupled coils representation, using the s1 n Ir
4.6 with N = 1 and L 1 = M. This transformation, and un-referring R 2n gives the representation
L m 1n Rrn /s1
Irn
Irn
I 2n
R2n s2 /s1 I2 n
V1
j1 L m 1n
j1 (L m 1n + L rn ) j1 L m 2n
V2
s2 s1
N1 Nb :1 N2 Na
Figure 6.5: An equivalent form of gure 6.2 The right hand side of gure 6.5 may be transformed to coupled coils form again using gure 4.6. We temporarily dene I2a = I2a , V2 = V2 , convert transformer and inductor of gure 6.5 into coupled coil form, using M N = L m 2n , L 2 = M/N = L m 2n
N2 Na N1 Nb 2
and L 1 = M N . To
152
PSfrag replacements
reinstate V2 and I2a , we redene them to be equal to V2 and I2a respectively. Alternatively the original
Ir These manipulations lead to gure 6.6. L m 1n
sign convention could be maintained by changing the sign of M, this can be deduced from gure 4.5.
I1
R1
Irn
Rrn /s1
Irn
L m 1n
V1
j1 L m 1n
j1 (L m 1n + L rn ) N2 Na N1 Nb
I 2n s2 s1
R2n s2 /s1
L m 2n
Irn
V2
j1 L m 2n
N2 Na N1 Nb
j1 L m 2n
I2
Figure 6.6: A further equivalent form of gure 6.2
Figure 6.6 may be represented as the following matrix equation:
V1
V2 = j1 0 0 s1 L m 1n
L m 1n
s1 s2
0 L m 2n s1 L m 2n
N2 Na N1 Nb N2 Na N1 Nb 2 s1 s2
L m 1n L m 2n
N2 Na N1 Nb
s1 (L m 1n + L rn + L m 2n
+ 0 0 )
R1
0 R 2n 0
0 I 2n Rrn Irn (6.18)
I1
Equation (6.18) is in the same form as equation (4.36), if the complex conjugate is taken of each side and the voltage and current symbols redened. Tracing back through the derivation of section 4.3, in particular from equation (4.31), with equations (4.15), (4.16) and (4.17) we may disaggregate the mutual inductance, Q, and resistance matrices. Therefore equation (6.18) may be re-written as:
V1
V2 = R + + r Q + + js M + I2 n 0 Irn
I1
(6.19)
6.5 Conclusion
153
where: R1 0 R 2n 0 p1 L m 1n 0 j p2 L m 2n 0
2 N2 Na N1 Nb 2
R+ = 0 0
Following back the derivation in section 4.2, which leads to the d-q axis version of the BDFM model in section 3.3, we may deduce the rotor reference frame d-q axis model:
Q+ = 0 0 0 Rrn 0 L m 1n 0 M+ = 0 L m 2n N2 Na N1 Nb L m 1n L m 2n N2 Na N1 Nb
j p2 L m 2n 0
j p1 L m 1n
L m 1n L m 2n
N2 Na N1 Nb
N2 Na N1 Nb
(L m 1n + L rn + L m 2n )
L m 1n
0 L m 1n 0 0 0 L m 1n L m 2n
0 0
N2 Na 2 N1 Nb N N
0 0 0 L m 2n
N2 Na 2 N1 Nb
L m 1n 0 L m 2n 0 (L m 1n + L rn + L m 2n ) 0 0
N2 Na 2 N1 Nb N2 Na N1 Nb
0 L m 1n 0 L m 2n N2 Na 1 b 0 (L m 1n + L rn + L m 2n ) p1 L m 1n
N N
0 0 vdq = s2 0 0 L m 1n vdqs1
0
0 L m 2n N2 Na 1 b 0 0 0 0
N2 Na 2 N1 Nb
0 L m 2n N2 Na 1 b 0 0 p2 L m 2n 0 0 0
N N
p1 L m 1n 0 r 0 0
0
p1 L m 1n 0 0 0 0 0 p2 L m 2n
p1 L m 1n 0 p2 L m 2n N2 Na 1 b 0 0
N N
0 0
0 idq N N s1 p2 L m 2n N2 Na 1 b idqs2 + 0 i 0 0
d idqs1 i dq + dt i s2 dqr
dqr
R1
0 R1 0 0 0 0
0 0 R 2n 0 0 0
0 0 0 R 2n 0 0
0 0 0 0 Rrn 0
0 0 0 0 0
0 i dqs1 0 i dq (6.20) s2 0 i dqr 0 Rrn
6.5 Conclusion
In this chapter a new method of estimating parameters for the BDFM, and indeed standard induction machines has been presented. The method is capable of producing all parameters needed for the referred equivalent circuit, and care has been taken to ensure that the precise meaning, and relationship to other forms of the equivalent circuit, has been maintained. The parameter estimation method has been demonstrated on ve of the seven rotor designs considered in chapter 5, as the remaining two rotor designs were not manufactured. The parameter estimation method has been shown to yield parameter values which lead to a model, the outputs of
154
which were veried by independent experiments at different operating conditions, and shown to be in reasonable agreement. Furthermore, where possible, the parameter values were compared with those provided by the manufacturer, and found to be in reasonable agreement. The estimated parameter values were compared with those calculated using the method described in chapters 2, 3 and 4. The parameter values were shown to be in good agreement generally, particularly so when estimated from cascade induction mode tests. The parameter values were demonstrated to give accurate predictions of the maximum and minimum torque envelope. The agreement of the parameter values increases condence in the accuracy of the parameter calculation method, including the calculation of excess harmonic inductance, which it was argued is a signicant parameter in BDFM design. The chosen solution method to the optimization problem posed by parameter estimation was found to be acceptable. However, particularly when solving the problem of nding cascade induction mode parameter values the crude random search algorithm was found to be relatively slow, as the parameter space is much larger. A signicant decrease in computational time was achieved by stopping the random search prematurely and then using a gradient-based algorithm. The gradient-based algorithm was started at the best 1000 points found by the crude random search. The disadvantage of this approach was that the information gained by the standard deviation was largely lost.
Chapter 7
Modelling for Control of the BDFM

7.1 Introduction
In the chapters 2, 3 and 4 dynamic and steady-state models have been developed for the BDFM. In chapter 4 the BDFM was analysed in steady-state and then from that steady-state analysis new rotor designs were considered in chapter 5. We now turn our attention to dynamic analysis of the BDFM, and use this to assist in the design of controllers. In order to progress the BDFM towards commercial service, not only must the machine be designed to give best possible performance in terms of high efciency, high output power, and other steady-state measures. However as a prerequisite, the machine must be fully controllable so that it can always maintain these desired operating conditions. Typically these two demands tend to be at odds with one another. For example, as the resistance of the machine windings decreases, the efciency will increase. However it will be shown in this chapter that as the resistance decreases the system dynamics become more lightly damped. For example, as a wind turbine generator, the machine must operate at a specic shaft speed to gain the maximum power output from the turbine, while also maintaining a desirable power factor [8]. Similar requirements apply for variable-speed drive applications, such as pump drives [10]. There is therefore a need for robust control algorithms which can regulate the desired quantities. Unfortunately, as noted in chapter 2, the BDFM dynamic equations are non-linear and time-varying, from equation (2.59). Although the transformation to a d-q reference frame synchronous with the rotor in chapter 3 removed the position-varying parameters, the model remains non-linear (the torque equations) and parameter-varying (dependence on shaft speed), as shown in equations (3.35) and (3.40). Although there are some systematic design procedures for designing non-linear control laws, they generally apply to a tiny subset of non-linear systems encountered, and vary signicantly in their performance. This is in contrast to linear systems theory which has a rich set of design methods. The array of advanced linear design tools motivates the derivation of a linearized model, even
155
156
if this model will have signicant limitations. Although it is possible to linearize any sufciently smooth system, the resulting linearized system may be time varying, even if the original system were not explicitly so [26, p. 18]. However, it will be shown that by careful choice of reference frame it is possible to derive a linearized system where the coefcients are constant under certain conditions, and furthermore these equilibrium conditions consist of constant valued inputs. This is in contrast to the only other presented linearized model for the BDFM (which incidentally only applies to a subset of the class of BDFMs analysed here) where the inputs required to keep the shaft speed in equilibrium are sinusoidally varying quantities, thus rendering the linearized system time varying [61]. We will show that in the special case that the machine being analysed has only a single set of rotor circuits, then the linearized model is in a similar form to that derived for the single frame cascade induction machine, [24, 25]. The derivation of a model where constant-valued system inputs are required to hold an operating point was rst proposed in [130]. Such a model is known as a synchronous reference frame model as the chosen reference frame is in synchronism with the currents. However in [130] only a currentfed model is presented, and it only applies to a subset of the class of BDFMs considered in this dissertation. In this chapter we derive a synchronous reference frame model for the entire class of BDFMs analysed in chapter 3. We also describe a direct, systematic method for determining equilibrium conditions that, in contrast to [60] does not require gradient-based optimization, and gives all possible solutions. We then show how the linearised model can be used to design linear control laws for the BDFM. The approach adopted assumes that the objective is speed regulation which should be robust to changes in load torque. This objective covers a wide range of applications. A simple linear controller was proposed for the BDFM in [127]. However no analysis was presented, and the performance of the controller is signicantly worse than that obtained here. We also demonstrate the efcacy of the linearized BDFM model, comparing it against the models of chapters 2 and 3 and against experimental results obtained for the prototype BDFM with rotor 2.
7.2 Synchronous reference frame model

We now derive a synchronous reference frame model for the class of BDFMs analysed in chapter 3. The approach used, although related, is somewhat different to that used in [130], specically because we present a general transformation for voltage source models, where in [130] it was necessary to separate rotor current components into those induced from stator 1 and 2 respectively, in order to derive the synchronous reference frame model. The d-q model derived in chapter 3 was aligned to the rotor angular position, which, we showed, removed the dependence on rotor position from the mutual inductance terms. Nevertheless, when the machine is running in equilibrium, that is at a xed shaft speed in the synchronous mode of
157
operation, the currents are not constant values. However, during the derivation of the steady-state equivalent circuit representation, it was shown in section 4.3 that under BDFM synchronous operation, the frequencies of currents in the two stators, and in the rotor, are the same in the d-q rotor reference frame. This fact suggests that if the reference frame is rotated so that it is in synchronism with these currents, then the equilibrium value of the currents will become constant values. It was derived in section 4.3 that when the BDFM is running in synchronous mode, that is, the shaft speed is related to the supply frequencies by: r = are given by: s = p1 r 1 = p2 r + 2 Hence we expect that if the reference frame is rotated by d-q pair which is a rotation matrix that rotates the signals by (7.2) 1 + 2 p1 + p2 (7.1)
then the frequencies of the currents in both stator windings and the rotor, in the rotor reference frame
s dt, then the currents will be constant

t 0
value in equilibrium. In order to rotate the reference frame we dene a transformation matrix for each s dt. However, from (7.2), s may be expressed in terms either of 1 , the stator 1 supply frequency, or 2 , the stator 2 supply frequency. It makes most conceptual sense to dene the transformation matrix, and therefore the synchronous reference frame, in terms of the stator 1 supply, which it is (initially) assumed is of xed frequency as it is grid connected. The transformation matrix for the stator 1 supply may now be dened: cos( p1 r 1 t + ) sin( p1 r 1 t + ) 0 Tsync ( ) = sin( p1 r 1 t + ) cos( p1 r 1 t + ) 0 0 0 1
(7.3)
where is some constant angular offset.
It is straight forward to show that equation (7.3) transforms stator 1 rotor reference frame voltages to constant values. The results of the transformation are given in table 7.1. In order to transform the stator 2 voltages and currents, the same form of the transformation matrix is used, however the direction of rotation of the transformation matrix must be reversed. This is because the stator 2 voltages and current rotate around the machine circumference in the opposite sense to the stator 1 voltages and currents in the rotor reference frame. This point was discussed in more detail in section 4.3. As (7.3) is a rotation matrix, the inverse (or equivalently the transpose) of the matrix represents a rotation in the opposite sense. In order to transform the rotor currents, the same transformation as used on stator 1 is applied. This is because when the rotor currents were transformed from their real values into the rotor reference
158
frame d-q model, in section 3.3, the transformation matrix used, equation (3.5), was dened using the pole-pairs corresponding to stator 1. Had equation (3.5) been dened using stator 2 pole-pairs, then the direction of the rotation of the rotor currents in the rotor reference frame would have been reversed and consequently the required rotation direction to transform the current into the synchronous reference frame would also have been reversed. We will refer to this new reference frame where the currents and voltages assume constant values in equilibrium, as the synchronous reference frame in the sequel. There is one additional degree of freedom in the transformation into the synchronous reference frame which has not yet been discussed. That is the constant angular offset, . can be used to align the stator 1 and stator 2 d and q axes within the synchronous reference frame. We may exploit this fact to simplify the synchronous reference frame model equations by rendering some (or possibly all) exclusively. the stator-rotor mutual inductance terms diagonal. For the rotor transformations we choose = 0 Table 7.1 (found at the end of the chapter) summarises the stator transformation matrices and the corresponding voltage waveforms in the synchronous reference frame. For convenience the transformations from terminal quantities to the d-q rotor reference frame, equations (3.1) and (3.2), are repeated too.
7.2.1
Transformation from the rotor reference frame to the synchronous reference frame
In section 3.3 the d-q axis model in the rotor reference frame was derived. The nal form of the model was given in equation (3.35). In order to transform the rotor reference frame d-q model into the synchronous reference frame we need to dene the complete transformation matrix, which will as discussed in section 3.3. As discussed in the preceding section the phase offset, is set to zero for the rotor circuits, and to values which diagonalize the stator-rotor mutual inductance matrices in the case of the stator transformations. The full transformation matrix is: Tsync (11 ) 0 0 0 0 0 Tsync (21 )1 0 0 0 0 0 Tsync (0) 0 0 Tsync = . . .. .. .. .. . . . . . . . . 0 0 0 Tsync (0) 0 0 0 0 0 Tsync (0) comprise of N + 2 copies of (7.3) along the diagonal, where N is the number of sets of rotor circuits,
(7.4)
where 11 and 21 are the angles of the rst stator-rotor mutual inductance for stator 1 and 2 respectively as used in equations (3.26) and (3.27). The use of the inverse transformation for stator 1 reects the opposing direction of the currents
159
The d-q model in the synchronous reference frame may now be derived. From (3.35):
s vdq0s1
v s = Tsync Rdq T 1 + r Tsync Q dq T 1 + Tsync Mdq d T 1 sync sync dq0s2 dt sync 0
s i dq0s1 1 d s Tsync Mdq Tsync i dq0 (7.5) dt s2 s i dq r
s i dq0s1 i s + dq0s2 s i dq r
s where Mdq , Q dq , Rdq are given in equation (3.35), and vdq0s1 denotes the stator 1 supply voltage
vector in the synchronous reference frame, as given in table 7.1. Similar notation is used for the other voltages and currents. The torque equation in the rotor reference frame is given in equation (3.38). Transforming into the synchronous reference frame gives: T s i dq0s1 0 1 s Te = i dq0 Tsync 0 2 s2 s i dq r Q dq sr 1 T We therefore dene: Rsync Q sync (1 , r ) Msync Ssync
1 Tsync Rdq Tsync 1 r Tsync Q dq Tsync + Tsync Mdq 1 Tsync Mdq Tsync 0 0 Tsync Q dq sr 1 T
0 0 Q dq sr 2 T
s i dq0s1 1 s Q dq sr 2 Tsync i dq0s2 s 0 i dq r Q dq sr 1
(7.6)
(7.7) d 1 T dt sync (7.8) (7.9) 0 0 Q dq sr 2

T
1 Q dq sr 2 Tsync 0
Q dq sr 1
(7.10)
In terms of the new variable the equations are: s i dq0s1 v s = Rsync + Q sync i s + Msync d dq0s2 dq0s2 dt s 0 i dq r T s s i dq0s1 i dq0s1 1 s s Te = i dq0 Ssync i dq0 s2 s2 2 s s i dq r i dq r
s vdq0s1
s i dq0s1 i s dq0s2 s i dq r (7.11)
160
where J is the moment of inertia of the machine, and Tl is the load torque.
Hence the full dynamic equations in state-space form become: s s s i dq0s1 vdq0s1 i dq0s1 1 i s Msync (Rsync Q sync ) i s + v s dq0s2 dq0s2 dq0s2 s s i dq 0 i dq r r d r r = dt T s s i dq0s1 i dq0s1 1 s r i Ssync i s Tl dq0s2 dq0s2 2J J s s i dq r i dq r
(7.12)
We now consider the result of the transformation to the synchronous reference frame, particularly we show that Rsync , Ssync and Msync are constant-valued, and Q sync is linearly dependent on 1 and r only.
7.2.2
Evaluation of component matrices in the synchronous reference frame
As Rdq is diagonal and the resistance values appear in pairs it is easy to show that:
1 Rsync = Tsync Rdq Tsync = Rdq
because each resistance pair may be represented at the identity multiplied by a scalar which therefore commutes with the transformation matrix. Similarly the stator-stator portions of Mdq must remain unchanged under the transformation as they too have the same structure. From consideration of the rotor-rotor terms, as given in equation (3.24), it can be seen that the rotor-rotor terms remain unchanged, as all the rotor transformation matrices have been chosen to be the same. The stator-rotor portions of M dq which comprise of a scaled rotation matrix of angle 11 for stator 1 and 21 for stator 2 will be diagonalised, and the terms only. remaining terms will be rotated by 11 or 21 respectively. Hence Msync comprises of constant As Q dq is closely related to Mdq similar comments apply. However it is convenient to observe that, from the denitions of Q dq , equations (3.17), (3.18), (3.28) and (3.29), the following relationship holds between Q dq and Mdq : p1 r Mdq0s1 0 Mdqsr 1 0 1 0 1 0 0 0 0 0 .. . = Q dq0s1 0 1 0 1 0 0 0 0 0
0 Q dqsr 1
(7.13)
161
where Mdq0s1 , Q dq0s1 are the stator 1-stator 1 components and Mdqsr 1 , Q dqsr 1 the stator 1-rotor of Mdq and Q dq respectively as given in (3.35). A similar relationship holds for the stator 2 components. From (7.3):
d 1 1 The remaining term, Q sync (r , 1 ) = r Tsync Q dq Tsync + Tsync Mdq dt Tsync will now be analysed.
Therefore we may write:
d 1 Tsync = ( p1 r 1 ) cos( p1 r 1 t + ) dt 0 0 1 0 1 = ( p1 r + 1 ) 1 0 0 Tsync 0 0 0 p1 r

0 1 0 1 0 0 0 0 0
sin( p1 r 1 t + ) cos( p1 r 1 t + ) 0
sin( p1 r 1 t + ) 0 0 0
Q sync (r , 1 ) = Tsync
p2 r
0 1 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0
..
1 It is easy to verify from (7.4) that Tsync and Tsync commute with their neighbouring matrices in (7.14),
( p1 r + 1 )Mdq
0 1 0 1 0 0 0 0 0
Mdq +
0 1 0 1 0 0 0 0 0
0 1 0 1 0 0 0 0 0
..
1 T sync (7.14)
therefore: p1 r
0 1 0 1 0 0 0 0 0
Q sync (r , 1 ) =
p2 r
0 1 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0
..
Therefore as Msync is constant, Q sync (r , 1 ) is linearly dependent on 1 and r .
1 ( p1 r + 1 )Tsync Mdq Tsync
0 1 0 1 0 0 0 0 0
Tsync Mdq T 1 + sync

0 1 0 1 0 0 0 0 0
..
0 1 0 1 0 0 0 0 0
(7.15)
Furthermore it can be noted that equation (7.14) shows that the rst three rows, that is the rows corresponding to stator 1, of Q sync (r , 1 ) are independent of r . This can be seen by noting that
162

0 1 0 0 1 0 0 0 0
therefore as there is no direct coupling between stator 1 and stator 2 it can be seen that the terms in r cancel out. Similarly it is easy to show that Ssync can be written as:
0 1 0 1 0 0 0 0 0
due to the structure of Mdq the 3 3 blocks of Mdq commute with matrices of the form:
p1
Ssync
= Tsync
p2
0 1 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0
..
Mdq which can be written as:

0 1 0 1 0 0 0 0 0
p1
0 1 0 1 0 0 0 0 0
Mdq +
0 1 0 1 0 0 0 0 0
p2
0 0 0 0 0 0 0 0 0
..
1 T sync (7.16)
p1
Ssync
p2
0 1 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0
..
p1
Tsync Mdq T 1 + sync

0 1 0 1 0 0 0 0 0
1 Tsync Mdq Tsync Hence as Msync is constant, then so is Ssync .
p2
0 1 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0
..
(7.17)
It is interesting to note that Youla and Bongiorno proved that a reference frame transformation will exist for any electrical machine, with linear electrical dynamics, which transforms the differential equations into an LTI system under the condition of constant rotor angular velocity [121]. It is not, therefore, surprising that such a transformation can be found for the BDFM. By way of an example consider a rotor with a single set of rotor circuits (or equivalently one to which the model reduction technique of chapter 3 has been applied), with zero sequence states removed:
7.3 Synchronous Reference Frame Model Equilibrium Conditions
163
s vdq s1 v s = dq s2 0
Rdq 1 0 0 0 0 0
0 Rdq 1 0 0 0 0 L dq 1 0 0 0 M1 1 0
0 0 Rdq 2 0 0 0 0 L dq 1 0 0 0 M 11
0
0 0 0 Rdq 2 0 0 0 0 L dq 2 0 M 21 0
1 L dq 1 0 0
0 0 0 0 Rdq r 0 0 0 0 L dq 2 0 M21
0 s i dq 0 s s1 i + dq 0 s s2 i dq r 0 Rdq r M 11 0 0 M 11 M 21 0 d 0 M21 dt L dq r 0 0 L dq r
0 0 0
s i dq s1 i s + dq s2 s i dq r
0 0
1 L dq 1 0 0 0 ( p1 r 1 )M11
(1 p1 r )M11 0
(( p1 + p2 )r 1 )L dq 2 0 ( p1 r 1 )M21 0 1 M11 0
(( p1 + p2 )r 1 )L dq 2 0 (1 + p1 r )M21 0 1 M 11 0
(( p2 + p1 )r 1 )M22 0 ( p1 r 1 )L dq r
The torque equation is given by: 0 0 0 0 0 p 1 M 11 0 0 0 0 p 1 M 11 0 0 0 0 0 0 p 2 M 21 0 0 0 0 p 2 M 21 0
s i dq s1 (( p2 + p1 )r 1 )M21 s i dq s2 (7.18) 0 is (1 p1 r )L dq r dq r
0
0 p 1 M 11 0 p 2 M 21 0 0
p 1 M 11 0
T s i dq s1 1 s Te = i dq 2 s2 s i dq r
s i dq s1 p 2 M 21 s i dq 0 s s2 i dq r 0 0
(7.19)

As it is assumed that stator 1 is connected to the grid, then (from table 7.1) 1 , 1 , 11 and V1 are constant, for a given machine and grid. 1 has the physical interpretation of a constant phase offset between the machine and the supply grid. As the instantaneous phase of the supply grid is, in a sense, arbitrary, 1 may be set freely. Therefore without loss of generality we set 1 = 11 .
164
From table 7.1 the phase of the stator 2 voltage in the synchronous reference frame is given by a constant value. Therefore in equilibrium 2 must be given by:
e 2 = ( p1 + p2 )r 1
s i dq0 s i dq
( p1 p2 )r +
2 dt + 1 t + 2 + 21 . It is immediate that when (7.1) holds, this phase must be (7.20)
The equilibrium currents may be found for any supply voltage by solving the synchronous reference frame dynamic equations, (7.11), for steady-state conditions, that is
d dt
s i dq0 s1 s2
It is assumed that it is desired to specify a value of the stator 2 voltage magnitude, and solve to nd the equilibrium currents, and stator 2 voltage phase offset in the synchronous reference frame. Once this phase offset has been determined, the value of 2 , which constitutes the stator 2 voltage phase offset may be calculated from table 7.1. Under steady-state conditions, from (7.11), the voltage and current equations are related as follows: v s = Rsync + Q sync dq0s2 0
s vdq0s1
= 0.
Using (7.21), the torque can be expressed in terms of the voltage (noting that R sync + Q sync will always be invertible as long as the resistances are non-zero). We omit the zero sequence components as they do not produce torque: 1 s v (Rsync + Q sync )1 T Ssync (Rsync + Q sync )1 v s dq s2 2 dq s2 0 0
T
s i dq0s1 i s dq0s2 s i dq r
(7.21)
Te =
s vdq s1
s vdq s1
(7.22)
partitioning (Rsync + Q sync )1 Ssync (Rsync + Q sync )1 into stator 1, stator 2 and rotor portions, and noting that it must be symmetric as Ssync is symmetric, the torque may be written as: T s s vdq s1 A11 A12 vdq s1 1 s s Te = vdq A12 T A22 vdq s2 2 s2 0 0
T
(7.23)
where A x y denotes the partition of A = (Rsync + Q sync )1 Ssync (Rsync + Q sync )1 and denote portions of A which do not contribute to the torque. Multiplying out (7.23), and combining terms, gives: Te = 1 s 1 s T s s s s vdq s1 A11 vdq s1 + vdq s1 T A12 vdq s2 + vdq s2 T A22 vdq s2 2 2
is used to
(7.24)
As A22 will be of the form of a scalar multiplied by a rotation matrix, then the nal term only depends
s on the magnitude of vdq s2 , (that is s s vdq s2 T vdq s2 ) rather than its phase. Therefore the equilibrium
conditions may be calculated from the following procedure:
165
1. Compute A = (Rsync + Q sync )1 Ssync (Rsync + Q sync )1 .

1 s s s s 2. Choose a magnitude for the stator 2 voltage, hence compute 2 vdq s2 T A22 vdq s2 (as vdq s2 T A22 vdq s2 s is independent of the phase of vdq s2 ).
3. Write the stator 2 voltage in terms of a magnitude,

s vdq s2 =
3V2 , and phase angle, : cos( ) sin( ) (7.25)
3V2
4. Determine the maximum and minimum possible values for the electrical torque, given this stator 2 voltage magnitude. From (7.24) the maximum and minimum torques are give by: Temax = Temin 1 s T s v A11 vdq s1 + 2 dq s1 1 s s = vdq s1 T A11 vdq s1 2 1 s s s s vdq s1 T A12 A12 T vdq s1 3V2 + vdq s2 T A22 vdq s2 2 1 s s s s vdq s1 T A12 A12 T vdq s1 3V2 + vdq s2 T A22 vdq s2 2 (7.26) (7.27)
5. For equilibrium the desired electrical torque, Te will be equal to the load torque, Tl . Clearly there is only one solution for the phase of the stator 2 voltage, however at all other conditions angle, :
s vdq s1 T A12
an equilibrium will only be possible if Temin Tl Temax . Note that in the equality conditions
there are two unique solutions. From (7.24), the solutions may be found by solving for the
cos( ) sin( )
1 3V2
1 s 1 s s s Tl vdq s1 T A11 vdq s1 vdq s2 T A22 vdq s2 2 2
=K
(7.28)
Therefore the solutions are given by: = arcsin(K / a 2 + b2 ) arctan2(a, b) = arcsin(K / a 2 + b2 ) arctan2(a, b) (7.29) (7.30)
where arctan2 denotes the 4 quadrant inverse tangent function of a/b (see [107, sect. 2-115]) and:
T s a b = vdq s1 A12
(7.31)
s 6. Having now determined vdq s2 completely, the equilibrium currents may now be found from s equation (7.21), as vdq s1 is already known.
7. If desired, the terminal voltages and currents may be calculated by using the inverse of the transformations given in table 7.1
166
7.4 Linearization of the Model

The model, given in equations (7.11), may now be linearized. The model is linearized by determining the rst term of the Taylor series expansion evaluated about the chosen equilibrium conditions (see for example [26, p. 18], or [54]). We seek a model of the form: d ( x) = A x + B u dt
(7.32)
where x Rn represents the state vector, and u Rm the input vector, and the original nonlinear system can be represented as x = f (x, u, t). Following [26, p. 18] A =
x
f (x e , u e , t), B = f1 x 1 f = f2 x1 . . .
f (x e , u e , t), where
f denotes:
f1 x2 f2 x2 . . .
.. .
As the stator 1 supply is constant, as discussed, it may simply be regarded as parameter which affects equilibrium conditions. We seek a model of the form:
s s i dq0 i dq0 V2 s2 d s s2 s i dq r = A i dq r + B 2 dt Tl r r r r where A and B constant matrices which depend on the particular equilibrium condition, and meanings from the previous sections.
s i dq0s1
s i dq0s1
(7.33)
is used
to denote small changes in the variables from their equilibrium values. The other symbol retain their
167
The linearized model is then:

s i dq0s1 s i dq0s2 s i dq r
d dt
= r r
1 Msync
1 e Msync Rsync Q sync (1 , r ) 0 0 0 1 se se i se T i dq0s2 T i dq r T Ssync J dq0s1
s vdq0s2 r 0 0 0
eqm.
0 s v 1 1 se 1 Msync vdq0 Msync dq0s2 2 s2 V2 0 0 0 0 0 0
se i s dq0s1 Q sync i dq0s1 1 se Msync (1 ) i dq0 s2 s r i se i dq r dq0s2 s i + dq r 1 r r 0 0 eqm. 0 V2 2 (7.34) 0 Tl 1 J
Q sync (1 ) denotes the partial r derivative of Q sync with respect to r , as Q sync was shown to be linearly dependent on r then the s vdq0s2 derivative will be a constant, that is only a function of the constant 1 . denotes a partial 2 eqm. derivative of the stator 2 supply voltage, which is subsequently evaluated at the equilibrium condise where the superscript e indicates an equilibrium value (e.g. i dq0s1 ).
tions; this will be discussed in the sequel.
1 se v V2 dq0s2
is the partial derivative of the stator 2 voltage with
respect to the voltage magnitude, V2 , and then evaluated at the equilibrium conditions. The presence of partial derivatives of the stator voltages in (7.34) deserves some clarication. From table 7.1 the stator 1 winding voltages are constant in the synchronous reference frame, that is they are independent of the input and state variables of (7.34). However the stator 2 winding voltage is a function of r , which is one of the states of (7.34) as well as the control inputs V2 and 2 . Therefore s vdq0s2 . the linearised system must contain a term by which r affects the machine currents, r eqm. Also the B contains similar terms which derive from the partial derivatives of the stator 2 voltages. It is signicant, however, that these terms are constant valued when evaluated at their steady-state
168
operating conditions. From table 7.1:

s vdq0s2
r
s vdq0s2
eqm.
V2
s vdq0s2
eqm.
eqm.
3V2e sin( e ) = ( p1 p2 ) 3V2e cos( e ) 0 3 cos( e ) 1 se = 3 sin( e ) = V2 vdq0s2 0 3V2e sin( e ) = 3V2e cos( e ) 0
(7.35)
(7.36)
(7.37)
7.4.1
Conclusions for the Linearized Model
From the derivation of the linearized model a number of conclusions can be drawn about BDFM behaviour in the synchronous mode of operation which corroborate well-known results: The BDFM has a synchronous mode of operation much a like a synchronous machine, that is changes in load torque do not affect the steady-state shaft speed. This can be seen as a consequence of the natural integral action, which is present in (7.34): because part of the state vector contains all others much imply r , then the only possible equilibrium conditions are when
d r dt
and the current, but not
r . This change in physical angle was observed using a strobe on the
= 0. Therefore any changes in load torque imply a change in
r = 0, as r
prototype machine, a phenomenon also noted by other authors (e.g. [115, 125]). Changes in both magnitude (of sufciently small amount) and phase (by any amount) of the stator 2 supply voltage do not affect the steady-state shaft speed. From similar reasoning to that applied to the load torque, changes to V2 cannot affect r , or else the equilibrium cannot be maintained. From table 7.1, it can be seen that any change in the phase of V2 , 2 , may be matched by a change in r and the value of the stator 2 supply voltage in the synchronous reference frame remain constant.
7.4.2
Simplication of Linearized Model
It is possible to simplify the linearized model derived if it is assumed that the electrical dynamics are sufciently fast such that they can be considered to be constant. It must be emphasised that this assumption is not true, and may not even be a good approximation. Even if it were a good approximation, it is entirely possible that the simplied model may not be stable when a full linearized model (and hence the BDFM) is, and vice-versa. Nevertheless the simplicity of the model, makes the simplication worth consideration, although its application may be very limited.
169
From (7.34), if the electrical dynamics can be considered constant, then the linearized system may be written in terms of only the mechanical states:
0 d s vdq0s2 = dt r K sync r 0
eqm.
0 0 s 1 vdq0s2 K se sync vdq0s2 K sync 2 V2 0 0
se i dq0s1 Q sync se (1 ) i dq0 K sync s2 r se i dq r 0
r r
eqm.
V2 1 2 (7.38) J Tl
where:
K sync =
1 se T se T se T e i dq0s2 i dq r Ssync Rsync + Q sync (1 , r ) i J dq0s1
(7.39)
which, from (7.23) can be written as:
1 se T v A J dq s2 12 0 A13 T
se vdq s1
A11
A12 A22 A23 T
A13 A33
A23
(7.40)
7.4.3
Linearization Example
By way of an example we consider the linearization applied to a BDFM with a single set of rotor coils, with equations given by (7.18) and (7.19).
170
0 0 0 M 11 0
L dq 1
0 L dq 1 0 0 0 M 11
0 0 L dq 2 0 M 21 0 0
0 0 0 L dq 2 0 M21
M 11 0 M 21 0 L dq r 0
0
1 L dq 1 Rdq 1 0 0
e (1 p1 r )M11 0
1 L dq 1 0 0 0 ( p e )M 1 r 1 11 0
se i qr p1 M11
Rdq 1
M 11 s i dq0s1 0 0 0 is dq0 s2 d M21 s i dq r = dt 0 r L dq r r 1 0 0 J

0 0 0 0
e (( p1 + p2 )r 1 )L dq 2 Rdq 2 e (1 + p1 r )M21 0
0 1 M11 0
e (( p2 + p1 )r 1 )M22 Rdq r e ( p1 r 1 )L dq r 0
e (( p1 + p2 )r 1 )L dq 2 0 e ( p1 r 1 )M21 0 se i qr p2 M21
Rdq 2
0
se i d p1 M11
r
0
se i d p2 M21
r
se se i qs1 p1 M11 i qs2 p2 M21
1 M 11 0
0 0 e ( p1 + p2 ) 3V2 sin( e ) e ( p1 + p2 ) 3V2 cos( e ) 0 0 0 0
0 0
se se ( p1 + p2 )L dq 2 i qs2 ( p2 + p1 )M21 i qr se ( p + p )M i se ( p1 + p2 )L dq 2 i d 2 1 22 dr s2 se se se p1 M11 i qs1 + p1 M21 i qs2 p1 L dq r i qr se se se p1 M11 i d + p1 M21 i d + p1 L dq r i d r s1 s2
e (( p2 + p1 )r 1 )M21
e (1 p1 r )L dq r Rdq r
0
se se i d p1 M 1 1 i d p2 M 2 1 s2 s1
s i dq0
s i dq0
s1 s2
s i dq r
r r
0 0 0 3 cos( e ) 3V2e sin( e ) 0 e V2 3 sin( e ) 3V cos( e ) 0 2 2 (7.41) 0 0 0 Tl 0 0 0 0 0 0 0 0 1
It is interesting to note that (7.41) is similar in form to the linearized model of the doubly-fed cascade induction machine analysed by Cook and Smith [24]. This similarity is not unexpected as the form of the BDFM equations in the synchronous reference frame, are similar to those derived in [24]. The simplied linearised model becomes:
7.5 Simulated Results
171
d dt
r r
=
0 1
se se K sync3 (( p1 + p2 )L dq 2 i qs2 ( p2 + p1 )M21 i qr )+
s2
e 3V K sync3 ( p1 + p2 )r 2e sin( e ) K ( p + p )e 3V e cos( e )

sync4 1 2 r 2
se se K sync4 (( p1 + p2 )L dq 2 i d ( p2 + p1 )M22 i dr )+ se se K sync5 ( p1 M11 i qs1 + p1 M21 i qs2 p1 L dq r qr se se + p M i se + p L K sync6 ( p1 M11 i d 1 21 ds2 1 dq r i dr ) s1
0 K sync3 3 cos( e ) K sync4 3 sin( e )
0 e e K sync3 3V2 sin( e ) K sync4 3V2 cos( e )
se )+ i
r r
+
V2 Tl
0 1 J
where K syncn denotes the n th element of the row vector K sync , dened in (7.40).
(7.42)

The synchronous reference frame model and linearised model was implemented using the Matlab/Simulink computer modelling environment. The nested-loop design rotor, rotor number 1 was simulated being subject to a step increase in instantaneous phase on the the stator 2 voltage supply. at an equilibrium point, which was determined by the method described in section 7.3. It is expected that if the machine is in a stable operating point the operating point will recover after the increase in phase. This phenomenon can be understood from the stator 2 voltage in the synchronous reference frame, given in table 7.1: as 2 changes r changes in such a way to cancel out the change. Five different model variants were compared: 1. Full 26 state coupled-circuit model of chapter 2, 6 states for the stator currents, 18 states for rotor currents and two mechanical states. 2. Full order d-q transformed model (with zero sequence states removed), as described in chapter 3. The prototype nested-loop rotor BDFM has 3 sets of 6 loops, therefore the model has 4 stator states, 6 rotor states and 2 mechanical states, giving a total of 12 states. 3. Reduced order d-q transformed model, using the new method described in section, 3.4.2. The rotor is reduced to 2 states, hence there are 8 states in total. 4. Linearized version of the full d-q model with 12 states. 5. Linearized version of the reduced order d-q model with 8 states. As discussed in chapter 3 the reduced order d-q model appears to be a very good approximation of the full order system. This proposition was corroborated during this series of tests, to the extent That is, 2 of table 7.1, was subject to a step increase at t = 0. The machine was congured to run
172
that at the scale of the graphs there was no visible difference between the two. For this reason the reduced order d-q model, and linearised version have been omitted from all but gure 7.2(b), to aide clarity. The responses of the shaft speed and d-axis stator 1 current in the synchronous reference frame have been compared at different operating points. These measurements were chosen as they illustrate the performance of the different models, the other current measurements had a similar accuracy. Figures 7.1(a) and 7.1(b) show the speed response to a small phase change at two different operating points, one stable, the other unstable. Figures 7.1(c) and 7.1(d) show the d-axis synchronous reference frame current under the same conditions. In each case the response of the 26 state full coupled circuit model, the full d-q model and the linearized d-q model were included. Figures 7.2(a) and 7.2(b) show the the speed response at the same stable operating point used in gure 7.1, but with a much larger phase change. Figure 7.2(a) may be compared with gure 7.1(a), where the phase changes were 6 and 60 respectively. Figures 7.2(a) and 7.2(b) compare the fullorder linearized d-q model and reduced order linearized d-q model. Figures 7.2(c) and 7.2(d) show the real part of the eigenvalues of the linearized full order system at the stable and unstable operating points. The gures show the change in eigenvalues as the operating point changes, brought on by a change in load torque. No values are given outside certain bounds, as this bound corresponds to the maximum torque available from the machine at those voltage magnitudes. Two sets of eigenvalues are plotted in each case, one set in blue and the other in red. The two sets of eigenvalues corresponding to the same load torque and voltage magnitude, but differing voltage phase, as described in section 7.3. Figure 7.3 shows the real parts of the eigenvalues of the full order linearised model and the reduced order linearised model. From gures 7.1, 7.2 and 7.3 we may draw the following conclusions: The full-order and reduced order d-q model are accurate representations of the full coupled circuit model in this case. The full-order linearized model is an excellent approximation of the full system, however as the system is perturbed away from the operating point the accuracy decreases, as expected. The reduced order linearised model is also an excellent approximations of the full system, this fact is corroborated by the eigenvalue plots, which are almost co-incident with those of the full order model, except for the extra eigenvalues, which are stable The stability of an operating point was predicted by the linearized model, as expected. The stability of the simulated prototype BDFM with a nested loop rotor varied signicantly with both load torque, stator voltage magnitudes and shaft speed. Certain regions were naturally unstable.
173
320 315 310
Full 26 state Full dq Linear full dq
900 850 800 750
Speed (rpm)
300 295 290
Speed (rpm) 0 0.1 0.2 0.3 0.4 0.5 t (s) 0.6 0.7 0.8 0.9 1
305
700 650 600 550 500
285 280
450 400 0 0.1 0.2 0.3 0.4 0.5 t (s) 0.6 0.7 0.8 0.9 1
(a) Speed response to 6 phase change at a stable operating point

4
(b) Speed response response to 6 phase change at an unstable operating point
20 15 10 daxis current (A) 5 0 5 10 15 20
2 daxis current (A)
0.1
0.2
0.3
0.4
0.5 t (s)
0.6
0.7
0.8
0.9
0.1
0.2
0.3
0.4
0.5 t (s)
0.6
0.7
0.8
0.9
(c) d-axis stator 1 current response to 6 phase change in sync. ref. frame at a stable operating point
(d) d-axis stator 1 current in sync. ref. frame at an unstable operating point
Figure 7.1: Nested-loop design rotor response to a step change in stator 2 voltage phase. The stable operating point was, 300 rpm, 240Vrms stator 1 phase voltage, 216Vrms stator 2 phase voltage, 5 N m load torque, and the phase change was 6 as indicated. The unstable operating point was 700 rpm, 240Vrms stator 1 phase voltage, 96Vrms stator 2 phase voltage, 5 N m load torque, and the phase change was 0.5 .
174
450
450
Full 26 state Linear red. dq Linear full dq
400
400
Speed (rpm)
300
Speed (rpm) 0 0.1 0.2 0.3 0.4 0.5 t (s) 0.6 0.7 0.8 0.9 1
350
350
300
250
250
200
200
150
150
0.1
0.2
0.3
0.4
0.5 t (s)
0.6
0.7
0.8
0.9
(a) Speed response to 60 phase change at a stable operating point

100
(b) Speed response to 60 phase change at a stable operating point showing reduced order linearized model
100 80 60 40
Soln. 1 Soln. 2
Soln. 1 Soln. 2
50
{} (rad/s)
{} (rad/s) 150 100 50 0 50 Torque(Nm) 100 150 200 250
20 0 20 40
50
100
60 80
150 200
100 400
350
300
250
200 150 100 Torque(Nm)
50
50
100
(c) real part of the eigenvalues of linearized model at a stable operating point
(d) real part of the eigenvalues of linearized model at a unstable operating point
Figure 7.2: The response of the prototype nested loop rotor to a 60 phase change, and the real part of the eigenvalues of the linearized model at the two operating points (shown in blue). Additionally the real part of the eigenvalues of the other solution to another operating point having the same voltage magnitudes, load torque but differing voltage phase (shown in red). The operating points are described in gure 7.1
7.6 Open-loop Experimental Results

20 0 20 40 {} (rad/s) 60 80 100 120 140 200 Linear reduced Linear full
175
150
100
50
0 50 Torque(Nm)
100
150
200
250
Figure 7.3: The real parts of the eigenvalues of the full order linearized model, and reduced order linearised model at the stable operating point described in gure 7.1

Rotor 2 was tested open-loop, in the prototype machine. Full details of the experimental apparatus are given in appendix E. As with the simulated results a step change in phase was applied to the stator 2 voltage source, and the response of the shaft speed logged. The experimental data was compared with simulated data generated using the parameter values found in section 6, and a calculated value for the moment of inertia of the drive train. However as the dimensions of the rotor of the DC load motor can only be estimated, this value could be signicantly in error, the value was therefore adjusted to natural frequency closer to that found in practice. The resistance values were increased to R 1 = 4 (from the estimated values of R1 = 3.6 J = 0.59kgm2 , which was still within estimation error range of the calculated value, but gave a and R2 = 4.3 ) to compensate for
and R2 = 5.5
additional supply impedances present. This increase in impedance was necessary for the simulated results to be stable at the second operating point.
Figure 7.4 shows the response at an unstable operating point, and gure 7.5 at a stable operating point. It is signicant to note that the two operating points differ only in their supply voltage, yet one is stable and the other unstable (as the oscillations in gure 7.4 are divergent and whereas those in gure 7.5 are convergent). Figure 7.6 shows the maximum real part of the linearized system eigenvalues over a range of torques at the two operating conditions. Although the same general shape is obtained in both conditions, the centre point moves, which made one operating point unstable at no-load. Figure 7.6 suggests that stability would be achieved if the load torque was increased, which is precisely what
176
was found in practice.
550
230
500
220
450
210
400
Speed (rpm)
Speed (rpm)
200 190 180
350 300 250
170 160 150 Experimental Full dq Linear full dq Simplified Linear dq 0 0.5 1 1.5 2 t (s) 2.5 3 3.5 4
200 150 0 5 10 15 20 t (s) 25 30 Experimental Full dq 35 40
(a)
(b)
90Vrms , V2 = 81.5Vrms , 0 N m load torque. Both sub-gures show the same event at different time scales
Figure 7.4: Speed response to a 67.5 phase change applied to the stator 2 voltage supply. V1 =
Figures 7.4 and 7.6 show that the full d-q model and its linearization are in reasonable agreement with reality. The simplied linearized d-q model (as described in section 7.4.2) is also included for comparison, however although the natural frequency is approximately right, the damping is in signicant error in both gures. However the damping of the full d-q model and its linearization is still at odds with that found in experiment. It is likely that effects, such as supply impedance and frictional forces are partly the cause, combined with the inevitable errors in the parameter values in the simulation model. Figure 7.4 shows that the natural frequency of the full d-q model and its linearized counterpart are somewhat in error. However the amount of error is comparable with that found in gure 7.2(a) between the non-linear and linearized models, allowing for the differing time scales. In conclusion, the results found by experiment on the prototype machine are in good agreement with those obtained from simulation. However, the machine dynamics themselves are very lightly damped, or unstable at the operating conditions investigated. Even the stable operating condition is hardly usable as a commercial drive, furthermore it was found that if the stator resistances were halved in value in simulation tests, then neither operating point was stable for any load torque with any stator 2 supply voltage. Therefore the use of feedback to stabilize the operating point will now be considered.
177
210
205
Speed (rpm)
200
195
190
185
Experimental Full dq Linear full dq Simplified Linear dq 0 1 2 3 t (s) 4 5 6
68Vrms , V2 = 70Vrms , 0 N m load torque.
Figure 7.5: Speed response to a 22.5 phase change applied to the stator 2 voltage supply. V1 =
0.25
V1=90, V2=81.5 V1=68, V2=70
0.2 max({}) (rad/s)
0.15
0.1
0.05
0.05 20
15
10
5 0 5 Load Torque (Nm)
10
15
20
Figure 7.6: Comparison of the maximum real part of the eigenvalues of the linearized system at two different supply voltage corresponding to the operating condition of gures 7.4 and 7.5, against load torque. The horizontal lines do not correspond to valid operating conditions as the particular combinations of supply voltages cannot deliver the torque.
PSfrag replacements
178 Modelling for Control of the BDFM
V1 , 2 (d ) 1 |V2 | d K 2 Tl BDFM r
Figure 7.7: Stator 2 phase angle control concept
7.7 Closed-loop stator 2 phase angle control

In section 7.6 it was found that there are operating points within the machine operation envelope which are very lightly damped or unstable. Such regions have been found by other groups (e.g [61]). In [84] the author described two control strategies which can be used to stabilize the BDFM. The rst of these, stator 2 phase angle control will now be described. The essential idea is to feed back the speed error as the value of 2 , the phase input to the stator 2 supply, as shown in gure 7.7. V1 and V2 represent the stator 1 and 2 voltages, which will be of the form given in table 7.1, and Tl is the load torque. Together these parameters set the operating point. The designer must choose K to stabilise the closed loop. The idea is to choose K so it robustly stabilizes the system, such that the load torque and stator 2 supply voltage magnitude can be set within some range (that is the operating point can be changed slightly) without loosing stability. However, although the linearised model is accurate in some neighbourhood of the operating point, as soon as the operating point changes, the model will no longer be correct. Furthermore the system is not (or may not) be accurately represented by the linearized model in transition between operating points. Typically if the transition is sufciently slow, then the linearized model will be accurate. This complicates the design of K . A full discussion on the design of K is beyond the scope of this dissertation, however some practical methods are proposed, and then some possible extensions are discussed in section 7.9. If K is restricted to controllers without integral action, then the operating point cannot be changed if the system is stable. This is because 2 can never be a ramp in the steady-state, and therefore the steady-state shaft speed must remain unchanged. Although this might appear restrictive, the operating point may still be changed by modifying 2 , and it is actually advantageous because it means that there will be no steady-state error. Furthermore if K does contain integral action, any error introduced will be constant in the steadystate. The error will be proportional to the xed errors in the measured and demanded value of 1 , 2 and r . However, as these errors can easily be made of the order 1 part in 2 16 (by the use of 16 bit precision electronics), then such steady state errors are small enough to make the use of integral
7.7 Closed-loop stator 2 phase angle control
179
action in the controller possible. The addition of integral action to the controller was found to improve the ability of the controller to stabilize large deviations from an operating point. Regardless of how K is chosen, the steady-state operating point may be set by choice of V1 , 1 , V2 , 2 , and then stabilized by K . The designer therefore has freedom to optimize for steady-state performance, without concern for stability which can be ensured by the design of K . In practice it was found that proportional or proportional-integral controllers could be tuned to give reasonable damping in experiments on the prototype BDFM. Furthermore, these controllers were found stabilize the system over a signicant range of operating points. During these experiments the value of 2 was set open-loop by the value of d from equation 7.1 as shown in gure 7.7. |V2 | was
chosen as a function of d : |V2 | = k max{ |dn n | , 0.2}, that is, |V2 | was proportional to the deviation
from natural speed with a boost at low values.
7.7.1
Experimental Results
The control algorithm was implemented in simulation and on the experimental test rig described in appendix E. Although the analysis in this section requires the synchronous reference frame model, the implementation of the control law simply requires the creation of 3 phase balanced voltages for the stator 2 supply. This is achieved by the inverter, the input of which was designed to accept three parameters, 2 , V2 and 2 . Thus 2 and V2 are xed for the operating point and 2 = K p (d r ) + K i (d r )dt. Figure 7.8 shows the same operating point as gure 7.5 stabilized with a PI controller, K p = 1,
K i = 3, with all measurements in radians (and radians per second).
We now consider the robustness of the controller by changing the operating point. This is achieved
by varying the load torque and shaft speed: gures 7.9(c) and 7.9(d) show the response to a step change in load torque, which is now well damped. Figure 7.9 shows 50 rpm step responses up and down for the BDFM stabilized with the same PI controller. Note that moving operating speed changes the operating point and so the system dynamics are no longer the same. Furthermore, as discussed, the dynamics in transit between operating points linearly with the demanded shaft speed: |V2 | = |V2o | the natural speed as given by (1.2), and |V2o | are likely to be poorly predicted by a linear model. The stator 2 voltage magnitude, |V2 | was scaled
d n n
where d is the demanded speed, n is
is a constant reference voltage.
The oscillations in the rising edge of gure 7.8(a) do not appear in the linearized model, and furthermore the linearized model suffers from a steady-state offset when the shaft speed is changed. This is to be expected from the form of the linearized model, equation (7.34). The results show that even simple PI control can achieve stability which is robust to modest changes in operating conditions, furthermore the linear model is shown to accurately predict the closed loop system even under changes in load torque, which constitute a change in operating point. However it was found that if a step change in equilibrium speed greater than about 60 rpm was de-
180
210
208
Experimental Full dq Linear full dq Simplified Linear dq
210
208
Experimental Full dq Linear full dq Simplified Linear dq
206 Speed (rpm) Speed (rpm) 0 0.1 0.2 0.3 0.4 0.5 t (s) 0.6 0.7 0.8 0.9 1
206
204
204
202
202
200
200
198
198 0 0.1 0.2 0.3 0.4 0.5 t (s) 0.6 0.7 0.8 0.9 1
(a) 67.5 step change in stator 2 voltage phase
(b) 22.5 step change in stator 2 voltage phase
Figure 7.8: Closed-loop speed response under stator 2 voltage phase angle control to a 67.5 and 22.5 step phase change applied to the stator 2 voltage supply. The step phase change was introduced by augmenting 2 of gure 7.7 by injecting a disturbance d, where d represents the step change. V1 = 68Vrms , V2 = 70Vrms , 0 N m load torque, the same operating condition as gure 7.5 manded then the system lost stability.
7.8 Control when
2 dt = ( p1 + p2 )r
1 dt
The concept of stator 2 phase angle control was to stabilize a open-loop operating point of the machine, however no attempt was made to stabilize transitions from one operating shaft speed to another. It can be seen from (7.34) that if is no longer a function of r . This has the effect of removing 2 dt = ( p1 + p2 )r 1 dt, then
s vdq0 s2
r as a state from the linearized
s = 0 as vdq0s2
model. Therefore the linearized model can now have many different equilibrium speeds. This implies a loss of the natural integral action maintaining the synchronous speed, and the machine now behaves more like an asynchronous machine where its steady-state operating point is not robust to torque perturbations. However the integral action can be restored though the design of a suitable controller, K . The conguration of the feedback loop is the same as that for stator 2 phase angle control, as shown in gure 7.7. It was found that reasonable performance could be achieved with K p = 0.3 and K i = 0.6. Figure 7.10 shows experimental and simulated results from this control strategy. The rst thing to note is that gures 7.10(c) and 7.10(d) show that the system was stable for 100 rpm demanded step changes in r . Furthermore it was found that the same controller gains permitted 100 rpm step changes within the speed range from 0 to 600 rpm. Beyond that range the robustness was found to
7.9 Future work on linear model based control
181
deteriorate, and different gains were required for adequate performance However gures 7.10(a) and 7.10(b) show a degraded torque disturbance rejection over that shown in gure 7.9. Furthermore there is a steady-state error introduced, as was also the case in the speed step responses. The steady-state error was due to the implementation of of using r measurement directly, r dt was used which therefore led to errors being introduced. It 2 dt = ( p1 + p2 )r 1 dt. Instead
was found by simulation that a 0.1% error in the value of r was sufcient to produce the steady-state error observed in the experiment. This order of the error is consistent with the limitations of the speed transducer (see appendix E). A steady-state error persisted despite integral action in the controller as the errors in r led to constant errors in angular speed error, which is therefore a ramp error in phase; hence single integral action cannot drive the steady-state error to zero. The steady-state error can, however, be reduced by as much as is desired by reducing the error in implementing the measured position, r (rather than integrating r ), and by generating 2 dt = ( p1 + p2 )r 1 dt. Practically speaking this was achieved by using 1 dt using a phase-locked
loop locked onto the stator 1 voltage. The input to the inverter was then the instantaneous phase of the desired stator 2 voltage, rather than the angular frequency. This alternative design was implemented in software using the experimental test rig, and was found to substantially remove the speed offset. However this was achieved at the expense of requiring the measurement of the instantaneous stator 1 voltage at a high sample rate.
7.9 Future work on linear model based control

Two linear control strategies have been presented in this chapter. Both strategies are scalar strategies and therefore assume that the magnitude of the stator 2 will be set open loop (or in a very slow closed loop). This is a reasonable assumption for general motor drive and generator applications where the bandwidth requirements of speed tracking are considerably more stringent than any other consideration. The speed step response rise-time of both strategies compared favourably to previously published results. For a 50 rpm step the rise time for stator 2 phase angle control was 750 ms and the fall time 350 ms. For the second strategy the rise time was approximately 1500 ms and the fall time 2000 ms. Previously published rise times are 15000 ms for so-called scalar control [127, 126], and between 300 ms and 1200 ms for various eld-orientated control schemes [126, 124]. It must be emphasised that the design of controllers for these strategies has not been thoroughly considered, so the presented results are likely to be worse than those that could be expected in the future. We will now briey consider methods of improving controller performance. Although there are fundamental limitations in linear design methods when applied to non-linear plants, a signicant improvement in performance is typically obtained by designing a series of lin-
182
ear controllers for different operating points and then scheduling between these controllers for the implementation. This technique, known as gain scheduling, seeks to nd a controller which is parameterised in some way on operating points. Traditionally such methods were ad-hoc methods where a plant was linearized at a number of operating points, and controllers designed for each. There are two problems with these methods, rstly that while each controller will stabilize its own operating point, and possibly neighbouring operating points, it may not stabilise the transition between the two. Secondly switching between two (or more) controllers which stabilize an operating point can actually introduce instability. Recent advances on the formulation of gain scheduling problems as LMIs has led to solutions to both these difculties in some circumstances. Rugh and Shamma give an thorough summary of research to date, with a good list of references, [89], and Apkarian and Adams is recommended for controller synthesis [3]. Rugh and Shamma describe two different methods of achieving gain scheduling. The more recent of these methods, quasi-LPV control, uses LMIs to nd parameter dependent controllers which guarantee stability and some performance measures [89, sect. 5]. Unfortunately the method requires that the scheduling variable (possibly a vector) is measurable in real time. This presents a signicant problem for application of these methods to the BDFM. The BDFM operating points depend on the load torque, shaft speed, voltages and currents, and the rotor current is very unlikely to be available for measurement in real time. Rugh and Shamma also give a summary of developments in the traditional gain-scheduling approach where a set of linear controllers are designed for a set of equilibria and then interpolated in some way to produce parameter dependent controllers [89, sect. 4]. In the BDFM the steady-state operating points of the BDFM depend only the supply voltages, shaft speed, and load torque (as the current can be computed in the steady-state). It is likely that the load torque dependence can be replaced by a dependence on the stator currents. Therefore the linear controllers would only depend on these measurable parameters. Under certain conditions, which include bounding the rate of variation of the parameters, stability can be guaranteed. Whether or not it is possible to obtain a stability guarantee gain scheduled controllers are likely to be applicable. One appealing aspect of the approach is that one might choose to specify the only one scheduling parameter, say the shaft speed, if all the other parameters can be expressed as a function of the shaft speed. In an application where the BDFM is connected to a device for which the load torque is a known function of speed (for example if the BDFM was operating as a pump), and the voltages were specied open loop, then a gain scheduled controller would only need to be parameterised on the shaft speed. In [84] the author presented results from a gain-scheduled version of stator 2 phase angle control, where the BDFM was stabilized over a wide speed range, which illustrates that such an approach is possible. There is also scope for including uncertainty explicitly in the model, and designing control laws
7.10 Conclusions
183
which achieve guaranteed stability, and possibly guaranteed performance in the face of this uncertainty. Apkarian and Adams (and others) have shown that it can be achieved for unstructured timevarying uncertainty in the case of quasi-LPV methods [3]. However at the time of writing no results have appeared for the inclusion of structured uncertainty, which is likely to limit the application to the BDFM. The linearization gain scheduling methods can incorporate uncertainty in the design of linear controllers for each operating point. However this does not imply any guarantees for the global properties of the closed loop system. However if a controller has been designed using the linearization gain scheduling methods (or indeed by the quasi-LPV methods) its stability can be investigated using LPV analysis techniques [120]. The practical implementation of the necessary LMIs is similar to that described in [3] for the LPV synthesis problem.
7.10 Conclusions
In this chapter we have presented a synchronous reference frame model for the full class of BDFM machines considered in chapter 2. The synchronous reference frame was shown to be simply an invertible state transformation of the rotor reference frame dq model of chapter 3 The synchronous reference was used to determine all equilibrium points, and it was noted that in general there are either 2, 1 or 0 possible equilibrium points for a given stator 1 voltage, load torque and stator 2 voltage magnitude. The synchronous reference frame model was then linearized about its equilibrium points. Simulation results were presented for the nested-loop design rotor, comparing the full range of models, from the coupled circuit model of chapter 2, through the full d-q and then reduced order d-q models of chapter 3, to the linearized versions of these models. Excellent agreement was found for all operating conditions investigated. The linearized models predicted the stability and damping accurately. The exception to this was the simplied linearized model, which did not correctly predict the damping. Experimental results were presented for the double-layer rotor design, rotor 2 in open-loop conditions, and the linearized model, using parameters estimated in chapter 6, gave good predictions of stability and dynamic performance. Two linear control strategies were presented, stator 2 phase angle control and control when stable operating points, and the performance of these controllers compared favourably to previously published results. It was concluded that the use of gain-scheduled controllers for these control strategies would be likely to further improve performance. 2 dt = ( p1 + p2 )r 1 dt. It was shown that both strategies could stabilize previously un-
184
260
260
250
250
Experimental Full dq Linear full dq
240 Speed (rpm)
240 Speed (rpm)

Experimental Full dq Linear full dq 0 0.5 t (s) 1 1.5
230
230
220
220
210
210
200
200 0 0.1 0.2 0.3 0.4 0.5 t (s) 0.6 0.7 0.8 0.9 1
(a) Speed step response from 200 rpm to 250 rpm

207 206 205 204 Speed (rpm) Speed (rpm) 203 202 201 200 199 198 197 0 0.5 1 1.5
(b) Speed step response from 250 rpm to 200 rpm

204 203 202 201 200 199 198 197 196 195 194 193 0 0.5 1 1.5
t (s)
t (s)
(c) Response to load torque step from 7 N m to 0 N m
(d) Response to load torque step from 0 N m to 7 N m
Figure 7.9: Closed-loop speed response under stator 2 voltage phase angle control to demanded step speed change from 200 rpm to 250 rpm and back down, and step change in load torque from 0 N m to 7 N m and back down. V1 = 68Vrms , V2 = 70Vrms , 0 N m load torque for speed steps, the same operating condition as gure 7.5
7.10 Conclusions
185
202
210
200
208
198 Speed (rpm) Speed (rpm) Experimental Full dq Linear full dq 0 0.5 1 1.5 2 t (s) 2.5 3 3.5 4
206
196
204
194 202 192 200
190
198
0.5
1.5
2 t (s)
2.5
3.5
(a) Response to 5 N m step increase in torque
(b) Response to 5 N m step decrease in torque
320 300 280 Speed (rpm) Speed (rpm) 260 240 220
320
300
280
260
240
220 200 180 0 0.5 1 1.5 t (s) 2 2.5 Experimental Full dq Linear full dq 3 3.5 200
180
0.5
1.5
t (s)
2.5
3.5
(c) Step response from 200 rpm to 300 rpm
(d) Step response from 300 rpm to 200 rpm
Figure 7.10: Closed-loop speed step response when
speed change from 200 rpm to 300 rpm and back down, and step change in load torque from 0 N m operating condition as gure 7.5
2 dt = ( p1 + p2 )r 1 dt to demanded step
to 5 N m and back down. V1 = 68Vrms , V2 = 70Vrms , 0 N m load torque for speed steps, the same
Terminal Phase Voltage Stator 1 Voltage: 2V1 cos(1 t + 1 2V1 cos(1 t + 1 + 2V1 cos(1 t + 1 )
2 ) 3 2 ) 3
Rotor ref. frame d-q quantity
Sync. ref. frame. d-q quantity
Transformation:
Stator 2 Voltage: 2V2 cos( 2 dt + 2 2V2 cos( 2 dt + 2 + 2V2 cos( 2 dt + 2 )

2 ) 3 2 ) 3
2 4 cos( p1 r ) cos( p1 (r 3 p1 )) cos( p1 (r 3 p1 )) 2 2 4 sin( p1 (r 3 p1 )) sin( p1 (r 3 p1 )) 3 sin( p1 r )

1 2 1 2 1 2
3V1 cos( p1 r 1 t 1 ) 3V1 sin( p1 r 1 t 1 ) 0
sin( p1 r 1 t + 1 ) cos( p1 r 1 t + 1 ) 0 1 1 0 0 1 3V2 cos(( p1 p2 )r + ( 2 dt + 1 t) + 2 + 21 ) 3V2 sin(( p1 p2 )r + ( 2 dt + 1 t) + 2 + 2 ) 1 0 sin( p1 r 1 t + 2 ) 1 0 cos( p1 r 1 t + 21 ) sin( p1 r 1 t + 21 ) 0 cos( p1 r 1 t + 21 ) 0 0 1
cos( p1 r 1 t + 11 )
3V1 cos(1 + 11 ) 3V1 sin(1 + 1 ) 1 0
sin( p1 r 1 t + 11 ) 0

2 3
Transformation:
Table 7.1: Comparison of stator voltages in different reference frames
sin( p2 r )
1 2
cos( p2 r ) cos( p2 (r sin( p2 (r

1 2
3V2 cos( p2 r 2 dt 2 ) 3V2 sin( p2 r 2 dt 2 ) 0

2 )) 3 p2 2 )) 3 p2
cos( p2 (r sin( p2 (r
1 2
4 )) 3 p2 4 )) 3 p2
186
Chapter 8
Feedback Linearization for the BDFM

In chapter 7 a synchronous reference frame model for the considered class of BDFM machines was derived, the model linearized, and two control strategies proposed based on the linearized model. The shortcoming of this approach is that linear controller designs for large deviations from an equilibrium point cannot be systematically approached. This is because the linearized model is only accurate for small deviations from the equilibrium point and, although the non-linear model can be used to predict the performance further away from the equilibrium point, it does not aid the design of controllers. There is, therefore, a need for a systematic method of controller design for large deviations from an equilibrium point. As discussed in section 1.1, a direct torque control method was proposed for the BDFM, although never implemented [13, 14, 128], and a number of variations of a eld orientated control scheme were proposed and some of these were implemented [125, 129, 130, 126, 124]. The eld-orientated control schemes all require a current source inverter. We therefore develop non linear control schemes for the BDFM which may be implemented using voltage source inverters. Feedback linearization (FBL) was rst applied to the BDFM by the author in [82, 84]. FBL was applied to the coupled circuit model for a nested-loop design rotor BDFM. However this led to a high order model for which the majority of the states (the rotor currents) could not easily be measured. An equivalent approach is used in this chapter in the d-q synchronous reference frame. It has the advantage of reduced state dimension, and constant current and voltage values in the steady-state. We give two applications of FBL to the BDFM, one regulating speed only and one regulating both speed and ux. We discuss the choice of observer to estimate unmeasurable states, and investigate the entire controller implementation in simulation. We conclude the chapter with some initial experimental results from an implementation of the FBL control schemes on the prototype machine.
187
188
8.1 Feedback Linearization

We start with some denitions and then give a concise introduction to FBL. The results presented here are taken from Isidori and Slotine and Li [50, 101], to which the reader is referred for full details.
8.1.1
Preliminaries
function f (x) Rn , where x Rn , then the Lie derivative of h with respect of f is written as L f h and is given by, L f h(x) = L f h(x) is scalar since used: L g L f h(x) =
x (L f h(x))g(x) x (L f h(x)) xh
Denition 8.1. [101, Def. 6.1]Lie Derivative: Given a scalar function h(x), h R and a vector
x (h)
f (x)
is a row vector, and f (x) a column vector. The following notation is also
L 2f h(x) = L f L f h(x) = L 0f h(x) = h(x)
f (x) f (x)
L kf = L f L k1 h(x) = f
k1 x (L f h(x))
where g Rn is also a vector function. Denition 8.2. [50, p. 137] Relative degree: Given a SISO nonlinear system, within input u R, output y R, and state x Rn : x = f (x) + g(x)u y = h(x) where f , g, and h are sufciently smooth functions, the system is said to have relative degree r at point x o if, 1. L g L kf h(x) = 0, for all x in a neighbourhood of x o and all k < r 1, 2. L g L rf1 h(x o ) = 0 Furthermore the system has relative degree r in the region D o D if the above conditions hold for all x Do . It may therefore be shown that for a SISO system the relative degree is exactly equal to the number of times that one has to differentiate the output, y until the input u explicitly appears in the expression [101, p. 247]. It is possible for a system not to have an well dened relative degree, this can occur
8.1 Feedback Linearization
189
if, for example, the function multiplying the input is not generally zero, but happens to be zero at precisely x o [50, p. 137]. The idea of FBL is to construct a nonlinear state transformation which renders the system linear, or at least the states which are observable at the output must be linear. In order for such a state transformation to exist, the Jacobian of the transformation must be invertible [101, p. 250]. A state transformation satisfying this requirement is the so-called normal form [50, Prop. 4.1.3]. If a system has a (well-dened) relative degree, r at x o , (hence r n), then: 1 (x) = h(x) 2 (x) = L f h(x) . . . r (x) = L rf1 h(h) If r is strictly less than n, it is always possible to nd n r more functions r +1 (x), , n (x) such that the mapping 1 (x) . (x) = . . n (x)
has a Jacobian matrix that is nonsingular at x o and therefore qualies as a local state transformation in a neighbourhood of x o . Furthermore it is possible to choose the value of these additional functions at x o arbitrarily, and it can be shown that they may be chosen in such a way that they are not a function of the input u. The benet of applying this transformation is shown when the original system is written in the new coordinates z i = i (x), 1 i n, hence zi = coordinates can be shown to be ([50, p. 143]): z1 = z2 x i x i . Therefore the system in transformed
z2 = z3 . . . zr 1 = zr zr = b(z) + a(z)u = L g L rf1 h(

1
zr +1 = qr +1 (z) + pr +1 (z)u . . . z n = qn (z) + pn (z)u y = h(x) = z 1
(z))u + L rf h(
(z))
where z Rn is the new state variable, and z k is the k th element of z. qk R and pk R are some functions of z (see [50, Remark 4.1.3]).
190
in this case the system is exactly linearised which is also described as input-state linearization. If r < n, then the input-output system is linear and the transformation is known as an input-output linearization. When the system is MIMO similar results apply. The system considered has m outputs, n states and l inputs and is of the form: x = f (x) +
l j=1
which renders the system from to y linear. When r = n the entire system becomes linear, and
It is immediate that if u is chosen so that u =
1 (b(z) ), a(z)
then the r th state becomes z r = ,
g j (x)u j
y1 = h 1 (x) . . . ym
= h m (x)
the number of times one has to differentiate the i th output yi in order to have at least one component of the input vector u explicitly appearing [50, p.220]. Differentiating the system r i times gives:
taken as the smallest integer such that for at least one g j L g j L rfi 1 h i (x o ) = 0. Therefore ri is exactly
where f Rn , g j Rn , h i R, n is the state dimension. For each output, yi the relative degree, ri is
r u1 y11 L rfi h 1 (x) . L g1 L rf2 1 h 2 (x) L g2 L rf2 1 h 2 (x) L gl L rf2 1 h 2 (x) . . . . = + . . (8.1) . . r ymm L rfm h m (x) ul L g1 L rfm 1 h m (x) L g2 L rfm 1 h m (x) L gl L rfm 1 h m (x) u1 . = B(x) + A(x) . (8.2) . ul For the relative degree to be well-dened, in addition to the condition for SISO systems, is it also necessary that A(x) in (8.2) be at least rank m at x o , where m is the dimension of the output [50, Rem. 5.1.3, Prop. 5.1.2]. In this case it can be shown that the system is exactly (input-state) linearizable iff
m i=1 ri
L g1 L rf1 1 h 1 (x)
L g2 L rf1 1 h 1 (x)
L gl L rf1 1 h 1 (x)
Sect. 5.3]. In order to linearize the system input, u is chosen so that: u1 1 . . . . =0 B(x) + A(x) . . ul m
= n, or if
m i=1 ri
< n then the system is input-output linearizable ([50, Lemma 5.2.1&
(8.3)
unique solution for u, however as long as (8.3) is satised the linearization will still hold. In this case
where v Rm is the new input to the linearized system. Note that if l > m then there will not be a
8.2 Application to the BDFM
191
the linearized system transfer function can be shown to be:

1
r
s 1 y1 . 0 . = . . . . ym 0
0
1 s r2
.. 0
. . ..
0 0 . . .
1 s rm
1 . . . m
(8.4)
As (8.4) comprises of m independent linear systems, m SISO controllers can be designed to stabilize the system. The robustness of the linearized system is dependent on robust design of these outerloop controllers. As FBL can be viewed as a state transformation, it is clear that unless the sum of the relative degrees of all the outputs are exactly equal to the number of states in the system, then the linearized which are unobservable at the output and uncontrollable from the new system input, . These states give rise to the zero dynamics of the linearized system [50, sect. 4.3]. Hence although it is straightforward to design linear controllers to stabilize the observable states, for which the transfer functions are simply of the form 1/s r , such a controller will not affect the zero dynamics of the system. Therefore if the zero dynamics happen to be unstable, then the system does not have internal stability, and therefore the controller cannot work satisfactorily in practice. system is not a complete description of the original system. Specically there are n
m i=1 ri
states

8.2.1 BDFM model in terms of ux linkages
The models derived for the BDFM in previous chapters may be represented wholly, or in part, in terms of circuit ux linkages, rather than circuit currents. From denition 2.7, the ux linkage of a circuit is = Mi where M is the mutual inductance and i the current, thus ux and current are related by a linear transformation, which, from Lemma 2.3, must be invertible. In the case that the BDFM has only one d-q pair representing the rotor states, then (ignoring zero sequence states) the machine has only 6 states. In this case the machine can be fully described by stator 1 and 2 currents, and either the stator 1, stator 2 or rotor uxes. The most useful choice is the stator 2 ux, as the stator 2 ux is strongly dependent on the stator 2 voltage, which is the control input. The rotor ux could be used, however it is considerably more difcult to measure the rotor ux than the stator ux.
192
Therefore in the synchronous reference frame the currents are transformed:
i s = 0 I dq s2 s dq s2 0 Msync22 s i dq s1 s = T i dq s2 s i dq r
s i dq s1
s i dq s1 s 0 i dq s2 s Msyncrr i dq r 0
(8.5)
(8.6)
where Msyncrr and Msync22 are the stator 2-rotor and rotor-rotor portions of M syncrr (the stator 1-stator 2 portion is always zero by design). Note that as M syncrr is rank 2 then the transformation is always invertible. It may be shown that the 2-norm of stator ux linkages is approximately proportional to the rms ux density in the air gap of the corresponding pole number eld. This can be shown by considering the ux linkage in the coupled circuit model in chapter 2, from where the air gap ux density can be related to the mutual inductance terms. It is signicant that the stator ux linkage is related to the rms air gap ux density because the allowable air gap ux density for a machine is limited by the iron laminations, as discussed in section 4.7. Therefore, from (7.12), the full dynamic model in terms of stator 2 ux becomes:
1 i s T Msync T1 (T Rsync T1 T Q sync T1 ) i s + T v s dq s2 dq s2 dq s2 s dq 0 s s2 dq d s2 r r = dt T s s i dq s1 i dq s1 1 s r T Ssync T 1 i s Tl i dq s2 2J dq s2 J s s2 s s2 dq dq
s i dq s1
s i dq s1
s vdq s1
(8.7)
We now consider the speed regulation problem for the BDFM using FBL. We assume that the torque is an unmeasured disturbance, and propose two methods, one where only the speed is regulated, and a second where both the speed and the ux are regulated.
193
8.2.2
Control strategy 1: Speed Only Regulation
s We have two control inputs, the d and q axis components of v dq s2 . We have one output, r . From
section 8.1.1 we differentiate the output, r until the input appears: y1 = r (8.8) T
is is dq s1 dq s1 Tl dr 1 s dy1 1 s = = (8.9) i dq s2 T Ssync T i dq s2 J dt dt 2J s s dq s2 dq s2 T s s i dq s1 i dq s1 d s 1 dTl d 2 y1 d r 1 s = = i dq T Ssync T1 i dq (8.10) s2 dt 2 dt J dt s2 J dt s s2 s s2 dq dq T s s s i dq s1 i dq s1 vdq s1 1 dTl d r 1 s 1 s s = = i dq T Ssync Msync (Rsync T1 Q sync T1 ) i dq + vdq s2 s2 J dt s2 dt J s s2 s s2 0 dq dq (8.11) Therefore the relative degree is 2 as long as the state vector is non-zero, and the system may be
s input-output linearized by choosing vdq s2 such that:
I 0 s s dTl 1 1 s + = P1 0 vdq s1 + P1 I vdq s2 P1 (Rsync T + Q sync T ) i dq + s2 dt J s 0 dq s2 0

s i dq s1
(8.12)
where:
s i dq s1
where is the new dummy input to the system, and therefore the system transfer function becomes: r (s) = 2 . s However it is clear that (8.12) is an under determined set of equations, that is for any chosen there are an innite number of valid solutions. We therefore propose choosing the solution giving the
s minimum value of vdq s2 2 , and which may be computed by:
1 s P1 = i dq T Ssync Msync s2 s s2 dq
(8.13)
s vdq s2
I s dTl s = (AT A)1 AT P1 (Rsync T1 + Q sync T1 ) i dq + + P1 0 vdq s1 s2 dt J s dq s2 0

s i dq s1
(8.14)
194
where: 0 I A = P1 0
As the linear dynamics comprise of a double integrator, they may be stabilized through the use of a phase lead compensator, 2 3 x s + x K1 = s + x
(8.15)
where x is the desired loop cross-over point (in radians per second), and is the desired spread of the pole and zero, which controls the amount of phase lead (and hence the damping and robustness). In this dissertation x = 100 and = 200, which gives a phase margin of 82 . We now consider the case of speed and ux regulation.
8.2.3
Control Strategy 2: Speed and Flux Regulation

output we differentiate it
be 2 from section 8.2.2. To determine the relative degree for the s s2 dq repeatedly:
s i dq s1
The system now has two outputs, r and s s2 2 . The relative degree for the r output is known to dq
2
y2 = s s2 dq
s s2 T s s2 = dq dq
d s s2 dy2 dq = dt dt
dq s s2 T s s2 dq s s2 T s s2 dq dq
s i dq s1
0 s i dq s2 0 s s2 T s s2 s dq dq dq s2 0 1
i s 0 0 0 i s (8.16) dq s2 dq s2 s s dq s2 0 0 I dq s2 T s s 0 0 0 i dq s1 i dq s1 1 i s 0 0 0 d i s (8.17) = dq s2 dq s2 dt s s2 T s s2 s dq dq 0 0 I dq s2 s s2 dq s s 0 0 i dq s1 vdq s1 1 s s 0 0 T Msync (Rsync T1 Q sync T1 ) i dq s2 + vdq s2 s 0 I dq s2 0 (8.18)

2
0 0 0
s i dq s1
Therefore the relative degree for the s s2 dq
s Therefore vdq s2 should be chosen such that:
output is 1, as long as s s2 = 0. dq
s i dq s1
1 1 P1 (Rsync T 1 Q sync T 1 ) i s + P1 v s 1 dTl J dq s2 J dt dq s2 J 0 s s2 dq = s s i dq s1 vdq s1 1 1 s 2 + P2 v s P2 (Rsync T Q sync T ) i dq dq s2 s2 s dq s2 0
s vdq s1
(8.19)
8.3 Towards A Practical Implementation
195
where P1 is given in (8.13) and: 1

s i dq s1
P2 =
Rearranging (8.19) gives:
i s 0 0 0 T M 1 sync dq s2 s s2 T s s2 s dq dq dq s2 0 0 I
s i dq s1
0 0 0
(8.20)
where:
I 1 1 P1 (Rsync T 1 Q sync T 1 ) i s + P1 0 v s 1 dTl J dq s1 dq s2 J J dt s 0 dq s2 + P3 v s = dq s2 s I i dq s1 s s 2 P2 (Rsync T1 Q sync T1 ) i dq + P2 0 vdq s1 s2 s dq s2 0 0 P1 I J 0 P3 = 0 P2 I 0
(8.21)
(8.22)
s s It may be veried that P3 is full rank if i dq s1 = 0 and i dq s2 = 0 and s s2 = 0. Therefore equation dq s (8.21) may be uniquely solved to give vdq s2 and the linearization is valid in all practical situations.
The FBL system may be stabilized by two linear controllers: K 1 (as given in (8.15)) for the speed constant gain as the ux output has relative degree 1. In this dissertation we choose: K 2 = 100 term, and K 2 = x for the ux term, where x is the desired loop bandwidth. K 2 need only be a (8.23)
It is noteworthy that equation (8.21) is similar to equation (15) of [14], as indeed were the methods used to reach it. However the authors do not appear to have realised that this constitutes FBL, and furthermore their derivation is not in the synchronous reference frame.

We now discuss issues related to the practical implementation the FBL control laws described. The rst point which is immediate is that the solution of (8.21) or (8.12) requires the derivative of the load torque to be known. This is very unlikely to be available as a measurement, however normally
196
the load torque will be constant, and therefore the derivative is zero. During torque transients, if the derivative term is assumed to be zero, then any change in torque appears as a disturbance at the input, which will be attenuated by the linear outer loop controller. Therefore we will assume that the torque derivative is zero for controller implementation. The second point which is immediate is that the full state must be measured in order to implement the controller. The stator currents can be measured, but it will not be desirable to measure the stator 2 ux. This is because it would require the addition of instrumentation to the inside of the machine, which would necessitate the controller manufacturer (who will probably not manufacture the machine) modifying the machine. Therefore it will be assumed that it is necessary to estimate at least the stator 2 ux.
8.3.1
Zero Dynamics and Idealized FBL Stability
As discussed in section 8.1.1, when the total relative degree of a system is less than the state dimension, then there are unobservable states, known as zero dynamics. These states must be stable if the system is to be internally stable, as they are not controllable. If the system is not internally stable then the controller is likely to be unusable. The analytical analysis of the zero dynamics is beyond the scope of this dissertation, however the local stability of equilibria can be investigated numerically by performing a standard linearization and then eigenvalue analysis. Calculation of Equilibrium Points The equilibrium points for both FBL methods were determined numerically by using a constrained optimization technique. For both methods the equilibria may be found by setting the derivative term
s zero in (8.7) and solving the resulting equation, omitting the r state. As the input, vdq s2 only appears
in the uppermost equations, the lower equation may be considered a non-linear constraint on the state
s vector. The equilibria are found by substituting (8.14) or (8.21) for vdq s2 , and then minimizing the
2-norm of the derivative vector subject to the aforementioned constraint. This was achieved using the Matlab function fmincon. However to reduce the chance of missing possible solutions the optimization was started from 100 randomly generated starting points, and all results with 2-norms suitably close to zero retained. In both cases multiple equilibria were found for a given load torque, shaft speed and stator 1 supply voltage. Local Stability Analysis of Idealized FBL Figure 8.1 illustrates the effect of stable and unstable zero dynamics for the rst FBL scheme. Figure 8.1(b) shows the maximum real part of all the eigenvalues of the system when linearized an a range of of operating points. It is clear that there are a substantial range of speeds where the zero dynamics are unstable for the load torques considered. We now show the transient response at two equilibria at 200 rpm, the rst with Tl = 0 N m with stable zero dynamics, and the second with Tl = 5 N m with
197
unstable zero dynamics. Figure 8.1(a), shows the resulting speed transient at startup. The noise on the signal for the unstable equilibrium is due to imperfections in implementation of FBL. As these imperfections are tiny (due only to errors in the ODE solver), it serves to illustrate why internal stability is necessary in practice there will be much larger imperfections in the implementation of FBL and consequently greater output errors. Figures 8.1(c) and 8.1(d) show the current and ux at the stable and unstable equilibriums. Therefore we conclude that, even in the case of (nominally) perfect implementation of FBL, if the zero dynamics are not stable, then the controller will not be practically usable, as expected. The same conclusion can be drawn from a similar analysis of the second FBL scheme.
8.3.2
Practical Implementation
As previously discussed, the implementation of FBL requires an observer to estimate the stator 2 ux, or rotor currents. As the BDFM electrical equations are linear parameter varying, parameterised on the speed, standard observer design techniques are not applicable. Four observer designs were considered. The rst three use the measurable stator currents along with the stator voltages, however the fourth only uses the stator voltages. 1. Time varying Kalman lter: as the Kalman lter equations have been solved for linear time varying systems the lter may be applied. However the computational burden is signicant as the state covariance estimate must be computed in real time [105, sect. 4.5]. 2. LPV observer design using LMIs, adapted from an observer for a conventional induction machine in [77], designed to give minimum induced 2-norm error. 3. Novel time varying observer design where stability can always be guaranteed using appropriate gain choices: presented by the author in [85]. 4. Open-loop observer comprising of the electrical system equations. Such an observer is stable for any variation in shaft speed from Theorem 3.5. Of the four designs it was found, through simulation studies, that the robustness of FBL using an observer was actually worse with the three observers which use current feedback. This is because the use of feedback tends to worsen the estimate of the phase of a signal, but generally gives an improved estimate of the magnitude. However, AC electrical machines are generally much more sensitive to changes in phase than changes in magnitude [67]. Therefore the open-loop observer, which simply consists of the rst row of (8.7), was used. In the implementation of the FBL controller it is necessary to convert the measured signals into the synchronous reference frame, and the demanded output voltage from the synchronous reference frame and back into terminal quantities. From a series of simulation tests it became apparent that
198
200.5 200 199.5 199 198.5 198

max({}) (rad/s)
50 40 30 20 10 0
Tl=0Nm Tl=5Nm
Speed (rpm)
197.5 197 0
T =0Nm l T =5Nm
l
10
0.5
1.5
2 time (s)
2.5
3.5
200
400
600 800 1000 Speed (rpm)
1200
1400
1600
(a) [Speed transient

40 30 20 10 0 10 20 30 40 0 0.5 1 1.5 2 time (s) 2.5 3 3.5 4
(b) Maximum real part of linearized system eigenvalues

40 30 20 10 0 10 20 30 40 0 0.5 1 1.5 2 time (s) 2.5 3 3.5 4
Current (A) / Flux (AH)
(c) Current and ux transients at Tl = 0 N m
Current (A) / Flux (AH)
(d) Current and ux transients at Tl = 5 N m
Figure 8.1: Local stability analysis, of FBL method 1 with V1 = 67.6Vrms , showing maximum real part of the linearized system eigenvalues, and the speed and current transients at an equilibrium point. At 5 N m the operating point is unstable, and at 0 N m it is stable, as can be seen from the the maximum real part of the eigenvalues.
199
the FBL controllers were very sensitive to any angular errors during this conversion. Such errors, although generally constant, were found in the experimental apparatus. Therefore a simple adaption law was derived to null out any offset prior to engaging the FBL controller. The adaption controller was: K adap =
t 0 s i dq s2 T s i dq s2 0 1 1 0 2
s i dq s2
s i dq s2
2
dt
(8.24)
s s i dq s2 and i dq s2 to zero by adjusting the angular offset of the d-q transformation. Once the angular error
s K adap is the integral of the sine of the angular difference between the measured stator 2 current, i dq s2 , s and that estimated by the observer i dq . Therefore the adaption law drives any angular error between
s2
has been nulled the adaption law was disengaged. The reference frame transformation is given in
table 7.1, however a software phase-locked loop was used to force 1i + 1 = 0 and then 2i , which is unknown, was set to 2i = T where T is the angular offset found using the adaption law. The value of 1 dt was determined from the phase-locked loop.
Figure 8.2 shows the block diagram of the implementation of the FBL controllers. Figure 8.3 shows simulated responses of the two FBL controllers, using this implementation, where a deliberate mis-match has been introduced between the parameters used in the simulated machine and those used in the simulated observer and controller. The machine model used is the full coupled circuit model developed in chapter 2, using calculated parameter values. The controller and observer used the estimated parameter values given in table 6.3. The model moment of inertia was also deliberately mis-matched by a factor of 2. The gures show that despite this modelling mis-match the controller performance was good, and the system stabilized, and furthermore that there is some decoupling between ux and speed in the second FBL scheme even with ux estimation. However the linearized system transfer functions are clearly not 1/s 2 and 1/s for speed and ux respectively because the transient response is around an order of magnitude slower than predicted. This suggests that the model mis-match has introduced some attenuation into the loop. Figure 8.4 shows that when V1 is increased the linearized system transfer function more closely approximates the ideal double integrator, as the damping factor is approximately correct, and the rise time was around 50 ms whereas under perfect linearization it would have been 30 ms. This is in contrast to gure 8.3 where the rise time is approximately 500 ms, and the damping factor is signicantly different. Figure 8.5 shows results from experimental implementation of the speed only regulation FBL controller. The steady-state offset is due to errors in the linearization. It was found that improved Nevertheless, the step down transient response is very good, and at 190 ms is faster than any previthe second FBL scheme implemented on the prototype machine. The cross coupling between speed and ux and the steady-state errors are due to shortcoming of the linearization. Again it was found that (relatively) large proportional gains gave better performance than the designed speed controller, performance was achieved with a controller with a very high proportional gain, therefore K 1 = 2000.
ously published controller results for the BDFM. Figure 8.6 shows speed and ux step changes for
200
i s1 , i s2 r r BDFM
Tl v1 v2
K adap T
a, b, c Sync. dq
s i dq s1
s s2 dq
Sync. dq a, b, c
s i dq s2
FBL Observer
s i dq s1 is dq s2 dq s
s2
s vdq s2
s vdq s1
K1 0
0 K2
FBL d ,
v1
Figure 8.2: Block diagram of FBL controller implementation
550 500 Shaft Speed, r (rpm) 450 400 350 300 250 200
d r || ||
2
3.5 3 Stator 2 flux, ||2||, (AH) Shaft Speed, r (rpm) 2.5 2
550 500 450 400 350 300 250 200
d r || || 2 d || ||
2
3.5 3 Stator 2 flux, ||2||, (AH) 2.5 2
1.5 1
1.5 1
0.5 0
0.5 0
10
12
time (s)
14
16
18
10
12
time (s)
14
16
18
(a) Speed regulation only FBL
(b) Speed and Flux regulation FBL
Figure 8.3: Transient response of simulated FBL speed and ux controller, using open-loop observer tween machine model and controller model. Step change in load torque at t = 15 s from 0 N m to and adaptive angle offset tuning with V1 = 67.6Vrms . Deliberate model mis-match introduced be-
10 N m, and step changes in speed and ux as shown. The actual, rather than estimated ux is shown.
8.4 Conclusion
550 4
201
500
d r ||2||d ||2||
3.5
400
2.5
350
300
1.5
250
200
10
12
time (s)
14
16
18
0.5
Figure 8.4: Transient response of simulated FBL controllers, using open-loop observer and adaptive angle offset tuning with V1 = 240Vrms . With a higher value of V1 the linearization error appears to decrease. Step change in load torque at t = 15 s from 5 N m to 105 N m, and step changes in speed and ux as shown. The actual, rather than estimated ux is shown.
lation. The need for high gains to minimize the steady-state error is an indication that that feedback linearization has not been effective at low frequencies, and therefore the ideal of innite gain has not been attained. The implementation bandwidth was 3.33 kHz for both controllers, in each period the cpu time was busy for 240 s in each case.
equation (8.15). It was found that K 2 could not be increased beyond K 2 = 10 without loosing regu-
8.4 Conclusion
In this chapter controllers using feedback linearization have been developed for the BDFM, one controlling speed only, and the other controlling ux and speed. The controllers were presented using a transformed version of the synchronous reference frame model where the stator 2 ux replaced the rotor current state. If the rotor has more than 2 states then the derivation is similar, however some of the rotor currents will still appear in the state vector. The control strategies were shown to give promising performance in simulation, even with introduced model mis-match. Initial attempts were made to implement the FBL strategies on the prototype machine, and met with some success. The FBL controller held an equilibrium, and could be made to control the machine. However there was a signicant departure from the ideal feedback linearized system transfer
Stator 2 flux, || ||, (AH)
450 Shaft Speed, (rpm)
202
350
350
300 speed (rpm) speed (rpm) 0 0.5 1 1.5 time (s) 2 2.5 3 3.5
300
250
250
200 0.5
200 0.1 0 0.1 0.2 time (s) 0.3 0.4
Figure 8.5: Experimental step up and step down speed transients for the speed only FBL implementation. The error in the linearization is apparent though the steady-state offset. V1 = 67.6Vrms , the load torque was nominally zero. The speed controller was K 1 = 2000.
360 340 320 300 speed (rpm) 280 260 240 220 200
d r || || 2 d ||2||
1.8 1.6 1.4 1.2 1 Stator 2 flux, || ||, (AH)
0.8 0.6 0.4 0.2
10
time (s)
15
20
25
Figure 8.6: Experimental results for the speed and ux regulation FBL controller at nominally no load. The cross coupling between speed and ux is due to errors in the linearization. Note that as the ux has not been measured, the red line shows ux estimated by the observer. V1 = 67.6Vrms , the load torque was nominally zero. The controllers for speed and ux were K 1 = 1000, K 2 = 10.
8.4 Conclusion
203
function in each case, most notably the absence of high gains at lower frequencies. Therefore further investigation into the cause of this error is required. However, despite the difculties the application of FBL to the BDFM deserves further attention.
204
Chapter 9
Conclusions and Future Work

9.1 Conclusions
The objectives of this dissertation have been: to derive generalised models for a broad class of BDFMs, and to use these models to analyse the steady-state and dynamic performance; to propose a new rotor design and new methods of speed control for the machine. The specic contributions of this dissertation will now be summarised. 1. A generalised framework was developed for the coherent derivation of models for a wide class of BDFMs, of which a machine with any nested-loop design rotor is a subset. The coherence between the different models allowed parameters calculated for the generalised coupled circuit to be transformed to provide parameter values for the other models: (a) A method of computing mutual inductance parameters where the user species the geometry and connection of the individual coils for each circuit was developed. Two calculation methods were proposed, one giving the total inductance, and the other decomposing the inductance into Fourier series terms. The method assumed nite width, but zero depth conductors. The estimation of leakage inductance was included in the method using standard techniques. The methods allowed parameter calculation for any electrical machine (Chapter 2). (b) A generalised coupled circuit model was derived and implemented in software using calculated parameter values which can model any electrical machine (Chapter 2). (c) The generalised coupled-circuit model was transformed into a d-q axis model in the rotor reference frame for a wide class of BDFMs by means of invertible matrix transformations, unlike all previously published models. The only additional assumption required was the lack of harmonic terms other than the fundamental in the stator-rotor mutual inductance matrices, this assumption was shown to be reasonable for all the examples considered. The invertibility of the transformation matrices showed that the transformed model was
205
206
precisely equivalent. It was shown that the position dependence of the model was removed. The proof made use of the circulant nature of the mutual inductance matrices for the class of machine considered (Chapter 3). (d) The rotor reference frame d-q axis model was shown to reduce to a model with 6 + 2N
states (including stator zero sequence states), where N is the number of sets of rotor circuits for the entire class of BDFMs considered and without any observable error. It was shown that the remaining states are stable, and uncontrollable/unobservable, the proof again made use of the circulant properties of the inductance matrices (Chapter 3).
(e) The dynamic rotor reference frame d-q axis model was transformed into complex sequence components for the entire class of BDFMs considered. It was shown that the dynamic sequence component model for the entire class of BDFMs considered could comprise of 2 + N complex states (excluding stator zero sequence states) (Chapter 4). (f) The dynamic rotor reference frame d-q axis model was transformed into a synchronous reference frame d-q axis model for the entire class of BDFMs considered. The transformation was achieved by invertible matrix transformation, therefore the full dynamic nature of the model was preserved, however the steady-state values became constant quantities (Chapter 7). (g) The complex sequence model was solved for steady-state conditions, and used to synthesize an equivalent circuit model for the full class of BDFMs considered. The coherence of the derivation allowed the physical meaning of the component parameters to be retained (Chapter 4). (h) The synchronous reference frame d-q model was linearized for the full class of BDFMs considered, and a method of calculation of all equilibrium points was given (Chapter 7). (i) A model reduction technique for the rotor reference frame d-q model was proposed for nested-loop rotor designs. The technique was shown to be a very good approximation of the full order model for all the examples considered. Furthermore the technique was shown to yield a model which is itself a member of the class of BDFMs considered, thus allowing the model to be transformed into the other forms derived (Chapter 3). 2. The equivalent circuit model was used to investigate the effect of rotor impedance parameters on the performance of the machine: (a) The rotor impedance was shown to limit the maximum torque available from the machine, or equivalently the torque available for a particular stator power factor. In particular the rotor impedance was shown to signicantly depend on excess harmonic inductance, L rh (Chapter 4).
9.1 Conclusions
207
(b) Seven rotor designs were investigated for the BDFM, ve rotors were manufactured and two studied in simulation using calculated parameter values (Chapter 5). (c) The use of cascade induction mode torque-speed measurements were shown to give a qualitative assessment of rotors ability to cross-couple, and hence their performance as BDFM rotors. Experimental measurements were taken for the ve rotors (Chapters 4 and 5). (d) Dimensionless measures were proposed to evaluate potential BDFM rotors. Each measure quantied different components of rotor impedance in the equivalent circuit (Chapter 5). (e) The rotors which exhibited cross-coupling were quantitatively assessed by these measures, and both the nested-loop design, rotor 1 and the new double layer design, rotor 2 were found to have good performance (Chapter 5). (f) The issue of magnetic loading for the BDFM was investigated, and was shown to translate to constraints on equivalent circuit voltages (Chapter 4). 3. A new method of parameter estimation for the BDFM, by experimental means, was proposed for a class of machines having one set of rotor circuits. This class includes machines with a nested-loop rotor when the new model reduction method is applied. The method was applied to the ve experimental rotor designs, and the parameter values veried by independent dynamic and steady-state experiments; by comparison against parameters calculated by the coupled circuit method, and, where possible, compared with manufacturers parameter data. The parameter values were found to be in good agreement, thus verifying both the model and parameter estimation method (Chapter 6 and 7). 4. The stability of the BDFM was investigated: (a) Theorem 3.5 proved that the electrical states of any BDFM in the class of machines considered are quadratically stable. That is for any nite 2-norm supply voltage and any nite shaft speed (possibly with an unbounded rated of change), the 2-norm of currents remains bounded (Chapter 3). (b) The linearized version of the synchronous reference frame d-q axis model was used to analyse the local open-loop stability of equilibria in the synchronous mode of operation. The analysis was found to be in close agreement to that found by experiment (Chapter 7). 5. Control strategies were proposed to stabilize and improve the damping of speed regulation for the BDFM: (a) A linear control scheme, stator 2 phase angle control was proposed to stabilize and improve the damping of these equilibria. The scheme was veried by experiment, and
208
was found to give dynamic performance similar to alternative non-linear control schemes when operating close to the operating set point (Chapter 7). (b) A control scheme where the stator 2 instantaneous voltage is phase-locked to the rotor position was proposed and its stability analysed using the linearized model. Experimental results were presented showing an improvement in robustness to changes in operating point (Chapter 7). (c) Two methods of applying feedback linearization to the BDFM were presented, shown to work in simulation, and an initial attempt made to implement them on the prototype machine (Chapter 8). 6. Possible pole number combinations for BDFMs were characterised for most practical situations: (a) Theorem 2.15 gave necessary and sufcient conditions for avoiding unbalanced magnetic pull, under the assumption that the air gap ux density contained no harmonic elds. (b) Theorem 2.12 gave necessary and sufcient conditions for non-coupling of isolated single layer stator winding pole combinations, which are sufcient for double layer windings.
9.2 Future Work

The work of this dissertation suggests a number of possible directions for research:
9.2.1
Analysis of the BDFM

to be most straight forward to proceed in the rotor reference frame, the results could be checked by experiment and against the coupled-circuit model. This would complement the steady-state harmonic analysis of Williamson et. al. [115].
1. The d-q axis model could be extended to include harmonic terms. In the rst instance it is likely
2. The general analysis framework presented assumed that leakage inductance terms only contributed to the self inductance of each coil. It is likely the analysis will generalise directly to allow mutual leakage inductance parameters. 3. The model reduction method proposed for nested-loop rotor designs, although shown to be accurate, comes with no performance guarantees. Further investigation may yield such guarantees, or at least give more specic guidance as to circumstances when the reduction method is likely to be accurate. 4. It was shown that the d-q model in the rotor reference frame could be reduced to a model with 6 + 2N states. It is believed that this is a minimal realisation for the system, i.e. that it is not
9.2 Future Work
209
possible to reduce the system order further without observing an error. However a proof of this is difcult as the system is parameterised by the shaft speed. There are, however, some results in the literature on necessary and sufcient conditions for the minimality of parameter varying systems which may apply to the BDFM [5]. 5. The class of rotors considered was neither necessary nor sufcient as functional BDFM rotor designs. Rotor 6 is a functional BDFM rotor, but is not within the class, and rotor 3 is within the class and is not a functional rotor. A simple necessary condition was given in chapter 2, and it may be possible to tighten this condition so that it becomes necessary and sufcient. One way to proceed would be to attempt to characterise the class of rotors encompassed by the necessary condition, although this is likely to be a challenging problem. 6. The new model order reduction method was used to reduce the order of nested-loop design rotors to a rotor which had a single effective set of loops. The reduction technique diagonalized the rotor-rotor self-inductance matrix, and then truncated the system. However the model could be left un-truncated (and therefore free from error) and then transformed into an equivalent circuit representation. This will, in general, lead to off-diagonal resistance terms, but these can be synthesised by a cage structure, as is the case with the squirrel cage rotor. This would lead to a full equivalent circuit representation for nested-loop rotor designs. 7. The analysis in this dissertation has exclusively been for BDFM machines where the synchronous speed is of the second stator would lead to a machine with a natural speed given by r =
1 +2 . p1 + p 2
However it was rst noted by Hunt that reversing the phase sequence
1 . p 1 p2
For
Hunts purposes this mode of operation was not ...of much commercial value. [49, p. 407]. This fact was rst noted by Field in comments on Hunts original paper [48, p. 669]. It is believed that the next mention of this differential mode of operation did not appear until 1997 in Williamson et. al. [115]. Williamson et. al. comment that the subtractive mode would require a rotor with an unfeasibly small numbers of rotor bars. This comment stems from the assumption that the rotor will be of nested loop design, and in differential mode the rotor that it will lead to a relatively high value of rotor leakage inductance due to excess harmonic inmust have a number of nests equal to p1 p2 , rather than p1 + p2 . The problem with this is ductance: L rh in the notation of chapter 4. However, as has been illustrated in this dissertation it is possible to reduce L rh by judicious rotor design, therefore the differential BDFM mode should be considered. It is anticipated that analysis in the differential mode will be similar to the additive mode, and it is likely that many of the results will carry over, including the analysis using circulant matrices. The differential mode offers a higher natural speed, and therefore the possibility to run the machine at a higher speed, which will deliver more power from the same frame size machine.
210
8. Although the modelling techniques presented have been shown to yield good agreement with experimental results, all the tests, with the exception of that shown in gure 6.3 were performed at relatively low ux densities. As the modelling techniques assume innite iron permeability the accuracy will deteriorate as the ux density increases. Initial attempts to derive saturation factors to compensate led to the conclusion that such factors would, in general, be time and load varying [34]. Therefore further work is required to investigate how allowances can be made for saturation.
9.2.2
BDFM Machine Design

tal verication, and the approach may not turn out to be the right compromise. The verication may take the form of loss measurements under different loading conditions, possibly through the use of calorimetric measurement, or through the use of nite elements analysis.
1. Although the magnetic loading for the BDFM was addressed, the treatment requires experimen-
2. This dissertation presented a quantitative method of evaluating BDFM rotors by the evaluation of easily calculable machine parameter values. This suggests the possibility of searching for an optimum rotor design. The search will require the parameterisation of a class of rotor designs over which to search, and the search may be performed numerically or possibly analytically. 3. The experimental procedure to estimate machine parameter values was used without the measurement of rotor bar currents. As described in chapter 6, if these measurements are taken, possibly using Bluetooth technology [83], then the stator-rotor turns ratio can be determined and it is likely that all parameter values will be determined with greater accuracy. 4. This dissertation has principally been concerned with the analysis of the BDFM, albeit with emphasis on its implication for machine design. However nowhere in the current literature are there any design guidelines for the BDFM. Such guidelines could be developed from the equivalent circuit, which retains accuracy whilst being elegant in its simplicity. Issues to consider include how the machine parameter values affect the rating of the machine, its efciency and power factor, and whether the stator 1-rotor-stator 2 turns ratios affect the performance. The proposed performance measures contribute in this direction, however they do not provide a complete picture.
9.2.3
Stability analysis and control, including parameter estimation

is likely to lead to a controller with improved performance over a wider range. Unfortunately recent developments in gain scheduling, offering guaranteed stability and performance are not immediately applicable as they require measurement of all the machine currents, including the
1. As discussed in chapter 7, the use of gain scheduling on the two control strategies proposed
9.2 Future Work
211
rotor currents. Nevertheless it is likely that gain scheduling will lead to satisfactory performance. 2. Prempain et. al. recently presented a LPV based control algorithm for the standard induction machine [77]. A similar scheme will apply to the BDFM, and offers a systematic method of non-linear controller design for the BDFM, however the stability of the overall system, and in particular the zero dynamics will need careful consideration. 3. Hopfensperger et. al. presented two control schemes for the cascaded doubly-fed machine [46, 45]. These schemes should be investigated, in particular their claimed validity to the BDFM. 4. Although a number of practical control schemes have been presented for the BDFM [125, 129, 130, 126, 124], the stability and performance of these schemes has not been investigated. Using the generalised transformation to the synchronous reference frame presented in this dissertation, the stability of such schemes may be analysed, possibly leading to directions to guarantee stability or even performance. 5. Recent developments in model predictive control have lead to the possibility of considering the problem of controlling an electrical machine as a search for an optimal inverter switching pattern. For the induction machine it has been shown, albeit only in simulation at present, that such a strategy can lead to reduced losses without any performance degradation [76]. A similar approach could be investigated for the BDFM. 6. A off-line parameter estimation method was proposed in this dissertation for the BDFM. However, as discussed above, model-based control algorithms often require accurate machine parameters, some of which change signicantly during machine operation (such that the resistance parameters). Therefore on-line parameter estimation techniques could be investigated. One possibility is to design an adaptive observer, such as the extended Kalman lter (although this has no convergence guarantees), alternatively approaches such as that proposed by Castaldi et. al. for the induction machine may be applicable to the BDFM [23]. 7. The initial attempts to implement feedback linearization met with limited success. Future work should investigate the stability of the proposed feedback linearization controllers, and seek to address the reasons for the partial experimental success. It is also possible to use feedback linearization to control different outputs other than the speed and ux, for example the speed and stator 1 power factor could be controlled. This can be achieved by appropriate choice of output function.
212
Appendix A
Mathematics
A.1 Trigonometric Results
and R. Hence
n1 k=0
Lemma A.1. Given w = exp( 2 j ) then n

n1 k=0
n1 k=0
sin( 2kp + ) = 0 n
cos( 2kp + ) + j sin( 2kp + ) = 0 and therefore n n
wkp exp( j) = 0 where p Z = tn where t Z,

n1 k=0
cos( 2kp + ) = n
Proof.
n1 k=0 n1
wkp exp( j) = exp( j)
wkp+
k=0
= exp( j) exp(0) + exp(
22 pj 2(n 1) p 2 pj ) + exp( ) + + exp( ) n n n
Noting that exp( 2npj ) = 1 due to the restrictions on p, and that exp( j) = 0:
n1
exp( j)
k=0
wkp 1 exp(
2 pj 2 pj 22 pj ) = exp( j) exp(0) + exp( ) + exp( ) + n n n + exp( 2(n 1) p ) n 1 exp( 2 pj ) n
= exp( j) 1 exp(2 pj) = 0
A.2 Linear Algebra

orthonormal basis of eigenvectors for Rn . Theorem A.2. [38, p. 270, vol. 1] A matrix A Rnn , such that A = AT (A is symmetric) has a
Theorem A.3. A symmetric matrix, A = A T Rnn , can be written as: A = T TT

213
214
Mathematics
where of A.
is a diagonal matrix of the eigenvalues of A, and T is a matrix of corresponding eigenvectors
the eigenvectors of A, and 1 , 2 , , n the corresponding eigenvalues. Furthermore let: T = x1 x2 xn
Proof. From Theorem A.2 the eigenvectors of A form a basis for Rn . Let x 1 , x 2 , , x n represent
as the eigenvectors for a basis for Rn then T is rank n (full rank) and thus T 1 exists. Furthermore as the basis is orthonormal hence T is orthogonal, that is T T T = I hence T 1 = T T . Therefore: AT = A x1 x2 xn = Ax 1 Ax 2 Ax n = 1 x1 2 x2 n xn = T
A = T TT
Lemma A.4. A real symmetric matrix has real eigenvalues. Proof. Let x Rn be an eigenvector of A = A T Rnn , and the corresponding eigenvalue. Note that we know x to be real from Theorem A.2. From the denition of eigenvectors: Ax = x and as both A and x are real Ax is real, hence is real. Lemma A.5. [131, sect. 2.3] Let A be a square matrix partitioned as follows: A where A11 and A22 are also square matrices. If A and A11 are non-singular then: A11 A12 A21 A22 where
1
A11 A12 A21 A22
A1 + A1 A12 11 11
1
A21 A1 11
A21 A1 A1 A12 11 11
1
is non-singular.
A22 A21 A1 A12 and furthermore it can be shown that A is non-singular if and only if 11
If A and A22 are non-singular then: A11 A12 A21 A22

1
A1 A21 1 22
1 A12 A1 22 A1 + A1 A21 1 A12 A1 22 22 22
where A11 A12 A1 A21 and furthermore it can be shown that A is non-singular if and only if 22 is non-singular.
A.2 Linear Algebra
215
the union of the eigenvalues of the (block) matrices on the diagonal.
Lemma A.6. The eigenvalues of a square upper or lower (block) triangular matrix, T C nn , are
Proof. Without loss of generality we assume T to be upper block triangular: T = A 0 B D
of D, with corresponding eigenvalue d C. Therefore it is immediate that: A 0 are eigenvalues of T . immediate that: B D 0 xd = d 0 xd
where A Cmm ,D Cnmmn , and B is compatibly partitioned. Let x d Cnm be an eigenvector
hence we conclude that d also an eigenvalue of T , and indeed by induction that all eigenvalue of D Similarly, let x a Cm be an eigenvector of A, with corresponding eigenvalue a C. Again it is A 0 triangular T follows the same lines. Lemma A.7. Given: A = A 0 + 1 A1 + + n An Rkk , C R jk and Q = Q T > 0, Q Rkk AT Q + Q A + C T C < 0
T
B D
xa 0
= a
xa 0
and hence we conclude that the eigenvalues of A are also eigenvalues of T . The proof for a lower
max min with n n < n R then:
Ai Q + Q Ai + C T C < 0 Ai where Ai denotes the i th extreme point of A. The extreme points are the points corresponding to the corners of the parameter box containing A. There are 2 n such corners. Proof. We will prove the result for a single parameter, the result for n parameters follows by inducmax min tion. Hence without loss of generality let A = A 0 + 1 A1 and 1 < 1 < 1 . max min Note that A can be written as A = A 0 + (1 (1 ) + 1 )A1 , 0 < < 1. By denition We T
have:
max max (A0 + A1 1 )T Q + Q(A0 + A1 1 ) + C T C < 0
min min (A0 + A1 1 ) Q + Q(A0 + A1 1 ) + C T C < 0
hence from denition 2.10, as > 0 and 1 > 0:

max min max min (A0 + A1 (1 + (1 )1 )) Q + Q(A0 + A1 (1 + (1 )1 )) + C T C < 0 T
and this is exactly equivalent to A T Q + Q A + C T C < 0, as all possible values of 1 are covered.
216
Mathematics
Lemma A.8. [36, Lemma 3.2] & [93, Prop. 1.34] (Shur Complement): Given P = P Cnn , R = R Cmm and S Cnm then: P S
S R
<0
R <0 P SR
1
<0
P <0 RS P
1
S <0
Lemma A.9. The following LMIs are equivalent: AT X + X A BT X 1 C BT X C AT X + X A BT X C XB 1 D I 1 C T I
AT X + X A X B I
1 D T < 0 CT
where R > 0 and X Rnn = X T .
DT < 0 D I XB CT I DT < 0 D I 2
Proof. Applying Lemma A.8 to the rst LMI yields the following equivalent condition: AT X + X A B X
T
XB I
1 C T
1
(I )1 1 C
1 D < 0,
I < 0
which, as > 0, is equivalent to: AT X + X A BT X XB I CT DT (I 2 )1 C D < 0, I 2 < 0
applying the Shur complement again, in reverse, yields the third LMI. The second LMI is obtained by multiplying through by . Lemma A.10. The rows of Q are orthonormal to one another, that is: Q Q T = I , and furthermore, if Q is square then Q T Q = I , where: Q= cos() cos( 2 p + ) cos( 22 p + ) cos( 2(n1) p + ) n n n 2 sin() sin( 2 p + ) sin( 22 p + ) sin( 2(n1) p + ) n n n n 1 1 1 1 2 2 2 2
where R, n N, n 3, p Z = tn, t Z.
A.2 Linear Algebra
217
Proof. cos() cos( 2 p + ) cos( 2(n1) p + ) n n sin( 2 p + ) cos( 2 p + ) 2 n n Q Q T = sin() sin( 2 p + ) sin( 2(n1) p + ) . . n n . . n . . 1 1 1 2 2 2 2(n1) p 2(n1) p cos( n + ) sin( n + ) From Lemma A.1, it can be seen that terms due to the product of non-trigonometric terms and trigonometric terms will be zero. Hence: cos()2 + cos( 2 p + )2 + + cos( 2(n1) p + )2 n n 2 2 p 2 p 2(n1) p = sin() cos() + sin( n + ) cos( n + ) + + sin( n + ) cos( 2(n1) p + ) n n 0 cos() sin() + cos( 2 p + ) sin( 2 p + ) + + cos( 2(n1) p + ) sin( 2(n1) p + ) 0 n n n n sin()2 + sin( 2 p + )2 + + sin( 2(n1) p + )2 0 n n n 0 2 cos() sin()
1 2 1 2 1 2
. . .
1 1 1 1 Using the trigonometric identities cos(A)2 = 2 + 2 cos(2 A), sin(A)2 = 2 2 cos(2 A), sin(A) cos(A) =
sin(2 A) with Lemma A.1 gives:
2 0 n 0
n 2
0
n 2
0
n 2
Denition A.1. A matrix, C Cnn is circulant if there exists c0 , , cn1 such that Ci j = c(i j+n)
mod n
means the k modulo n i.e. the remainder after dividing k by n.
where the rows and columns of C are numbered between 0 and n 1. k mod n
For n = 6 a circulant matrix is: c5 c4 C = c3 c2 c1 c0 c1 c2 c3 c4 c5 c0 c1 c2 c3 c4 c5 c0 c1 c2 c3 c4 c5 c0 c1 c2 c3 c4 c5 c0 c1 c2 c3 c4 c5 c0
218
Mathematics
Denition A.2. A matrix F Cnn is a Fourier matrix if Fqr = w (q1)(r 1) / n, where w = exp( j2 ). A Fourier matrix is of the form: n 1
1 w w2 w2 w4 1 1 F = . . . . . n . . . . 1 w (n2) w2(n2) 1 w (n1) w2(n1) and hence full rank.
.. .
w(n1) w2(n1) . . . (n2)(n1) w (n1)(n1) w
Lemma A.11. Any Fourier matrix, F Cnn , as detailed in denition A.2 is unitary, i.e F 1 = F ,
Proof. From the denition F may be written: 1 1 1 1 1 w w2 w 2 w4 1 1 F = . . . . . n . . . . 1 w (n2) w2(n2) 1 w (n1) w2(n1) hence: 1 1 1 1 1 1 1 1 1
.. .
w(n1) w2(n1) . . . (n2)(n1) w (n1)(n1) w
1 w 1 w2 w(n1) 1 w w2 w(n1) w 2 w4 w2(n1) 1 w2 w4 w2(n1) 1 1 F F = . . . . . . . . .. .. . . . . . . n . . . . . . . . . . . . 1 w (n2) w2(n2) w(n2)(n1) 1 w (n2) w2(n2) w(n2)(n1) 1 w (n1) w2(n1) w(n1)(n1) 1 w (n1) w2(n1) w(n1)(n1) by considering the unit circle it is easy to see that
n1 0
wkn = 0, k Z = 0. Hence:
FF = I
orthonormal. Then there exists a matrix, D Rn(nk) such that D
Lemma A.12. [131, p.19] Let D Rnk (n > k) be such that D D = I , thus the columns of D are D is a unitary matrix.
A.2 Linear Algebra
219
where 1 i n denotes the column number of F. Furthermore, if x C n is an eigenvector of C
ble dimension are the eigenvectors of C, and that the corresponding eigenvalue is =
Theorem A.13. For any circulant matrix C Cnn the columns of the Fourier matrix, F of compatin1 j=0
c j w j (i1) ,
then x is also an eigenvector, and if C Rnn and the corresponding eigenvalue for x is then the eigenvalue of x is . Finally, from Lemma A.11, C is diagonalisable, that is the eigenvectors of C are distinct. Proof. We shall show that for each column of F, i : C Fi = i Fi , where Fi denotes the i th column of F. C F may be written: c0 c1 c2 c3 cn1 1 1 1 cn1 c0 c1 c2 cn2 1 w w2 c1 cn3 1 w2 w4 1 cn2 cn1 c0 CF = . . . . . . . .. .. . . . . . . n . . . . . . . . . . c 1 w (n2) w2(n2) c3 cn1 c0 c1 2 c1 c2 c3 cn1 c0 1 w (n1) w2(n1) c0 c1 c2 c3 1 cn2 1 cn3 1 1 = . . . . n . . c 1 1 c0 1 cn1 1
.. .
hence:
w(n1) 2(n1) w . . . (n2)(n1) w w(n1)(n1)
noting that w n = 1 (from denition A.2): c0 c1 c2 c3 cn1 1 cn1 c0 c1 c2 cn2 w 2 c1 cn3 w 1 cn2 cn1 c0 C F2 = . . . . . . .. .. . . . . n . . . . . . . c w(n2) c3 cn1 c0 c1 2 (n1) w c1 c2 c3 cn1 c0 c0 + wc1 + w 2 c2 + + w (n1) cn1 w[c0 + wc1 + w 2 c2 + + w (n1) cn1 ] 2 2 (n1) cn1 ] 1 w [c0 + wc1 + w c2 + + w = 2 F2 = . . n . w(n2) [c + wc + w 2 c + + w (n1) c ] 0 1 2 n1 (n1) 2 (n1) w [c0 + wc1 + w c2 + + w cn1 ]
cn1 c0 c1 c2 c1 1 cn2 cn1 c0 C F1 = . . . .. . . n . . . . . c c3 cn1 2 c1 c2 c3
.. . c0
n1 i=0 ci n1 i=0 ci n1 i=0 ci
cn1
= 1 F1 n1 ci i=0 n1 i=0 ci . . .
220
Mathematics
n1 j=0
by induction we conclude C Fi = i Fi . By inspection the i th eigenvalue, i = w

kn
c j w j (i1) .
If x denotes any column of F, then there will always exist a column of F which is equal to x, as
of 1 gives a column of F.
= 1k Z and w = w 1 , so multiplication of each element of x by an appropriate representation
real.
Therefore if x if an eigenvector of C, then x is also an eigenvector, the and if is the eigenvalue n1 j (i1) corresponding to x then, as i = , it is clear that the eigenvalue of x is if C is j=0 c j w
Lemma A.14. Given a circulant (not necessarily symmetric) matrix, C R nn , n N 3, then: || QC Q T = sin() cos() 0 cos() sin() 0 0
vector to 0 . Furthermore, if C is symmetric then the rst and second columns of Q T are eigenvectors of C corresponding to eigenvalue , and of course, from Theorem A.4 must be real. Where: cos() cos( 2 p + ) cos( 22 p + ) cos( 2(n1) p + ) n n n 2 sin() sin( 2 p + ) sin( 22 p + ) sin( 2(n1) p + ) n n n n 1 1 1 1 2 2 2 2
n1 i=0 n1
where C, 0 R are eigenvalues of C,and the third column of Q T is the corresponding eigen-
Q=
= 0 =
ci wi
ci
i=0
where p N, p = nt, t N. ci is the element on the i th column of row 1 of C, and w = exp( j2 ). n Proof. Firstly, we know that the third column is an eigenvector from Theorem A.13. We now assume, without loss of generality, that p < n, as for all p > n p1 > 0 : p = p1 + nk, k, p1 N and exp( 2nk ) = 1. n
exp( j)
exp(0)
exp( 2 pj + j) exp( 2 pj ) cos( 2 p ) + j sin( 2 p ) n n n n 22 pj 22 pj 22 p 22 p C exp( n + j) = exp( j)C exp( n ) = exp( j)C cos( n ) + j sin( n ) . . . . . . . . . 2(n1) pj 2(n1) pj 2(n1) p 2(n1) p exp( exp( cos( n ) + j sin( n ) + j) ) n n
cos(0) + j sin(0)
A.2 Linear Algebra
221
recalling that C is real, taking real and imaginary parts gives: sin() sin( + ) sin( 2 p + ) sin( 2 p + + ) n n sin( 22 p + ) = sin( 22 p + + ) C n n . . . . . . C cos( 2(n1) p + ) n
and from Theorem A.13, for some C: cos(0) + j sin(0) cos( 2 p ) + j sin( 2 p ) n n cos( 22 p ) + j sin( 22 p ) = exp( j) n n . . .
cos( 2 p + + ) + j sin( 2 p + + ) n n = || cos( 22 p + + ) + j sin( 22 p + + ) n n . . . cos( 2(n1) p ) + j sin( 2(n1) p ) cos( 2(n1) p + + ) + j sin( 2(n1) p + + ) n n n n
cos( + ) + j sin( + )
C:
hence if C is symmetric then = 0, and we can conclude that the rows of Q are all eigenvectors of 0 0
sin( 2(n1) p + ) sin( 2(n1) p + + ) n n cos( + )) cos() 2 p cos( 2 p + + )) cos( n + ) n cos( 22 p + ) = cos( 22 p + + )) n n . . . . . .
cos( 2(n1) p + + )) n
The eigenvalue results come directly from Theorem A.13.
C Q T = Q T 0 0 0 0 0
then with the previous discussion and from Lemma A.13: C x x = x x
If C is not necessarily symmetric the we proceed as follows. Dening: exp( j) exp( 2 pj + j) n 22 pj x = exp( n + j) . . . 2(n1) pj exp( + j) n 0 0
222
Mathematics
now notice: C x = x x x
1 2 1 2 1 2 1 2 j 2 j 2 j 2 j 2 1 2 1 2
= x 1 j
j 2 j 2
x 1 j
0 0 0 0
1 2 1 2 1 2 1 2
j 2 j 2 j 2 j 2
= x Now notice that x x

1 2 1 2 j 2 j 2
is proportional to the rst two columns of Q T , hence recalling
that the third column of Q T is an eigenvector of C then we may write: cos() sin() || 0 C QT = QT sin() cos() 0 0 Lemma A.15. Given a symmetric circulant matrix, C Rnn , n N 3, that is, a circulant matrix such that C = C T , then: 0 0
QC
where =
n1 i i=0 ci w
Q=
where p N, p = nt, t N. ci is the element on the i th column of row 1 of C, and w = exp( j2 ). n Proof. From Lemma A.14 we may write: 0 0 QC = 0 0 Q 0 0 0 0 0 d d (Q T ) = 0 0 Q (Q T ) QC d d 0 0 0
and: cos() cos( 2 p + ) cos( 22 p + ) cos( 2(n1) p + ) n n n 2 sin() sin( 2 p + ) sin( 22 p + ) sin( 2(n1) p + ) n n n n 1 1 1 1 2 2 2 2
d d 0 0 (Q T ) = dt dt 0 0 0
post multiplying by
d (Q T ): d
A.2 Linear Algebra
223
Differentiating Q gives: sin() sin( 2 p + ) sin( 22 p + ) sin( 2(n1) p + ) n n n 2 22 p 2(n1) p 2 p cos() cos( n + ) cos( n + ) cos( n + ) n 0 0 0 0
dQ = d
Using a similar argument to that found Lemma A.10: 0 0
QC
Applying the chain rule then gives the result.
d (Q T ) = 0 0 d 0 0 0
Lemma A.16. Given a circulant matrix (not necessarily symmetric), C R nn , n N 3, then: Q Q Q Q

T
0 0 = 0 0 0 0 0 0
0 0 S
if C is symmetric, or otherwise: cos() sin() || sin() cos() = 0 0
Q Q
Q Q
0 0 0
0 0 S
where , 0 R1 and S R(n3)(n3) and: 2 sin() n

1 2
Q=
cos() cos( 2 p + ) cos( 22 p + ) cos( 2(n1) p + ) n n n sin( 2 p + ) n

1 2 1 2
sin( 22 p + ) sin( 2(n1) p + ) n n

1 2 n1 i i=0 ci w j2 ). n
complement of the subspace spanned by the rows of Q. Furthermore =

n1 i=0 ci , th
where p N, p = nt, t N, and Q an orthogonal matrix whose rows span the orthogonal where ci is the element on the i column of row 1 of C, and w = exp(
and 0 =
Proof. The proof follows the same steps whether or not C is symmetric, therefore without loss of generality we assume C is symmetric. Lemma A.14 proves that each column of Q T is an eigenvector
224
Mathematics
from Lemma A.10:
with corresponding eigenvalues , , 0 . Therefore we may write: 0 0 C Q T = Q T 0 0 0 0 0 0 0 QC Q T = Q Q T 0 0 0 0 0 0 0
We now consider the entire transformation. Firstly note that an orthogonal Q exists as Q Q T = I (Lemma A.12). Therefore we may write: Q Q C Q Q
0 0 0 0 0 0 0 T
QC Q T = 0 0 0 0 0
Q Q
CQ
CQ
QC Q T Q C Q T
QC Q
T T
Q C Q
Noting that C Q T = Q T
l 0 0 0 l 0 0 0 0l
, and because C T is also circulant from Lemma A.14 that QC =
Q, and that by denition Q Q T = 0: 0 0 0 0 0 = 0 0 0 T 0 Q C Q
Q Q
Q Q
QC Q T 0
0 Q C Q
T
Lemma A.17. Given M as dened below, and Q, Q as dened in Lemma A.16 (with n = p, p = p1 , = 0) then M is rank 2 and:
M1 2 p1 + p 2 2 p1 + p 2 2
Q p1 0
0 Q p2
Q Q
M1 2 2 =3 3 M2 2 M2 2
cos(1 p1 ) M1 2 sin(1 p1 )
M1 2
p1 + p 2 sin(1 p1 ) 2 p1 + p 2 cos(1 p1 ) 2
0
p1 + p 2 2 p1 + p 2 2
cos(2 p2 )
M2 2
sin(2 p2 ) M2 2
0 0 0 0 p1 + p 2 sin(2 p2 ) 0 2 p1 + p 2 cos(2 p2 ) 0 0 2 0 0
A.2 Linear Algebra
225
where: M1 cos(( 1 ) p1 ) M1 cos(( 2 1 ) p1 ) p M1 cos(( 2 1 ) p1 ) M1 cos(( 2 2 1 ) p1 ) 3 p1 p 3 p1 4 2 4 M1 cos(( 3 p 1 ) p1 ) M1 cos(( p 3 p 1 ) p1 ) 1 1 M = M2 cos(( 2 2 ) p2 ) M2 cos(( 2 ) p2 ) p M cos(( 2 ) p ) M cos(( 2 2 ) p ) 2 2 2 2 2 2 3 p2 p 3 p2 M2 cos((
4 3 p2
2 ) p2 ) M2 cos((
2 p
1 ) p1 ) M1 cos(( ( p 1) 2 1 ) p1 ) p M2 cos(( ( p 1) 2 2 ) p2 ) p 2 M2 cos(( ( p 1) 2 3 p2 2 ) p2 ) p 2 4 M2 cos(( ( p 1) p 3 p2 2 ) p2 ) M1 cos(( ( p 1) 2 p

2 3 p1 4 3 p1
M1 cos(( ( p 1) 2 1 ) p1 ) p
4 3 p2
2 ) p2 )
and p= p1 + p2 , M1 , M2 R1 , n N, p1 , p2 N, 1 , 2 , R1 , and: 2 4 )) cos( p1 ( )) 3 p1 3 p1 4 2 2 )) sin( p1 ( )) sin( p1 ) sin( p1 ( 3 p1 3 p1 3 1 1 1 2 2 2 2 2 )) cos( p2 ( )) cos( p2 ) cos( p2 ( 3 p2 3 p2 2 4 2 )) sin( p2 ( )) sin( p2 ) sin( p2 ( 3 p2 3 p2 3 1 1 1 2 2 2 cos( p1 ) cos( p1 (
Q p1 =
Q p2 =
Proof.
Recalling that cos(A) cos(A ) =

1 2
1 2
sin(2 A ), and using Lemma A.1:

M1 2 M 1 2
cos() +
M1 2 M1 2 M2 2 M2 2
1 2
cos(2 A ), sin(A) cos(A ) =

2 p1 ) p 2 p1 ) p 2 p2 ) p 2 p2 ) p M1 2 M1 2
1 2
sin() +
cos(1 p1 )
Q p1 0
0 Q p2
sin(1 p1 ) 0 2 M =3 M2 3 cos(2 p2 ) 2 M2 sin( p ) 2 2 2 0
cos(1 p1 + sin(1 p1 + 0
cos(1 p1 + sin(1 p1 + 0
cos(2 p2 + sin(2 p2 + 0
M2 cos(2 p2 + 2 p2 ( p1) ) 2 p M2 sin(2 p2 + 2 p2 ( p1) ) 2 p 0
2 p1 ( p1) ) p 2 p1 ( p1) ) p
(A.1)
226
Mathematics
Now from Lemma A.19 (A.1) is rank 2, with the rst two rows spanning the entire row space. Furthermore it is clear that the rst two rows of Q are in the row space of the rst two rows, hence Q is in the null space. As cos() = cos( + 2), and similarly for sin(), the RHS of the previous expression may be written as: M1 M1 cos(1 p1 ) cos(1 p1 + 2pp1 ) 2 2 M 2 p1 M1 1 sin(1 p1 ) sin(1 p1 + p ) 2 2 0 0 2 3 M2 2 p2 2 p M2 3 cos(2 p2 ) cos(2 p2 + p p ) 2 2 M2 sin( p ) M2 sin( p + 2 p2 2 p ) 2 2 2 2 2 2 p p 0 0 M1 M1 cos(1 p1 ) cos(1 p1 + 2pp1 ) 2 2 M M1 1 sin(1 p1 + 2pp1 ) 2 sin(1 p1 ) 2 0 0 2 =3 M2 M2 3 cos(2 p2 ) cos(2 p2 + 2pp1 ) 2 2 M2 sin( p ) M2 sin( p + 2 p1 2 2 2 2 2 2 p 0 0 ), sin(A + ) cos(A) =
1 2 1 sin() + 2 sin(2 A + ), M1 2 M1 2
multiplying this by Q T , and then simplifying using cos(A + ) cos(A) =

1 1 ), cos(A + ) sin(A) = 2 sin() + 2 sin(2 A + ),
with Lemma A.1 gives:
1 cos(2 A + 2 1 1 cos(A + ) sin(A) = 2 sin() + 2 sin(2 A + 1 1 sin(A + ) sin(A) = 2 cos() 2 cos(2 A + ) 1 2
0 2 p2 ( p1) 2 p( p1) M2 cos(2 p2 + )) 2 p p 2 p2 ( p1) 2 p( p1) M2 sin(2 p2 + ) 2 p p 0 M1 cos(1 p1 + 2 p1 ( p1) ) 2 p M1 sin(1 p1 + 2 p1 ( p1) ) 2 p 0 M2 cos(2 p2 + 2 p1 ( p1) ) 2 p M2 sin(2 p2 + 2 p1 ( p1) ) 2 p 0 sin(1 p1 + cos() +
cos(1 p1 +
2 p1 ( p1) ) p 2 p1 ( p1) ) p
2 3
hence:
0 0 0 0 2 M 2 p1 + p 2 p1 + p2 M2 p1 + p2 cos(2 p2 ) sin(2 p2 ) 0 2 2 2 2 p +p M2 1 2 sin(2 p2 ) M2 p1 + p2 cos(2 p2 ) 0 2 2 2 2 0 0 0 sin(1 p1 )

M1 2 p1 + p 2 2 p1 + p 2 2
M 1 p1 + p 2 2 2 M p +p 1 1 2 2 2
cos(1 p1 ) M1 2
p1 + p 2 sin(1 p1 ) 2 M 1 p1 + p 2 cos(1 p1 ) 2 2
cos(1 p1 ) sin(1 p1 ) cos(2 p2 )
Q p1 0
0 Q p2
Q Q
M1 2 2 =3 3 M2 2 M2 2
M1 2
M1 2
p1 + p 2 sin(1 p1 ) 2 p1 + p 2 cos(1 p1 ) 2
0
p1 + p 2 2 p1 + p 2 2 M2 2
sin(2 p2 ) M2 2
0 0 0 0 p1 + p 2 sin(2 p2 ) 0 2 p1 + p 2 cos(2 p2 ) 0 0 2 0 0
A.2 Linear Algebra
227
Lemma A.18. Given M as dened in Lemma A.17, and Q, Q are dened as in Lemma A.16 (with n = p, p = p1 , = 0) then M is rank 2 and:
M1 2 p1 + p 2 p1 sin(1 p1 ) 2 p1 + p 2 p1 cos(1 p1 ) 2 p1 + p 2 p1 cos(1 p1 ) 2 p1 + p 2 p1 sin(1 p1 ) 2
Q p1 0
0 Q p2
Q d (M) d Q
where: p= p1 + p2 , M1 , M2 R1 , n N, p1 , p2 N, 1 , 2 , R1 , and Q p1 , Q p2 are dened in Lemma A.17 Proof. Differentiating M by changes the cosine terms to negative sine terms, which is equivalent to augmenting the angular offset in each term by /2, and multiplying the resultant by p 1 , or p2 respectively. Following this change through the proof in Lemma A.17 gives the desired result. Lemma A.19. The matrix Q R4( p1 + p2 ) is rank 2, and furthermore the rst two rows span the whole of the row space of Q, where Q is dened as: 1 A cos(1 ) A cos(1 + p2 pp2 ) 1+ A sin( ) A sin( + 2 p1 ) 1 1 p1 + p 2 Q= B cos(2 ) B cos(2 + 2 p2 ) p1 + p 2 2 B sin(2 ) B sin(2 + p2 pp2 ) 1+ Proof. Firstly note that from Lemma A.10, that the rst two rows are linearly independent. To proceed with of Q: the proof we dene X C1( p1 + p2 ) as follows, noting that X is a linear combination of rows 1 and 2 X = A exp( j1 ) rearranging X gives: = A 1 exp( j2 j1 ) B exp( j2 ) B exp( j2 j p2 pp ) B exp( j2 j 2 p1 ( p1 + p2 1) ) p1 + p 2 1+ 2 B A p +p 1 = exp( j2 j1 ) B exp( j2 ) exp(0) B exp( j2 p2 pp ) exp( j2 p1 + p2 ) 1+ 2 1 2 B A 2 exp( j2 j1 ) B exp( j2 ) B exp( j2 + j p2 pp ) B exp( j2 + j 2 p2 ( p1 + p2 1) ) p1 + p 2 1+ 2 B
p B exp( j2 j 2 p1 ( p1 + p2 1) )) exp( j2 ( p1 + p2p)(+1p+ p2 1) ) p1 + p 2 1 2
1 A exp( j1 j p2 pp2 ) 1+
M1 2 2 =3 3 M2 2 M2 2
M1 2 M1 2
0
p1 + p 2 p2 sin(2 p2 ) 2 p1 + p 2 p2 cos(2 p2 ) 2
M2 2
M2 2
0 0 0 0 p1 + p 2 p2 cos(2 p2 ) 0 2 p1 + p 2 p2 sin(2 p2 ) 0 0 2 0 0
B cos(2 + B sin(2 +
A cos(1 + A sin(1 +
where 1 , 2 , A, B R and p1 , p2 N.
2 p1 ( p1 + p2 1) ) p1 + p 2 2 p1 ( p1 + p2 1) ) p1 + p 2 2 p2 ( p1 + p2 1) ) p1 + p 2 2 p2 ( p1 + p2 1) ) p1 + p 2
A exp( j1 j 2 p1 ( p1 + p2 1) ) p1 + p 2
228
Mathematics
similarly we may dene Y , also a linear combination of the rst two rows of Q: Y = A 2 exp( j2 + j1 ) B exp( j2 ) B exp( j2 j p2 pp ) B exp( j2 j 2 p2 ( p1 + p2 1) ) p1 + p 2 1+ 2 B
jB 2A B 2A
Now taking j2 )X +
exp( j1 j2 )Y gives the 4th row of Q.
j exp( j1 + j2 )X + 2B exp( j1 j2 )Y gives the 3rd row of Q and 2 B exp( j1 + A A
Appendix B
Prototype Machine Stator and Rotor Design Details

In this chapter the details of the prototype machine stator and rotor designs, as described in chapter 5, are given. Table B.1 summarises the conguration of the prototype machine stator.
Parameter Frame size Stator core length Stator slots Stator winding 1 Poles coils per phase turns per coil Conguration wire diameter rated current Measured air gap ux density at 50Hz, at phase voltage given Stator winding 2 Poles coils per phase turns per coil Conguration wire diameter rated current Measured air gap ux density at 50Hz, at phase voltage given Value D180 190mm 48 4 16 (series connected) 10 star 1.2mm 6.78Arms (with 6A/mm2 ) 0.31Trms at 238Vrms 0.12Trms at 95Vrms 8 16 (series connected) 20 star 1.2mm 6.78Arms (with 6A/mm2 ) 0.30Trms at 238Vrms 0.12Trms at 95Vrms
Table B.1: Prototype machine stator summary specication. measured at the winding synchronous speed with a single pole pitch search coil.
229
230
B.1 Prototype machine frame details

The prototype machine frame was a standard induction machine frame, manufactured by Marelli Motori SpA. The details are given in table B.2. Parameter Frame size Marelli Part No. Rated output power Rated Speed Rated Torque Peak torque Rated efciency Moment of Inertia Weight Value 180mm A4C180L4 22kW 1465rpm 143Nm 358Nm 90.3% 0.11kgm2 122kg
Table B.2: Original Marelli Motori SpA. 4 pole induction machine specication used for the Prototype BDFM frame
B.2 Prototype Machine Stator Windings

We now detail the stator windings design and analysis on the prototype machine. We begin by reviewing the standard winding factors used in 3-phase machine.
B.2.1
Machine Winding Factors
Winding factors are commonly used in the design of AC electrical machine windings. The per-phase magnetizing inductance and air gap magnetic ux density of a balanced three phase winding when excited with a balanced three phase source is given by [43, eqs. (78), (81), p. 40]:
2 3wd0 Nph
L=
nQ
Kn n Kn n
(B.1) cos(n t) (B.2)
30 I Nph B(, t) = g
nQ
where w is the machine stack length, g is the effective air gap width (as dened in section C.2), d is the machine diameter, N is the number of turns per phase, n is the pole number (number of pole pairs) of the eld being considered, K n is the combined winding factor for the n pole number eld and the current in the winding in one phase is I cos(t).
231
For a balanced three phase winding only odd harmonics of the fundamental p pole number eld appear, with the exception of third harmonics. It can be shown that for a balanced three phase winding: Q = p|6c + 1| where c = {0, 1, 2, 3, }. For proof see [43, eq. (36b)]. (B.3)
The winding factor for the n th harmonic, K n comprises of the following terms:
K dn =
sin(n M p s /2) M p sin(ns /2) sin(nws /2) nws /2
(B.4) (B.5) (B.6) (B.7)
K pn = cos(n Ms s /2) K sn =
K n = K dn K pn K s n
where M p is the number of slots per phase per pole, s , ws are the slot pitch and slot mouth width in radians as in gure 2.2, Ms is the number of slots that the winding is short-pitched by . Equations (B.4) and (B.5) can be found in [99] as equations (10.130) and (10.131). Equation (B.6) is found in [112], although it is effectively proved by noting that (2.24) contains K sn as a factor.
B.2.2
Stator Winding details
Table B.3 shows the details of the stator windings. Figures B.1 and B.2 show colour coded winding diagrams for the 8 and 4 pole windings respectively. Further machine parameters relating to dimensions and the like can be found in tables C.2 and C.1. Throughout this dissertation unless otherwise stated, the stator state order will be 4 pole states then 8 pole states. The numerical values for the stator mutual inductance given are for the nested-loop design rotor, Rotor 1. The values differ slightly for the different rotors because the effective air gap changes. Mutual inductance (excluding leakage) is inversely proportional to effective air gap width, therefore the stator mutual inductance terms for other rotors can be closely approximated by scaling the terms given as appropriate. The subscripts are in the order of: coil group number,connection number within a coil. So for means return path current direction. example a21 means coil number 2, connection number 1, and current in the out direction. A bar (a)
No. slots short pitch. Ms
No. slots /phase/pole Mp
No. layers
Turns per coil
No. coils per phase
No. turns per phase Nph
Wire dia.
Wire area (mm2 )
Fund. dist. fac. Kd
Fund. pitch fac. Kp
Fund. slot w. fac. Ks
Fund. winding fac. K
(mm) Stator 4 pole winding 2 Stator 8 pole winding 1 2 2 20 16 320 1.2 4 2 10 16 160 1.2
1.13
0.9577
0.9659
0.9998
0.9248
1.13
0.9659
0.9659
0.9991
0.9322
Table B.3: Stator winding details and fundamental winding factors
232
233
Slot: 8 pole: 4 pole:
1 a11 a82 A11 A43
2 a12 A12 A44 14 a32 c 22 A c
3 c c
4 b A14 C 16 b 24 A C c c
5 b11 b C
6 b a11 C C
7 a12 C B a21
8 a22 c C
9 c11 c B11 B 21 c c B
10 c b
11 b B 11 A 23 b b B b
a21 B 12 A 24 b
12 b
A13 C 15 c 23 A C c
B B 22 c b
13 a31 a22 A21
17 b b C11 C 29 b b C
18 b a31 C C 30 b a51 C C
19 a41 a32 C B
20 a42 c C B 32 a62 c C
A13 25 a51 a42
A14 26 a52 c
B B 34 c b
A21 35 b B A31 47 b b B b
A22 36 b
a41 B
Slot: 8 Pole: 4 pole:
27 c c
28 b A34 C 40 b A44 c c
31 a61 a52 C B
33 c c B B 45 c c B
a61 B A32 48 b
A31 A23
A32 A24 38 a72 c A42
A33 C 39 c A43 c
B B 46 c b
37 a71 a62 A41
41 b b C
42 b a71 C
43 a81 a72 C B
44 a82 c C B
A33 A34 C C C C B A41 A42 Thus for phase a of the power (8 pole) winding the connections are as follows: (a51 , a51 ), (a52 , a52 ), (a61 , a61 ), (a62 , a62 ), (a71 , a71 ), (a72 , a72 ), (a81 , a81 ), (a82 , a82 ) (a11 , a11 ), (a12 , a12 ), (a21 , a21 ), (a22 , a22 ), (a31 , a31 ), (a32 , a32 ), (a41 , a41 ), (a42 , a42 ),
B B
a81 B
Thus phase a starts in slot 1, top and ends in slot 44 top. Phase B starts at slot 5 top, and ends at slot 48 top, phase C starts at slot 9 top, and ends at slot 4, top. Phase A for the control (4 pole) winding goes as follows: (A11 , A11 ), (A12 , A12 ), (A13 , A13 ), (A14 , A14 ), (A21 , A21 ), (A22 , A22 ), (A23 , A23 ), (A24 , A24 ),
Thus phase A starts in slot 1, top and ends in slot 40 top. Phase B starts at slot 9, top and nishes in slot 48, top, phase C starts at slot 17, top, and ends at slot 8, top. Stator combination matrices With reference to gures B.2 and B.1, the coil group order for the mutual inductance matrices is as follows:
(A31 , A31 ), (A32 , A32 ), (A33 , A33 ), (A34 , A34 ), (A41 , A41 ), (A42 , A42 ), (A43 , A43 ), (A44 , A44 )
234
slot 44 top slot 45 top slot 46 top slot 47 top slot 48 top 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 slot 1 bottom slot 2 bottom slot 3 bottom slot 4 bottom slot 5 bottom
8 Pole winding (to be wound closest to the airgap) 20 turns per coil
U out
W return V out
W out
U return
V return
Figure B.1: 8 pole winding diagram
235
4 Pole winding (to be wound furthest from the airgap) 10 turns per coil
slot 41 top slot 42 top slot 43 top slot 44 top slot 45 top slot 46 top slot 47 top slot 48 top 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 slot 1 bottom slot 2 bottom slot 3 bottom slot 4 bottom slot 5 bottom slot 6 bottom slot 7 bottom slot 8 bottom slot 9 bottom slot 10 bottom U out
W return V out
W out
U return
V return
Figure B.2: 4 pole winding diagram
236
Element 17 32 1 16
Description 4 pole winding red phase 4 pole winding yellow phase 4 pole winding blue phase 8 pole winding red phase 8 pole winding yellow phase 8 pole winding blue phase
33 48
49 64
65 80
81 96
where:
The combination matrix to combine individual coil inductances into winding inductances is given by: Q4 0 0 0 0 0 0 Q4 0 0 0 0 0 0 Q4 0 0 0 (B.8) Ts = 0 0 Q8 0 0 0 0 0 0 0 Q8 0 0 0 0 0 0 Q8
Q 4 = Q a Q a Q 8 = Q b Q b
Q a Q a , Q b Q b Q b Q b Q b Q b ,
Qa = 1 1 1 1 Qb = 1 1
The resulting mutual inductance matrix, leakage inductance matrix as calculated by the methods described in appendix C.1, and excess harmonic inductance is given by: 3.5 210 96.3 96.3 0 0 0 Mss = 103
hl Mss = 103 96.3 210 96.3 0 0 0 0 0 0 96.3 96.3 210 217 97.4 97.4 0 0 0 0 0 0 97.4 217 97.4 0 97.4 217 6.60 5.6 5.6 0 0 0 97.4 0 5.6 6.6 5.6 0 0 0 5.6 5.6 6.6 0 0 0 0 0 0 10.0 6.1 6.1 0 0 0 6.1 10.0 6.1 0 0 0 6.1 6.1 10.0
Mssl = 103
0 0 0 0 0 0 3.5 0 0 0 0 0 0 3.5 0 0 0 0 0 0 8.0 0 0 0 0 0 0 8.0 0 0 0 0 0 0 8.0
(B.9)
(B.10)
Transforming these using the d-q transformation matrices, (3.1) and (3.2) (the leakage inductance matrix is unchanged as it is diagonal): 307 0 0 0 0 Mdq0s = 103
0 0 307 0 0 0 0 0 0 17.8 0 0 0 0 0 0 314 0 0 0 0 0 0 314 0 0 0 0 0 0 22.2
hl Mdq0s = 103
1.04
0 0 0 0 0
0 0 0 0 0 1.04 0 0 0 0 0 17.8 0 0 0 0 0 3.88 0 0 0 0 0 3.88 0 0 0 0 0 22.2
(B.11)
Stator winding arrangement for parallel coil groups With reference to the coil pitch table in section B.2.2: Thus for phase a of the 8 pole winding the connections are as follows: A: (a11 , a11 ), (a12 , a12 ), (a21 , a21 ), (a22 , a22 )
B.3 Rotor 1: Nested-loop Rotor Design Details
237
where A,B,C,D are the 4 sets of coil groups to be connected in parallel. Phase A for the control 4 pole winding goes as follows: A: (A11 , A11 ), (A12 , A12 ), (A13 , A13 ), (A14 , A14 ) B: (A21 , A21 ), (A22 , A22 ), (A23 , A23 ), (A24 , A24 ), C: (A31 , A31 ), (A32 , A32 ), (A33 , A33 ), (A34 , A34 )
D: (a71 , a71 ), (a72 , a72 ), (a81 , a81 ), (a82 , a82 )
C: (a51 , a51 ), (a52 , a52 ), (a61 , a61 ), (a62 , a62 )
B: (a31 , a31 ), (a32 , a32 ), (a41 , a41 ), (a42 , a42 )
D:(A41 , A41 ), (A42 , A42 ), (A43 , A43 ), (A44 , A44 )
where A,B,C,D are the 4 sets of coil groups to be connected in parallel.
B.3 Rotor 1: Nested-loop Rotor Design Details

The state order used for rotor 1 is: 6 inner rotor loops, 6 middle rotor loops, 6 outer rotor loops. This order is preserved when transformed to dq axes.
B.3.1
Rotor-rotor inductance terms
Figure B.3 shows the arrangement of the rotor bars in the prototype nested-loop rotor. This arrangement leads to a rotor-rotor mutual inductance matrix given by (B.12), calculated using the method described in section 2.7.1: Mrr = 108 14 14 14 14 14 468 43 43 43 43 43 439 71 71 71 71
14
465 14 14 14 14 14 465 14 14 14 14 465 14 14 14 465 14 14 465 14
468 43 43
43 468 43
43 468 43
43
14 14 14 14
14 14 14
14 14
465
43 43 43 43 43
43 43
43
468 43 43 43 43 128
1372 128 128 128 128 128
43
43
43
43
43
43
43 468 43
43
43
43
43
43 468 43
43
43
43
43
43 468
43
43
43
43
439 71 71
71 439 71
71 439 71
71
468 43 43 43 128 128
1372 128 128 128 128 213
1318 213 213 213 213 1318 213 213 213 1318 213 213
71
71
71
71
71
71
71 439 71
71
71
71
71
71 439 71
71
71
71
71
43 43 43
468 43 43 128 128 128
1372 128 128 128 213 213
43 43 43 43
468 43 128 128 128 128
1372 128 128 213 213 213
71 71 71 71 71
468 128 128 128 128 128
1372 128 213 213 213 213
1318 213 1318
71 71
71
439 71 71 71 71 213
1318 213 213 213 213 213
1372 213 213 213 213 213 2165 354 354 354 354 2165 354 354 354 2165 354 354 2165 354
439 71 71 71 213 213
1318 213 213 213 213 354
71 71 71 71 71
71 71 71
439 71 71 213 213 213
1318 213 213 213 354 354
439 71 213 213 213 213
1318 213 213 354 354 354
439 213 213 213 213 213
1318 213 354 354 354 354
2165
1318 354 354 354 354 354
71 71 439 213 213 213 213 213 1318 354 354 354 354 354
71 2165
71
71
(B.12)
The leakage inductances as calculated using the methods described in appendix C.1 are 9.7753 107 , 1.0278 106 , and 1.0781 106 for the inner, middle and outer loops respectively. As
238
explained in appendix C.1 it has been assumed that the leakage inductances only affect the selfinductance terms, hence the full Mrr matrix has, additionally, the appropriate leakage terms added on the diagonal. When converted to d-q axes and the observable/uncontrollable parts removed, as discussed in section 3.3, the rotor-rotor mutual inductance matrix becomes: 479 0 0 510 0 0 510 0 0
0 479 0 0 0 0 510 0 0 510 0 0
Mrrdq
0 510 8 0 = 10 0 510
0 0 0
394 0 0
255 0 0
1499 0 0
1531 0 0
510 0 0 510 0
1499 0 0 1531 0
1531 0 0 2520 0
255 0 0 85
734 0 0 255
1531 0 0
2520 0 0
394
0 0 0 l 8 0 0 Mrrdq = 10 255 0 0 0
85 0 0
98
0
0 98 0 0 0 0 0 0 0
0 0 98 0 0 0 0 0 0
0 0 0 103 0 0 0 0 0
0 0 0 0 103 0 0 0 0
0 0 0 0 0 103 0 0 0
0 0 0 0 0 0 108 0 0
0 0 0 0 0 0 0 108 0
0 0 0 0
h2 Mrrdq
0 242 8 0 = 10 0 371
0 0
84
0 84 0 0 242 0 0 371 0
0 0 0 0 0 0 0 0 0
242 0 0 698 0 0 1069 0 0
0 242 0 0 698 0 0 1069 0
0 0 0 0 0 0 0 0 0
371 0 0 1069 0 0 1638 0 0
0 0 0 1069 0 0 0
371 0
1638 0
0 0 0 0 0
0
h4 Mrrdq
0 206 8 0 = 10 0 235
0 0
82
0
0 82 0 0 206 0 0 235 0
0 0 0 0 0 0 0 0 0
206 0 0 523 0 0 594 0 0
0 0 0 0 0 594 0
0 0 0 0 0 0 0
206 0
594 0
(B.13) 235 0 0
0 235 0 0 0 0 0 594 0 0 0 0
108
0 0 0
0 0
523 0
676 0 0
676 0
(B.14)
0 0 0
0
l where Mrrdq is the full d-q rotor-rotor mutual inductance matrix excluding leakage effects, M rrdq is h2 h4 the leakage inductance matrix, Mrrdq and Mrrdq are the 4 pole and 8 pole harmonic components (the
rst two non-zero components) of the rotor-rotor mutual inductance matrix. This d-q model may be reduced in order, after removal of the zero-sequence terms, as described in section 3.4.2. The matrix used to reduce the order of the rotor states for this machine is: T = 0 0.2108 0.2108 0 0 0.5768 0.5768 0 0 0.7892 0.7892 0 (B.15)
Applying (B.15) to (B.13), the resulting rotor-rotor mutual inductance terms become: Mrrdq = 108
h2 Mrrdq = 108
3777 0 2411 0
0 3777 0 2411
l Mrrdq = 108 h4 Mrrdq = 108
106 0 0
0 106 0 1268
(B.16) (B.17)
1268
l where, Mrrdq is the reduced order d-q rotor-rotor mutual inductance matrix, Mrrdq is the reduced order h4 h2 leakage inductance matrix, Mrrdq and Mrrdq are the 4 pole and 8 pole harmonic components (the rst
two non-zero components) of the rotor-rotor mutual inductance matrix. Subtracting equations (B.17) from Mrrdq gives the harmonic leakage (or differential leakage) inductance in the rotor, that is the additional series inductance that appears due to the production of unwanted space harmonics by the rotor:
hl Mrrdq = 108
98 0
0 98
(B.18)
B.4 Rotor 2: New Double Layer Rotor Design Details
239
B.3.2
Rotor-Stator inductance details
Figure B.4 shows the rotor-stator mutual inductance for a inner, middle, and outer loop between both 4 and 8 pole stator phases. The data is overlaid with the fundamental component in each case, calculated from the FFT. Equation (B.19) shows the mutual inductance between stator and rotor for the reduced order dq model. 2408 1253 = 106 345 1954 1253
Mdqsr
2408 1954 345
(B.19)

B.4.1 Rotor-rotor inductance terms
Figure B.5 shows the arrangement of the rotor bars in the prototype double-layer winding rotor. This arrangement leads to a rotor-rotor mutual inductance matrix given by (B.20), calculated using the method described in section 2.7.1: 312 83 31 31 312 31 83 312 83 83 31 (B.20)
6 Mrr = 10
31
31
83
83
83
83
83
83
83
31
31 312
83
83
31
31 312
83
83
31
83 83 83 31 312
The leakage inductance for each rotor phase as calculated using the methods described in appendix C.1 are 8.8 106 . As explained in appendix C.1 it has been assumed that the leakage inductances leakage terms added on the diagonal. When converted to d-q axes, as described in section 3.3, rotor-rotor mutual inductance matrix only affect the self-inductance terms, hence the full Mrr matrix has, additionally, the appropriate
240
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
Figure B.3: Rotor 1: Prototype nested-loop design rotor
241
1.5
x 10
Inner loop Middle loop Outer loop
1.5
x 10
Inner loop Middle loop Outer loop
Mutual Inductance (H)
Mutual Inductance (H) 60 120 180 240 Rotor angle (degrees) 300 360
0.5
0.5
0.5
0.5
1.5 0
1.5 0
60
120
180 240 Rotor angle (degrees)
300
360
(a) Stator 1 - Rotor Mutual Inductance
(b) Stator 2 - Rotor Mutual Inductance
Figure B.4: Stator-Rotor mutual inductance for Rotor 1, the prototype nested loop rotor for the individual loops of a particular nest. The dotted lines show the fundamental component of the elds. becomes: Mrrdq = 106
h2 Mrrdq
342 0 0 293 0 0
0 342 0 0 293 0
= 10
0 0 0 0 0
l Mrrdq = 106 0 0
h4 Mrrdq
0 9 0 0 44 0
= 10
44 0 0
0 9 0 0 0
(B.21)
(B.22)
rst two non-zero components) of the rotor-rotor mutual inductance matrix. Subtracting equations (B.22) from Mrrdq gives the harmonic leakage (or differential leakage) inductance in the rotor, that is the additional series inductance that appears due to the production of unwanted space harmonics by the rotor: hl Mrrdq = 106 0 0 5 0 5 0 0 0 0
(B.23)
B.4.2
Figure B.6 shows the rotor-stator mutual inductance between a rotor phase and a 4 and 8 pole stator phase. The data is overlaid with the fundamental component in each case, calculated from the FFT.
242
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 slot 1 bottom slot 2 bottom slot 3 bottom slot 4 bottom

Figure B.5: Prototype rotor 2: double-layer design rotor
B.5 Rotor 3: Isolated loop rotor design
243
Equation (B.24) shows the mutual inductance between stator and rotor for the dq model.
x 10
2 1.5
x 10
1 0.5 0 0.5 1 1.5 2 0 60 120 180 240 Rotor angle (degrees) Actual Fundamental 300 360
5 0
Actual Fundamental 60 120 180 240 Rotor angle (degrees) 300 360
Figure B.6: Stator-Rotor mutual inductance for Rotor 2, the prototype double-layer rotor. The dotted lines show the fundamental component of the elds.
Mdqsr
6 = 10
9119 0 3690
2875
9119 0
2875
651 3690 0
651 0
0 0 0 0 0
(B.24)
B.5 Rotor 3: Isolated loop rotor design

The state order used for rotor 3 is: 6 loops spaced by 60 , then another 6 loops again spaced by 60 and offset from the previous set by 20 , then a third set of 6 loops again evenly spaced by 60 and offset from the previous set by 20 . The 3 sets are colour-coded on the winding diagram, gure B.7.
B.5.1
Rotor inductance terms
Figure B.7 shows the arrangement of the rotor bars in the prototype isolated loop rotor. This arrangement leads to a rotor-rotor mutual inductance matrix given by (B.25), calculated using the method
244
described in section 2.7.1: Mrr = 106

29
29
7 29
7 19
7 8
2 2
7 2 2
19 8 7 7
29
7 29
7 19 8 7 7
7 19 8
8 7 7
8 19 7 7
7 7 19 7 7 7
7 29
7 8
7 19 7 7 7
7 8
19
7 19
19 7 7
19 7 7 7 7 8 7 7
7 19
7 8
7 2
7 29
7 8
7 19
2 2
7 29
7 2 2
7 7 29
7 19 8 7 7
7 19
7 19
7 8
7 19
8 7 7
19
7 19 8
7 8
19
7 8
7 7 19
2 2
7 29
7 19 8 7 7 7 7
7 7
7 19
7 7 8 8 7 7 7
2 2
7 29
19 8
7 8
19
7 19 8
19
19
7 19 8
7 19
7 29
7 2 2
7 7 29
7 19 8 7 7
19 8
7 19
29
7 19
7 8
7 19
2 2
7 29
7 19 2
2 2
7 29
7 7
2 2
7 29
7 29
(B.25)
The rotor loop leakage inductance as calculated using the methods described in appendix C.1 is 3.3055 106 H. As explained in appendix C.1 it has been assumed that the leakage inductances only terms added on the diagonal. Transformation to d-q axes is achieved by the method described in section 3.3. This rotor design can be considered as three sets of 6 evenly spaced coils, and thus the method of section 3.3. The resulting mutual inductance matrix (with zero-sequence state removed for brevity) is:
315 103 0 315 103 0 0 0 328 103 0 0 0 0 328 103 736 2784 122 617 736 34 0 18 13 3 22 122 2784 736 0 2211 832 480 18 1696 480 603 2720 328 3 2720
affect the self-inductance terms, hence the full Mrr matrix has, additionally, the appropriate leakage
Mdq
6 = 10
1696
0 0 2784 122 2211
2211 832
0 2784 122
617 736 832 480 328
617 34 13 18 22
13 0
902 22
902 22
603 328 3 13 18 0 34
1696
1696
2720
603
2211 2720 603
480
617
13 34 0 18 13
18 34 13 18
18
832
902
902
13 34 0
328
(B.26)
where the leakage inductance is included, and is unaffected by the dq transformation. The rotor-rotor portion of the mutual inductance matrix may be diagonalised by an appropriate state transformation, the eigenvectors of the rotor-rotor mutual matrix. Applying this transformation
B.6 Rotor 4: Isolated loop design rotor with one set of loops removed
245
to the whole mutual inductance matrix (using the identity matrix for the stator potion) gives: Mdq = 106
315 103 0 0 0 211 0 4822 0 0 0 0 0 315 103 0 0 0 0 211 4822 0 0 0 0 0 4822 0 0 0 0 211 328 103 0 4822 0 79 0 0 0 0 0 211 0 0 79 0 0 0 0 0 0 0 0 0 1140 0 0 1212 328 103 1212 0 0 0 1140 1212 0 0 1140 0 14 0 0 0 1140 0 0 0 14 0 0 1212
0 0 0 0 0 0 0 0 0 9
0 0
0 0 0 0 9 0
(B.27)
As can be seem from (B.27) this rotor decouples the different stator supplies. There is no longer any cross-coupling achieved by the rotor, as is expected.
B.6 Rotor 4: Isolated loop design rotor with one set of loops removed
This rotor is identical rotor 3 but with one set of 6 loops removed. Figure B.7 shows rotor 3. Rotor 4 is the same, but with the loops marked in blue removed.
B.6.1
Rotor inductance terms
As rotor is a subset of rotor 3 so are the mutual inductance terms. The rotor-rotor inductance is given by the upper-left 12 12 partition of equation (B.25). The leakage inductance per coil is the same, at 3.3055 106 H. The d-q transformed machine mutual inductance is the upper-left 8 8 partition of (B.26). How-
ever when diagonalizing the d-q transformed model the results differ signicantly from those obtained for rotor 3: 315 103 0 315 103 0 0 0 328 103 0 0 0 0 328 103 103 167 3939 55 706 57 0 0 0 103 55 3939 103 0 117 1143 0 0 12 0 2
Mdq
= 106
0 0 3939 55 117 2
0 55 3939 117 2
706
706 57 0 0
167
1143
103
706
167
1143
(B.28)
117 167 1143 0 0 0 12
Thus equation (B.28) demonstrates that unlike rotor 3 where the 4 and 8 pole stator currents were decoupled from each other, there is cross-coupling in the case of rotor 4.
246
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 slot 2 bottom slot 4 bottom slot 6 bottom

Figure B.7: Prototype rotor 3: Isolated loop rotor design
B.7 Rotor 5: 6 bar cage rotor design
247
B.7 Rotor 5: 6 bar cage rotor design

A cage rotor may be analysed by mesh loop analysis. As such the for an n bar cage there will be n such loops to fully describe the cage (assuming that no ux is linked though the end ring). Figure B.8 shows an example cage, with two loops illustrated, with currents I 1 and I2 owing respectively. Having dened the rotor circuits as such the bar current may be expressed in terms of the loop currents as illustrated in gure B.8. For the purposes of mutual inductance calculation an n bar cage is therefore identical to n isolated loops replacing the virtual loops forms from the analysis. Therefore the mutual inductance calculation is the same as for the other rotor designs. However the resistance and leakage inductance matrices will be of the following form: 2Rbar + 2Rend Rbar 0 0 Rbar Rbar 2Rbar + 2Rend Rbar 0 0 . .. .. .. .. . . . . . 0 . R= . .. .. .. .. . . . . . . 0 0 0 Rbar 2Rbar + 2Rend Rbar Rbar 0 0 Rbar 2Rbar + 2Rend
PSfrag replacements I1 I1 In I1 I1 I2 I2 I1 I2 I2 I2 I3
where Rbar and Rend are the resistance (or inductance) due to the bar and end ring respectively.
Figure B.8: Mesh loop analysis of cage rotors The rotor-rotor mutual inductance may therefore be calculated by the method described in section 2.7.1: Mrr = 107 271 51 56 51 271 51 56 271 56 56 51 56 56 56 51 271 (B.29)
51 51
56 271
56
56
56
51 51
56
56 271
51
56
56
56
51 51
56
56
248
The leakage inductance for each bar and end winding was calculated using the methods described in Therefore the leakage inductance matrix is: appendix C.1. The bar leakage was found to be 2.40 10 7 H and the end ring leakage 1.02 107 H.
2 0 l 7 Mrr = 10 0 0 2 ductance matrix becomes: 9 0 9 0
0 2 2 0 0 7
0 0 2 2 0 7
0 0 0 2 7
2 0 0 0
0 0 0 2 7
(B.30)
When transformed to d-q axes using the method described in section 3.3, the rotor-rotor mutual in-
l Mrrdq = 107 0
h2 Mrrdq 107
0 0 0 3 230 0 0 0 230 0 0 0 0
Mrrdq 107
h4 Mrrdq 107
322 0 0
0 322 0 0 57 0
57 0 0
0 0 0 0 0
(B.31)
(B.32)
rst two non-zero components) of the rotor-rotor mutual inductance matrix. Subtracting equations (B.32) from Mrrdq gives the harmonic leakage (or differential leakage) inductance in the rotor, that is the additional series inductance that appears due to the production of unwanted space harmonics by the rotor: 35 0 0 35 0 0
hl Mrrdq = 107 0
0 0
(B.33)
B.7.2
Figure B.9 shows the rotor-stator mutual inductance between a rotor phase and a 4 and 8 pole stator phase. The data is overlaid with the fundamental component in each case, calculated from the FFT. Equation (B.24) shows the mutual inductance between stator and rotor for the dq model.
B.8 Rotor 6: Wound Rotor Design Details
249
1.5 1 Mutual Inductance (H) 0.5 0 0.5 1
x 10
8 6 Mutual Inductance (H) 4 2 0 2 4 6
x 10
1.5 0
8 0
Figure B.9: Stator-Rotor mutual inductance for Rotor 5, the prototype double-layer rotor. The dotted lines show the fundamental component of the elds. 2660 839 0 243 0 0 (B.34)
Mdqsr
2660 0 6 6 = 10 10 1378 243 0
839
1378
0 0 0 0 0
B.8 Rotor 6: Wound Rotor Design Details

Figures B.1 and B.2 show the winding diagram for the stator winding of the prototype machine. Rotor 6 has the same windings as the stator winding connected together in a star short-circuit including a neutral connection. The connection is achieved using the following combination matrix: 1 0 0 0 0 1
Tc = 0 1 0 0 1 0 0 1 1 0
0 0
(B.35)
Therefore the rotor-rotor mutual inductance matrix given by (B.36), calculated using the method
250
described in section 2.7.1: 400 181 181 400 181 181
Mrr = 1107 181
181 400
(B.36)
The leakage inductance is calculated in exactly the same was as for the stator using the methods described in appendix C.1. The leakage inductance matrix is given by: 10 0 0 10 0 0
l Mrr = 107 0 becomes: 10 0 0 10 0
0 10
(B.37)
When converted to d-q axes, as described in section 3.3, rotor-rotor mutual inductance matrix
Mrrdq = 107 0
h2 Mrrdq
= 10
286 0 0
0 0 10 0 0 286 0 0 0
l Mrrdq = 107
h4 Mrrdq
581 0 0 290 0 0
0 581
= 10
0 0 37 0 0 290 0 0 0
(B.38)
(B.39)
rst two non-zero components) of the rotor-rotor mutual inductance matrix. Subtracting equations (B.39) from Mrrdq gives the harmonic leakage (or differential leakage) inductance in the rotor, that is the additional series inductance that appears due to the production of unwanted space harmonics by the rotor: 5 0 0 5 0 0
hl Mrrdq = 107 0
0 37
(B.40)
B.8.2
Figure B.10 shows the rotor-stator mutual inductance between a rotor phase and a 4 and 8 pole stator phase. The data is overlaid with the fundamental component in each case, calculated from the FFT. Equation (B.41) shows the mutual inductance between stator and rotor for the dq model.
B.9 Rotor 7: Standard Squirrel Cage Rotor Details
251
x 10 2 1.5 Mutual Inductance (H)
x 10 2 1.5 Mutual Inductance (H) 1 0.5 0 0.5 1 1.5 2 0
1 0.5 0 0.5 1 1.5 2 0 60 120 180 240 Rotor angle (degrees) Actual Fundamental 300 360
Figure B.10: Stator-Rotor mutual inductance for Rotor 6, the prototype double-layer rotor. The dotted lines show the fundamental component of the elds.
Mdqsr
1639 0 6 = 10 1866 2224 0
2341
1639
2341 2224
1866
0 0 0 0 0
(B.41)
B.9 Rotor 7: Standard Squirrel Cage Rotor Details
B.9.1
Rotor-rotor inductance terms
The rotor is a standard 40 bar squirrel cage rotor with Boucherot slots to give a higher rotor resistance at high slips though the skin effect. The calculation procedure is identical to that for Rotor 5 described in section B.7. Using this
252
The rotor bar and end ring leakage inductance is calculated using the methods described in appendix parameters are added to the rotor-rotor mutual inductance as described in section B.7 C.1. The bar leakage is 1.25 106 H and the end ring leakage is 1.34 106 H. These inductance Transformation to d-q axes is achieved by the method described in section 3.3. However as the cage rotor has 40 bars it cannot directly be considered to be part of the class of rotor designs. Therefore a different d-q transformation was used: Cdq = where: 2 sin(0) 40
1 2
method the rotor-rotor mutual inductance matrix is: 471 14 13 13 13 14 14 471 14 13 13 13 13 14 471 14 13 13 .. . .. .. .. .. .. . . . . . . . . Mrr = 108 13 13 14 471 14 13 13 13 13 14 471 14 14 13 13 13 14 471
(B.42)
Cdq p1 0
0 Cdq p2
(B.43)
Cdq p =
p p p cos(0) cos( 240 ) cos( 440 ) cos( 640 ) cos( 78 p ) 40 p sin( 240 ) 1 2 p sin( 440 ) 1 2 1 2
and p1 = 2, p2 = 4 for the prototype stator winding. states.
p sin( 640 ) sin( 78 p ) 40

1 2
(B.44)
It can be shown that this non-square transformation matrix truncates only unobservable/uncontrollable
The resulting mutual inductance matrix (with zero-sequence state removed for brevity) is: 363645 0 0 0 574 1260 0 0 0 363645 0 0 1260 574 0 0 0 0 377345 0 0 0 1374 96 0 0 0 377345 0 0 96 1374 Mdq = 106 574 1260 0 0 6 0 0 0 1260 574 0 0 0 6 0 0 0 0 1374 96 0 0 6 0 0 0 96 1374 0 0 0
(B.45)
B.10 Machine slot utilisation
253
where the leakage inductance is included. The other rotor-rotor inductance parameters in d-q axes are given by: 54 0 54 0 0 0 53 0 0 0 0 0 4 0 4 0 0 0 0 0 0 0 0 11 0 0 0 52 0 0
Mrrdq
h2 Mrrdq
0 = 107 0 0 53 0 = 107 0 0
0 53 0 0 53 0 0 0 0 0 0 0 0
l Mrrdq
h4 Mrrdq
0 = 107 0 0 0 0 = 107 0 0
0 0 11 0 0 0 52
(B.46)
(B.47)
rst two non-zero components) of the rotor-rotor mutual inductance matrix. Subtracting equations (B.47) from Mrrdq gives the harmonic leakage (or differential leakage) inductance in the rotor, that is the additional series inductance that appears due to the production of unwanted space harmonics by the rotor: 2 0 2 0 0 0 0 9 0 0
hl Mrrdq
= 10
0 0 0
0 0 9
(B.48)

Table B.4 shows the slot utilisation of the machine for different rotors.
254
Conductor area (mm2 ) Stator 67.86
Conductor metal
Slot area (mm2 )
Fill factor
Cu
135.47
0.500
Rotor 1: Nested-loop design 133 Cu 147.23 0.903
Rotor 2: New double layer design* 51.46 Rotors 3 and 4* 51.46 Cu 70 0.74 Cu 70 0.74
Rotor 5: 6 bar squirrel cage design (simulation only) 205.84 Cu 280 0.74
Rotor 6: wound rotor (simulation only) 135.72 Cu 135.47 1.00
Rotor 7: standard squirrel cage rotor 116.60 Al 116.60 1.00
Table B.4: Slot utilisation of the stator and rotors for the prototype machine. (* values for slots which contain conductors only)
255
M530-65A 0.65mm thick 5.3W/kg loss at 1.5T B (T) 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 H (A/m) 44 59.5 69.6 78.2 86.6 95 104 113 125 138 159 196 270 454 1040 2630 5620 10100
Table B.5: Magnetisation data for stator laminations for prototype BDFM
256
Appendix C
Leakage Inductance and Effective Air Gap

In this chapter the procedure used to estimate the leakage inductance are described, and the notion of effective air gap reviewed and the chosen method of estimation described. The symbols used in this chapter are similar to those used in gures 2.2 and 2.1, however in this chapter all dimension are in metres unless otherwise specied, whereas in gure 2.2 all measurements other than the air gap width were in radians. In this chapter is used to denote permeance, y s is used ws as depicted in gure 2.2). to denote slot pitch in metres (hence ys s d , and wo ws d where d is the diameter, and s and 2 2
C.1 Leakage Inductance

The calculation of leakage inductance is greatly simplied by the calculation of specic permeance. Permeance is the reciprocal of reluctance. Specic permeance is dened as the self-inductance per unit length of coil per turns squared. Thus to calculate the self inductance of a coil due to leakage the permeances around the length of the coil should be summed and the the result multiplied by the number of turns squared. The sources of permeance will now be considered. For full details, and derivation of results see [31], and [18] for trapezoidal slot geometry (and simplications adopted here). An estimation of the slot bridge permeance for the cage rotor considered is taken from [9].
C.1.1
Slot and Tooth-top Permeance
Figure C.1 shows a trapezoidal slot. The dotted boxes within the slot represent different conductor positions that will be considered. A square slot, can of course, be considered as a special case of a trapezoidal slot. It will be assumed that slot leakage can be adequately modelled by trapezoidal slot
257
PSfrag replacements
258 Leakage Inductance and Effective Air Gap
ys g h6 h4 h3 w3 h2 w2 h1 w1 Figure C.1: Slot shape used for permeance calculations geometry even if the slot it not exactly trapezoidal (such as having a semi-circular end). For the conductors in the bottom position in gure C.1 the slot permeance is: 2h 1 2(h 2 + h 3 + h 4 ) 2h 5 h6 g + + + + 3(w1 + w2 ) (w5 + w2 ) (w5 + wo ) wo ys wo w5 w4 h5
s1 = 0
(C.1)
The rst term of C.1 accounts for the permeance inside the conductor being considered, with the average slot width being taken. The second to fourth terms account for the permeance in the region above the conductor and the nal term is the tooth-top permeance. For other conductors the rst and second terms are adjusted accordingly: 2(h 3 + h 4 ) 2h 5 h6 g 2h 2 + + + + 3(w2 + w3 ) (w5 + w3 ) (w5 + wo ) wo ys 2h 4 2h 5 h6 g 2h 3 + + + + 3(w3 + w4 ) (w5 + w4 ) (w5 + wo ) wo ys 2h 4 2h 5 h6 g + + + 3(w4 + w5 ) (w5 + wo ) wo ys
s2 = 0 s3 = 0 s4 = 0
(C.2) (C.3) (C.4)
C.1.2
Overhang Permeance
Overhang permeance accounts for coupling arising from the end windings which overhang the lamination stack. Figure C.2 shows a diagram of an end winding. Under the assumption that the overhang forms two sides of a square at each end coil (see gure C.2) [31] gives an empirical formula: o = 0 0.366 loge 2a b + 0.11 (C.5)
259
Rotor Laminations
Overhang
a PSfrag replacements L o = 2a + 2b
b Figure C.2: Assumed end winding conguration for overhang permeance calculations which for typical values of a and b gives: 0 . It is assumed that there is zero mutual coupling between adjacent coils (if this is not the case [31] gives suitable modications to (C.6). (C.6)
C.1.3
Zig-zag Permeance
Zig-zag permeance accounts for coupling where the ux density path zig-zags between teeth on opposing sides of the air gap. It is only applicable when both rotor and stator have open slots. Assuming that the slot pitch can be assumed to be average of the rotor and stator pitch, and that that half the zig-zag ux links the stator and half the rotor, [31] gives the permeance as:
ys wo + wor z = 0 a 1 12g ysa where gure C.3 explains the symbols.
(C.7)
C.1.4
Leakage ux per coil
The leakage ux per coil can then be calculated from the specic permeances previous stated. The permeances described in sections C.1.1 and C.1.3 both act over the lamination stack length, ws , whereas the overhang permeance acts over the overhang length, L o . A description of the calculation is given for each rotor in described in chapter 5.
260
ysa PSfrag replacements g wo ysa Figure C.3: Assumed slot conguration for calculation of zig-zag permeance Stator and rotor 6 The machine being considered has two two layer windings on the stator for the power and control winding supplies. Therefore the stator leakage for the winding closest to the air gap (in our case the 8 pole winding) is as follows:
2 L coil8 = t8 (ws (s3 + s4 + 2z ) + 2L o o )
wor
(C.8)
For the 4 pole winding:

2 L coil4 = t4 (ws (s1 + s2 + 2z ) + 2L o o )
(C.9)
Rotor 6, which is considered in simulation only, is assumed to have the same slot shape and winding arrangement as the stator. Rotor 1: nested-loop design, and rotor 5: squirrel cage rotor with 6 bars On both rotors1 and 5 the rotor bars take up the entire slot therefore the slot permeance is:
sn = 0
h6 2h 5n g 2(h 1n + h 2n + h 3n + h 4n ) + n + + 3(w1n + w5n ) w5 n + w o n wo n ysn
(C.10)
However the zig-zag permeance and overhang permeance expressions are the same. Therefore the leakage inductance is:
L nested = 2ws (sn + z ) + L on o
(C.11)
ignoring any leakage due to the end ring at one end, and noting that L on is dependent on the location of the particular loop within a nest.
261
ysa g wo h2
h1 PSfrag replacements
w1
h2
h1
Figure C.4: Rotors 2, 3 and 4 slot detail Rotor 2: new double layer design The rotor has a dual layer winding and the slot shape is unusual (and non-ideal), due to economic reasons which would not affect a commercial machine, a diagram of the slot is shown in gure C.4. Rotor 2 only uses the top slot shown in gure C.4, and each phase consists of 4 loops. Each of the loops comprises of a conductor taking half the area of the top slot; the out and return paths of the loop are in the top and bottom portion of the slot respectively. Therefore: h2 g h 1 /2 h 1 /2 + 2z ) + 2L o o ) + +2 +2 3w1 w1 wo ys
L phase = 4(ws (0 2
(C.12)
Rotor 3: A rotor design consisting of 18 isolated loops Rotor 3 has 18 separate loops comprised of copper strip which takes the up the whole slot area of the top then bottom slot of gure C.4 respectively. Rotor 4 is identical in construction to rotor 3, but with fewer loops, hence the inductance per loop is unchanged. The leakage inductance of a single loop is therefore: h1 h2 h2 g h1 + + +2 +2 + 2z ) + 2L o o 3w1 w1 wo wo ys
L loop = ws (0 2
(C.13)
Rotor 7: Standard squirrel cage design Rotor 7 has closed slots, and therefore a different approach is required to calculate the permeance. Birch and Butler give a computational method for estimating the slot permeance as a function of the
262
rotor m.m.f. (magneto-motive force) [9]. However the slot geometries of the cage rotor considered are close enough to the example given in the paper to take the results straight from gure 3 of [9]. The prototype rotor slot bridge depth on the slot line is 0.85mm, the slot top radius is 2.5mm and the air gap is 0.55mm. 7.5 3.5 2.8 2.5 m.m.f. (A-turns) 100 200 300 400
Thus assuming around 300 A-turns then the specic permeance is 2.8. This permeance must be added to the slot permeance to determine the leakage inductance for a single bar: L bar = 2.80 ws + ws (0 g h 1 /2 + + z ) 3w1 ys
The leakage due to the end ring is approximated using the same method as for an end winding: L end = L o o Therefore the total leakage inductance matrix will be a symmetric circulant matrix with 2L bar + 2L end on the diagonal, and L bar offset by 1 column from the diagonal. the progressive loop rotor. Parameter Values for Leakage calculation Table C.1 shows the slot and overhang dimensions for leakage ux calculation on the stator rotor designs considered. Table C.2 contains physical geometries of the considered rotors and stator, and the effective air gap as calculated using Carters factor. The dimensions for the slot leakage calculations have been assumed to be the same as those for
C.2 Effective air gap

In 1900 Carter [21] rst derived a correction factor for the air gap to allow for the effect of having an open slot. The initial result was expanded and given in a more complete form in [22], equation (8). Broadly the assumption made is that slots are parallel and innitely deep, slot pitch length is innite. Furthermore it is assumed that the teeth themselves are made of material of innite magnetic permeability. Carters factor is given by: K = ys ys g (C.14)
wo (mm)
w1 (mm)
w2 (mm)
w3 (mm)
w4 (mm)
w5 (mm)
g (mm)
h1 (mm)
h2 (mm)
h3 (mm)
h4 (mm)
h5 (mm)
h6 (mm)
ys (mm)
w (mm)
Lo (mm)
Stator (and rotor 6 with t = 0.1 and w = 189.9) 4 pole (furthest from the air gap) 3.2 8.111 7.731 7.335 6.471 5.473 2.845 2.992 6.532 7.556 0.656 1.13 11.457 195.5 160 10
Stator (and rotor 6 with t = 0.2 and w = 189.9) 8 pole (closest to the air gap) 3.2 8.111 7.731 7.335 6.471 5.473 2.845 2.992 6.532 7.556 0.656 1.13 11.457 195.5 75 20
Rotor 1: Nested-loop design 2.8 5.68 9.433 0.547 19.48 0 0 0 1.207 0.65 15.182 189.9 100,60,20 1
Rotor 2: New double layer design 2.1 3.74 0.555 18.71 2.27 15.180 190.2 110 1
Rotors 3 and 4: isolated loop designs 2.1 3.74 0.555 18.71 2.27 15.180 189.9 140 1
Rotor 5: 6 bar squirrel cage design 8.4 22.72 37.72 0.547 19.48 0 0 0 1.207 0.65 91.091 189.9 110 1
Table C.1: Slot and overhang dimensions for calculation of leakage ux for the different rotors, see gures C.4, C.1 and C.2 for details. the value used in calculation of zig-zag permeance was 15.182mm, to coincide with rotor 1.
263
diameter d (mm)
Stack length w (mm)
rotor-stator stack overlap (mm)
No. slots
Carter Factor K
Carter Factor (Approx.)
Carter Factor (Ossana Approx.)
air gap g (mm)
air gap (tol.)
slot mouth width wo (mm)
air gap (eff.) geff (mm)
(mm)
Stator (3 values are given for three different air gaps) 175.065 175.065 175.065 195.5 195.5 195.5 48 48 48 1.179 1.178 1.186 1.193 1.192 1.198 1.253 1.252 1.259 0.547 0.555 0.510 3.2 3.2 3.2 -
Rotor 1: Nested-loop design 173.971 189.9 189.9 36 1.104 1.114 1.167 0.547 0.03 2.8 0.712
Rotor 2: New double layer design 173.955 190.2 189.4 36 1.064 1.138 1.120 0.555 0.03 2.1 0.696
Rotors 3 and 4: isolated loop designs 173.955 189.9 188.3 36 1.064 1.138 1.120 0.555 0.03 2.1 0.696
Rotor 5: 6 bar squirrel cage design (simulation only) 173.971 189.9 189.9 6 K = 1 by assumption 0.547 0.03 8.4 0.645
Rotor 6: wound rotor (simulation only) 173.971 189.9 189.9 48 1.181 1.194 1.255 0.547 0.03 3.2 0.762
Rotor 7: standard squirrel cage rotor 174.046 190.3 189.3 40 closed slots 0.510 0.02 0.605
264
Table C.2: Physical data for the experimental (and simulation) BDFM stator and rotor designs
265
where = 4 wo wo arctan 2g 2g loge 1+ wo 2g (C.15)
and where wo is the slot mouth width, ys is the slot pitch (in distance units) and g is the air gap width as per gure C.1. When both rotor and stator have open slots it is assumed that the resultant Carter factor is the product of the Carter factors for the stator and rotor respectively (see [43, sect. 6.1] and references therein for details and experimental justication).
K total = K r K s where K r and K s are the rotor and stator Carter factors respectively.
(C.16)
As noted in [43] the assumption of innite slot pitch length is not always reasonable. In [22] and [39, ch. 17], Carters original calculation is generalised for the case of a regular slotting pattern. [39] claims that Carters initial assumption i.e. equations (C.14) is quite sufcient for practical application, which is at odds with [43, p. 56]. In practice as long as the ratio the ratio
wo ys wo wo g
is sufciently large (or equivalently
sufciently small, however in normal machines this is rarely the case) then (C.14) is a
good approximation. Sometimes these assumptions are not reasonable, and [43, sect. 6.1] gives the following two alternative approximations which have empirical justication: When
wo g
< 12: K ys 3 ys + g 4 wo ys + 8g ys wo + 8g (C.17)
Or an alternative approximate formula is given as K (C.18)
266
Appendix D
Previous BDFMs
267
Previous BDFMs
Location
Frame name
Frame size (mm)
Air gap (mm)
Air gap dia. (mm)
Stack Length (mm)
Poles
No. Rotor Slots
No. Stator Slots
Original Torque (Nm)
Claimed power kW
Ref.
Year
Broadway/Burbridge Bristol, UK Bristol, UK Spe/Wallace Oregon NEMA 254T 158.75 0.6 210 75 6/2 44 36 40 [110, 104, 127, 11] 1989-1995 177.8 69.85 6/2 18/12 48 (40 used) 60 54 [17] [17] 1970 1970
Brune/Spe/Wallace Oregon Williamson Cambridge McMahon/Ali Lotia Cambridge McMahon/FKI Cambridge Shoudao et. al. China Runcos/WEG Brazil 12/8 100 [90] 2004 6/2 30 [96] 2000 D180 180 0.55 175 190 8/4 36 48 140 2001 D100 100 0.42 91.7 96 6/2 32 36 15 2 [68] 2000 D160 160 273 8/4 36 48 70 [114] 1997 NEMA 182T 114.30 6/2 12 [19] 1994
Table D.1: Prototype nested-loop rotor BDFMs. This value is hard to believe from the evidence presented.
268
Appendix E
Experimental Apparatus
The test rig for the prototype BDFM is shown in gure E.3. Details on the prototype machine can be found in appendix B. The test rig was designed and assembled for the purposes of this research. The purpose of the design was to produce a test rig which would allow steady-state and dynamic tests to be performed on the prototype BDFM machine over its full potential operating range. The operating range was considered to be 0-150 N m and 0-2000 rpm. In practice the extremes of this operating envelope were not attainable, due to limitations that will be discussed in the sequel. The apparatus will now be described.
E.1 Apparatus Description

The apparatus is depicted in gures E.2 and E.3. The apparatus consists broadly of three parts: the prototype BDFM machine and power supplies; the DC load motor and power supplies and control and instrumentation hardware. In addition a Yokogawa 3 phase power meter, part number WT130 was available for conrmatory measurements. The DC machine armature was supplied from two unregulated 220V, 30A DC supplies which could be connected in series or parallel giving a maximum power of 13.2 kW, this was the maximum power the rig could supply. The DC machine eld winding was supplied from a regulated 600V, 4A DC supply. The BDFM stator 1 winding was supplied from the 3 phase, 50 Hz grid, connected through a 25A per phase Variac, variable from 0 to 245Vrms (phase). The stator 2 winding was connected to an inverter through a L-C line lter. The PWM switching frequency was variable, but normally set to 10 kHz. The control and instrumentation was based around a dedicated slave PC, with an AMD Athlon 1.4 GHz processor. This slave PC ran a real-time operating system, specically designed to be programmed by, and interface to Simulink/Matlab, called xPC Target. This slave PC will be referred to as the target PC. The target PC contained a number of peripheral boards to perform data acquisi269
270
tion and output data. The target PC performed data logging, this data was accessible over a TCP/IP network connection for subsequent downloading. The target PC was programmed, and monitored from a host PC running Matlab/Simulink under MS Windows. The host PC simply functioned as an interface to the target PC. Figure E.1 shows a typical screen shot of the the target PC when running. The screen shot shows the basic signal monitoring features of the target computer. The screen shot shows current and voltage wave forms, along with some numerical data.
Figure E.1: Typical screen shot of the target PC, showing current and voltage wave forms and numerical data Due to the extremely high levels of E-M interference generated by the inverter the elimination of noise from measurements was particularly problematic. Where possible signals were isolated using opto-isolators and low impedance earthing was provided throughout. These measures combined with some ltering gave acceptable performance.
E.1.1
xPC Target PC and peripheral boards
Table E.1 gives details of the peripheral boards in the target PC. The FPGA based inverter output board was specied by the author and designed by George Makrides[70].
E.1.2
Torque Transducer
The transducer was manufactured by Hottinger Baldwin Messtecknik (HBN), Germany part number T30FN with display unit and amplier part numbers DA3417 and MD555. The transducer was rated 0-1000 N m at up to 5 kHz and up to 3000 rpm. In practice, the power supply to the transducer was
271
Manufacturer/Part No. Measurement Computing / PCI-DIO96H
Description 96 bit digital I/O
Specication Organised in 8-bit groups, 5V levels, read latency approx. 750 ns, write latency approx 250 ns.
Purpose Receiving parallel data
from voltage A/D, total read time approx. 9 s for 6 16 bit inputs. Receiving calculations, parallel total data read from speed and position time approx. 4 s. Measurement of analogue current signals from Hall effect sensors, torque reading from transducer, ux reading from integrating amplier. Total read time, approx 40 s.
Measurement Computing / PCI-DIO96H
96 bit digital I/O
Organised in 8-bit groups, 5V levels, read latency approx. 750 ns, write latency approx 250 ns.
Measurement Computing / PCI-DAS1602/16
16 single-ended (or 8 differential) input 16-bit A/D, 2 16-bit D/A, 24 bit digital I/O
A/D 200kSamples/s (one input), hence read latency approx. 5 s per input.
PLX-9030 and Xilinx Spartan 2e FPGA
Custom designed implementation of space-vector modulation, to generate switching patterns for the inverter. Variable dead-time (0-4 s), switching frequency (1-50 kHz), modulation frequency (-200 to 200 Hz), av. latency switching period, hence 50 s at 10 kHz Table E.1: xPC Target PC peripheral boards
1 2
272
poorly decoupled resulting in large amounts of noise being picked up from the inverter. Therefore the useful bandwidth was only a few Hz.
E.1.3
DC load motor
The DC load motor specications are given in table E.2. For further details see [69]. As previously mentioned the maximum power the DC machine could deliver was 13.2 kW, which was the supply limit. The maximum power the machine can dissipate is 13.9 kW which was limited by the load resistors. Parameter Manufacturer Frame size Nature Rated power Rated speed Armature voltage Armature current Field winding voltage Field winding current Armature resistance Armature inductance Field winding resistance Field winding inductance Value GEC, UK SD160XLB separately excited 24 kW 1750 rpm 460V 61A 360V 1.83A 0.428 7.5 mH 196.7 74 H
Table E.2: DC load motor specication
E.1.4
Inverter Output Filter
The inverter output lter was a 3 phase L-C lter. The lter was manufactured by Schaffner, part number FN 5010-18-99 and was a star connected L-C lter with L=3.5 mH and C=1.5 F giving the resonant pole at approximately 2.2 kHz. The lter is rated at 18A rms per phase, and 400Vrms line at 40 C. The manufacturer states that the lter can be used with switching frequencies from 4 kHz to 16 kHz.
E.1.5
Inverter
The inverter comprised of 3 half-bridge IGBT modules. The modules were Semikron SKiiP 432 GB 120 - 207 CTV modules, which contain two N-channel IGBT devices in series and accompanying
273
gate drive circuitry, including measures to protect the devices from overheating due to desaturation, undervoltage and short circuit protection. The latter ensures that a suitable dead-time between ring high and low side devices is employed. The devices were rated at 1200V and 400A. The devices were supplied from a 600V DC link derived from a uncontrolled 3 phase rectier connected directly to the mains. The devices were mounted in an earthed protective enclosure on a heat sink with forced air cooling. The overall inverter rating was calculated as 200Arms per phase with a 600V DC link at 8 kHz switching frequency, or 100Arms per phase at a 600V DC link at 25 kHz switching frequency. The inverter also included a brake-chopper, Semikron part number SKAI 100 E, which provided bang-bang regulation of the DC link voltage by switching in the inverter protection resistors when the DC link voltage exceeded around 680V. This facility was necessary when testing the BDFM, as when the machine is in operation as a generator, above natural speed, or as a motor below natural speed, the inverter connected winding regenerates power, which would otherwise cause the DC link voltage to increase.
E.1.6
Position and Speed Measurements
The position and speed measurements were derived from an optical encoder (or resolver), which produced 2500 pulses per revolution. A state-machine was implemented on a fast 8-bit micro controller, a Scenix SX28 running at 50MIPS, to determine the rotor angular position to an accuracy of 0.036 . The data was sent as 15 bit parallel data, with one bit representing the direction of rotation for the last transition. The data was fed into the input of one of the digital I/O cards. Handshaking was achieved by an assertion from the digital I/O card that a read was imminent; during this time the SX28 refrained from modifying its output pins. The latency between position updates was approximately 2 s. The speed measurement was achieved with another SX28 micro controller, which used a 50MHz clock to time pulse widths. This data was output as a oating point number with a 5 bit exponent, 12 bit mantissa, and a saturation bit, to one of the digital I/O cards. The speed was computed by the target PC by performing the requisite division. As the optical encoder gave 2500 pulses per revolution the minimum speed measurable before saturation occurred was approximately 0.07 rpm. At 1000 rpm the speed measurement was accurate to better than 1 rpm, at 2000 rpm the accuracy was better than about 4rpm. As with the position measurement, handshaking was achieved by means of an assertion from the I/O board. The outputs from the optical encoder were isolated using opto-isolators to ensure that no ground loops were present and to protect the equipment.
274
E.1.7
Analogue to Digital Converters
The stator 1 and stator 2 phase currents were measured using Hall effect current sensors, LEM part number LTA-100P, wound for a maximum current of 25A rms . The Hall effect sensors provide galvanic isolation, however it was necessary to place the sensors in the return current path, rather than the out current path, to reduce the capacitive coupling between the sensor circuitry and the noisy supply voltage to stator 2 from the inverter. The sensors were read by the DAS1602/16 16 bit A/D card. The sensor bandwidth was approximately 100 kHz, hence the bandwidth was limited by the sample rate. The voltage measurement was achieved with a custom designed A/D board designed by the author and implemented by Davor Dukic and Iskandar Samad. The board consists of 6 parallel 16 bit A/D converters, AD677, driven by a differential input with a 2 kHz 4 pole Butterworth anti-aliasing lter. The converters were isolated via opto-isolators, and the data conversion controlled by 7 PIC micro controllers. The data was acquired by one of the digital I/O boards, in parallel format. The maximum input voltage was 1000V peak, and the input impedance was 1 M .
E.1.8
DC Machine load resistors and Inverter dump resistors
The inverter dump resistors were six 24 , 10A resistors connected as three in series and the two such sets in parallel, giving a total resistance of 36 at 20A, or 10 kW at 600V. Therefore the inverter could withstand 10 kW of regeneration. The DC machine load resistors comprises of three resistors each congured at 10.5 of 3.5 and rated at 21A. Generally either two or three resistors were connected in parallel giving a minimum resistance at a current of 63A, and total power dissipation of 13.9 kW, which was the maximum power which could be dissipated by the test rig.
3 phase instantaneous voltages
DC machine armature Torque transducer
Power winding 3 phase voltage Control winding 3 phase voltage
AC Variac 3 phase Halleffect current transducers LC filter
load resistor
DC supply
DC machine
BDFM
incremental encoder signal
Inverter Opto Isol. switching signals
Opto Isol. Integrating amplifier field winding DC supply flux search coil (analog V) shaft torque (analogue voltage)
Rectified 3 phase mains giving 600V DC link over voltage protection to dissapate regenerated energy.
Speed decoder
Posn. decoder
16bit A/D
3 phase PWM
xPC Target PC
Opto Isol. Opto Isol. PCIDAS1602/16

PCI Bus
PCIDIO96H
115kbit RS232
xPC Target PC running realtime kernel AMD Athlon 1.4GHz processor 512Mb RAM
Bluetooth receiver
PCIDIO96H
Rogowski coil current Transducer Bluetooth Transmitter (internal to machine)
TCP/IP network
Host PC MS Windows ME Matlab/Simulink/xPC
PLX PCI Bridge Xilinx FPGA
275
Figure E.2: Test rig functional diagram for the prototype BDFM
276
Figure E.3: Test rig for the prototype BDFM
Bibliography
[1] E. Abdi-Jalebi, P. C. Roberts, and R. A. McMahon. Real-time rotor bar current measurement using a rogowski coil transmitted using wireless technology. In 18th Intl. Power Systems Conf. (PSC2003), Iran, volume 5, pages 19, October 2003. [2] Gerald C. Alexander. Characterization of the brushless, doubly-fed machine by magnetic eld analysis. In Proc. IEEE Industry Applications Society (IAS) Annual Mtg., pages 6774, Seattle,WA, October 7-12 1990. IEEE. [3] Pierre Apkarian and Richard J. Adams. Advanced gain-scheduling techniques. IEEE Transactions on Control Systems Technology, 6(1):2132, January 1998. [4] Carolyn Beck and Pascale Bendotti. Model reduction methods for unstable uncertain systems. In Proc. Conference on Decision and Control. IEEE CSS, December 1997. [5] Carolyn Beck and John Doyle. A necessary and sufcient minimality condition for uncertain systems. IEEE Trans. Automatic Control, 44(10):18021813, October 1999. [6] Carolyn Beck, John Doyle, and Keith Glover. Model reduction of multidimensional and uncertain systems. IEEE Trans. Automatic Control, 41(10):14661477, October 1996. [7] G. Becker and A. Packard. Robust performance of linear parametrically varying systems using parametrically-dependent linear feedback. Systems and Control Letters, 23, 1994. [8] Shibashis Bhowmik, Ren Spe, and Johan H. R. Enslin. Performance optimization for doubly fed wind power generation systems. IEEE Trans. Industry Applications, 35(4):949958, July/August 1999. [9] T. S. Birch and O. I. Butler. Permeance of closed-slot bridges and its effect on induction motor current consumption. Proc. IEE, 118(1), January 1971. [10] M. Boger, A. Wallace, and R. Spe. Investigation of appropriate pole number combinations for brushless doubly fed machines as applied to pump drives. IEEE Transactions on Industry Applications, 31(5):10221028, 1996.
277
278
BIBLIOGRAPHY
[11] M. Boger, A. Wallace, R. Spe, and R. Li. General pole number model of the brushless doublyfed machine. IEEE Transactions on Industry Applications, 31(5):10221028, 1995. [12] M. S. Boger. Aspects of Brushless Doubly-Fed Induction Machines. PhD thesis, University of Cambridge, 1997. [13] W. R. Brasseld, R. Spe, and T. G. Habetler. Direct torque control for brushless doubly-fed machines. In Conf. Record of 1992 IEEE Industry Applications Society (IAS) Annual Meeting, pages 615622. IEEE, 4-9 October 1992. Houston, TX, USA. [14] W. R. Brasseld, R. Spe, and T. G. Habetler. Direct torque control for brushless doubly fed machines. IEEE Transactions on Industry Applications, 32(5):10981103, 1996. [15] A. R. W. Broadway. Cageless induction machine. Proc. IEE, 118(11):15931600, November 1971. [16] A. R. W. Broadway. Brushless cascade alternator. Proc. IEE, 1974. [17] A. R. W. Broadway and L. Burbridge. Self-cascaded machine: a low-speed motor or high frequency brushless alternator. Proceedings, Institution of Electrical Engineers, 117:1277 1290, 1970. [18] G. S. Brosan and J. T. Hayden. Advanced Electrical Power and Machines. Sir Isaac Pitman & Sons Ltd., 1966. [19] Chris S. Brune, Ren Spe, and Alan K. Wallace. Experimental evaluation of a variable-speed, doubly-fed wind-power generation system. IEEE Trans. Industry Applications, 30(3):648655, May/June 1994. [20] A. Bruce Carlson. Circuits: Engineering Concepts and Analysis of Linear Electric Circuits. John Wiley & Sons Inc., New York, 1996. [21] F. W. Carter. Note on air-gap and interpolar induction. Institution of Electrical Engineers Journal, 29, 1900. [22] F. W. Carter. The magnetic eld of the dynamo-electric machine. Institution of Electrical Engineers Journal, 64, 1926. [23] Paolo Castaldi, Walter Geri, Marcello Montanari, and Andrea Tilli. A new adaptive approach for on-line parameter and state estimation of induction motors. Control Engineering Practice, 13(1):8194, January 2005. [24] C. D. Cook and B. H. Smith. Stability and stabilisation of doubly-fed single frame cascade induction machines. IEE Proceedings, 126(11):11681174, 1979.
BIBLIOGRAPHY
279
[25] C. D. Cook and B. H. Smith. Effects of machine parameter values on dynamic response and stability regions of doubly-fed cascade induction machines. IEE Proceedings - B Electrical Power Applications, 130(2):137142, 1983. [26] Peter A. Cook. Nonlinear Dynamical Systems. Prentice Hall, 1994. [27] F. Creedy. Some developments in multi-speed cascade induction motors. Institution of Electrical Engineers, Journal, pages 511537, 1920. [28] Henry DAngelo. Linear Time-Varying Systems: Analysis and Synthesis. Series in Electrical Engineering. Allyn and Bacon, 1970. [29] J. A. de Kock, F. S. van der Merwe, and H. J. Vermeulen. Induction motor parameter estimation through an output error technique. IEEE Trans. on Energy Conversion, 9(1):6975, March 1994. [30] A. Draper. Electrical Circuits Including Machines. Longmans, 1 edition, 1964. [31] A. Draper. Electrical Machines. Longmans, 2 edition, 1967. [32] Guang-Ren Duan and Ron J. Patton. A note on hurwitz stability of matrices. Automatica, 34(4):509511, 1998. [33] A. C. Ferreira. Analysis of the Brushless Doubly-Fed Induction Machine. PhD thesis, University of Cambridge, 1996. [34] A. C. Ferreira and S. Williamson. Time-stepping nite-element analysis of brushless doubly fed machine taking iron loss and saturation into account. IEEE Transactions on Industry Applications, 35(3):583588, 1999. [35] A.E. Fitzgerald, Jr. Charles Kingsley, and Stephen D. Umans. Electric Machinery. Electrical Engineering. McGraw-Hill, New York, 4 edition, 1983. [36] Pascal Gahinet and Pierre Apkarian. A linear matrix inequality approach to H control. Int. J. of Robust and Nonlinear Control, 4, 1994. [37] Pascal Gahinet, Arkadi Nemirovski, Alan J. Laub, and Mahmoud Chilali. Control Toolbox Users Guide. The Mathworks Inc., 1995. LMI
available from
http://www.mathworks.com/access/helpdesk/help/pdf_doc/lmi/lmi.pdf. [38] F. R. Gantmacher. The Theory of Matrices. Chelsea, New York, 2 edition, 1990. 2 vols. [39] William. J. Gibbs. Conformal Transformations in Electrical Engineering. Chapman & Hall Ltd., London, 1958.
280
BIBLIOGRAPHY
[40] Keith Glover. All optimal hankel-norm approximations of linear multivariable systems and their L -error bounds. Int. J. Control, 39(6):11151193, 1984. [41] B. V. Gorti, G. C. Alexander, and R. Spe. Power balance considerations for brushless doublyfed machines. IEEE Transactions on Energy Conversion, 11(4):687692, December 1996. [42] B. V. Gorti, G. C. Alexander, R. Spe, and A. K. Wallace. Characteristics of a brushless doublyfed machine in current-fed mode of operation. In Proc. Conference on Industrial Automation & Control (I,A & C), pages 143148. IEEE/IAS, IEEE, 5-7 January 1995. Hyderabad, India. [43] Bedich Heller and Vclav Hamata. Harmonic Field Effects in Induction Machines. Elsevier r Scientic Publishing Co., 1977. [44] Y. C. Ho and D. L. Pepyne. Simple explanation of the no-free-lunch theorem and its implications. J. of Optimization Theory and Applications, 115(3):549570, December 2002. [45] B. Hopfensperger, D. J. Atkinson, and R. A. Lakin. Stator ux orientated control of a cascaded doubly-fed induction machine. Proc. IEE B - Electric Power Applications, 146(6):597605, November 1999. [46] B. Hopfensperger, D. J. Atkinson, and R. A. Lakin. Combined magnetising ux orientated control of the cascaded doubly-fed induction machine. Proc. IEE B - Electric Power Applications, 148(4):354362, July 2001. [47] Roger A. Horn and Charles R. Johnson. Topics in Matrix Analysis. Cambridge University Press, 1991. [48] Louis J. Hunt. A new type of induction motor. Institution of Electrical Engineers, Journal, pages 648677, 1907. [49] Louis J. Hunt. The cascade induction motor. Institution of Electrical Engineers, Journal, pages 406434, 1914. [50] A. Isidori. Nonlinear Control Systems 3rd Edition. Springer, 1995. [51] Morris Jevons. Electrical Machine Theory. Blackie & Son Limited, London, 1966. [52] Charles V. Jones. The Unied Theory of Electrical Machines. Butterworths,London, 1967. [53] Alson Kemp, Michael Boger, Ernesto Wiedenbrg, and Alan Wallace. Investigation of rotorcurrent distributions in brushless doubly-fed machines. In Proc. IEEE Industry App. Soc. Annual Mtg., pages 638643. IEEE, 6-10 October 1996. San Diego, CA. [54] Hassan K. Khalil. Nonlinear Systems. Prentice-Hall, Inc., New Jersey, 2 edition, 1996.
BIBLIOGRAPHY
281
[55] Brian Kock, Ren Spe, and Bryan Clever. A comparison of stack preparation methods for bar insulation in diecast rotors. In Proc. IEEE Industry Applications Society Annual Meeting, volume 1, pages 182187. IEEE IAS, IEEE, New Orleans, October 5-9 1997. [56] Paul C. Krause, Oleg Wasynczuk, and Scott D. Sudhoff. Analysis of electric machinery and drive systems. IEEE Press Wiley, New York, second edition, 2002. [57] Alexander Kusko and Clement Somuah. Speed control of a single-frame cascade induction motor with slip-power pump back. IEEE Trans. Industry Applications, IA-14(2):97105, 1978. [58] Sanjay Lall and Carolyn Beck. Error-bounds for balanced model-reduction of linear timevarying systems. IEEE Trans. Automatic Control, 48(6):946956, June 2003. [59] H. K. Lauw. Brushless doubly-fed motor control system. US Patent No.: 5,239,251, August 1993. [60] R. Li, R. Spe, A. K. Wallace, and G. C. Alexander. Synchronous drive performance of brushless doubly-fed motors. IEEE Transactions on Industry Applications, 30(4):963970, July/August 1994. [61] R. Li, A. Wallace, and R. Spe. Determination of converter control algorithms for brushless doubly-fed induction motor drives using Floquet and Lyapunov techniques. IEEE Transactions on Energy Conversion, 10(1):7885, 1995. [62] Ruqi Li, Alan Wallace, and Ren Spe. Determination of converter control algorithms for stable brushless doubly-fed drives using oquet and lyapunov techniques. In Proc. Power Electronics Specialists Conference (PESC), pages 571577. IEEE, 24-27 June 1991. Cambridge, MA. [63] Ruqi Li, Alan Wallace, and Ren Spe. Dynamic simulation of brushless doubly-fed machines. IEEE Transactions on Energy Conversion, 6(3):445452, 1991. [64] Ruqi Li, Alan Wallace, and Ren Spe. Two-axis model development of cage-rotor brushless doubly-fed machines. IEEE Transactions on Energy Conversion, 6(3):453460, 1991. [65] Yuefeng Liao. Design of a brushless doubly-fed induction motor for adjustable speed drive applications. In Proc. IEEE Industry App. Soc. Annual Mtg., pages 850855. IEEE, 6-10 October 1996. [66] Antonio Marcus Nogueira Lima, Cursino Brando Jacobina, and Eurico Bezerra de Souza Filho. Nonlinear parameter estimation of steady-state induction machine models. IEEE Trans. on Industrial Electronics, 44(3):390397, June 1997.
282
BIBLIOGRAPHY
[67] R. D. Lorenz. Observers and state lters in drives and power electronics. In Proc. IEEE IAS OPTIM, Brasov, Romania, May 16-18 2002. IEEE/IAS. Keynote paper. [68] Mohammed Ali Lotia. Control strategies for variable speed brushless doubly-fed induction motor drives. Masters thesis, University of Cambridge, 2000. [69] R.A. Majzoub. The role of damper windings in suppressing electromechanical oscillations of diesel-driven synchronous generators. PhD thesis, University of Cambridge, June 2001. [70] George Makrides. The design and implementation of a hardware platform for driving and controlling brushless doubly-fed machines. Mphil., University of Cambridge, August 2004. [71] M. Matsumoto and T. Nishimura. equidistributed on Modeling uniform and pseudorandom Computer Mersenne twister: number generator. 8(1):330, Simulation, A 623-dimensionally ACM January Trans. 1998.
http://www.math.sci.hiroshima-u.ac.jp/~m-mat/eindex.html. [72] R. G. Meadows. Electric Network Analysis. Penguin Education, 1972. [73] Alexandre Megretski, Chung-Yao Kao, Ulf Johnsson, and Anders Rantzer. Guide to IQC: Software for Robustness Analysis, February 2003. http://www.math.kth.se/~cykao/. [74] Alexandre Megretski and Anders Rantzer. System analysis via integral quadratic constraints. IEEE Trans. Automatic Control, 42(6):819830, June 1997. [75] S. Mller, M. Deicke, and Rik M. de Doncker. Doubly fed induction generator systems for wind turbines. IEEE Industry Applications Magazine, pages 2633, May/June 2002. [76] Georgios Papafotiou, Tobias Geyer, and Manfred Morari. trol of three-phase symmetric induction motors. Optimal direct torque conA
available from
Technical Report AUT03-07, Eid-
gen ossische Technische Hochschule (ETH), Zrich, Switzerland, April 2003. available from http://control.ee.ethz.ch/research/publications/publications.msql?id=1614. [77] E. Prempain, I. Postlethwaite, and A. Benchaib. A linear parameter variant H control design for an induction motor. Control Engineering Practice, 10(6):633644, June 2002. [78] Amuliu Bogdan Proca and Ali Keyhani. Identication of variable frequency induction motor models from operating data. IEEE Trans. on Energy Conversion, 17(1):2431, March 2002. [79] Ashok Ramchandran and Gerald C. Alexander. Frequency-domain parameter estimations for the brushless doubly-fed machine. In Conference Record of PCC 1993, pages 346351, Power Conversion Conference, Yokohama 1993, 1993. IEEE. 19-21 Apr 1993.
BIBLIOGRAPHY
283
[80] Ashok Ramchandran, Gerald C. Alexander, and Rn Spe. Off-line parameter estimation for the doubly-fed machine. In Proceedings of Conference on Industrial Electronics, Control, Instrumentation, and Automation, 1992. Power Electronics and Motion Control, volume 3, pages 12941298. IEEE, 1992. 9-13 Nov 1992. [81] N. S. Risley and Richard S. Burington. A matric theory development of the theory of symmetric components. London, Edinburgh & Dublin Philosophical Magazine, 27(184):605613, May 1939. [82] P. C. Roberts. Control of a bdfm motor. Masters thesis, University of Cambridge, 2000. [83] P. C. Roberts, E. Abdi-Jalebi, R. A. McMahon, and T. J. Flack. Real-time rotor bar current measurements using bluetooth technology for a brushless doubly-fed machine (BDFM). In Int. Conf. Power Electronics, Machines and Drives (PEMD), volume 1, pages 120125. IEE, March 2004. [84] P. C. Roberts, T. J. Flack, J. M. Maciejowski, and R. A. McMahon. Two stabilising control strategies for the brushless doubly-fed machine (BDFM). In Int. Conf. Power Electronics, Machines and Drives (PEMD) (Conf. Pub. No. 487), pages 341346. IEE, 4-7 June 2002. Bath, UK. [85] P. C. Roberts, J. M. Maciejowski, R. A. McMahon, and T. J. Flack. A simple rotor current observer with an arbitrary rate of convergence for the brushless doubly-fed (induction) machine (BDFM). In Proc. IEEE Joint CCA, ISIC, CACSD, pages 266271, Taipai, Taiwan, 2-4 September 2004. paper no. 683. [86] P. C. Roberts, R. A. McMahon, P. J. Tavner, J. M. Maciejowski, and T. J. Flack. An equivalent circuit for the brushless doubly-fed machine (BDFM) including parameter estimation. Proc. IEE B - Elec. Power App., 2004. accepted for publication, November 2004. [87] P. C. Roberts, R. A. McMahon, P. J. Tavner, J. M. Maciejowski, T. J. Flack, and X. Wang. Performance of rotors for the brushless doubly-fed (induction) machine (BDFM). In Proc. 16th Int. Conf. Electrical Machines (ICEM), pages 450455, September 2004. 5th-8th, Cracow, Poland. [88] P. Rochelle, R. Spe, and A. K. Wallace. The effect of stator winding conguration on the performance of brushless doubly-fed machines in adjustable speed drives. In Conference record of the IEEE Industry Applications Society Annual Meeting, pages 331337, Seattle,WA, October 7-12 1990. IEEE. [89] Wilson J. Rugh and Jeff S. Shamma. Research on gain scheduling. Automatica, 36(10):1401 1425, October 2000.
284
BIBLIOGRAPHY
[90] F. Rncos, R. Carlson, A. M. Oliveira, P. Kuo-Peng, and N. Sadowski. Performance analysis of a brushless doubly-fed cage induction generator. In Proc. 2nd Nordic Windpower Conf., 2004. 1-2 March. [91] Mulukutla S. Sarma. Electric Machines: Steady-state Theory and Dynamic Performance. West Publishing Company, 1986. [92] M. G. Say. The Performance and Design of Alternating Current Machines. Sir Isaac Pitman & Sons Ltd., 2 edition, 1948. [93] Carsten Scherer and Siep Weiland. Lecture notes DISC course on linear matrix inequalities in control. Dutch Institute of Systems and Control, Delft, Netherlands, April 1999. available from http://www.ocp.tudelft.nl/sr/personal/Scherer/lmi.pdf. [94] Fukuo Shibata and Toshio Kohrin. A brushless, self-excited polyphase synchronous generator. IEEE Trans. on Power Apparatus and Systems, PAS-102(8):24132419, August 1983. [95] Fukuo Shibata and Kouji Taka. Speed control for brushless cascade induction motors in control range of slips s1 > 1 and s2 > 1. IEEE Trans. on Energy Conversion, EC-2(2):246253, June 1987. [96] Huang Shoudao, Huang Keyuan, Zhou Lawu, and Lai Lieen. A study of the control strategy on rotor eld orientation for brushless doubly-fed machine. In Proc. 3rd Power Electronics and Motion Control Conference (PIEMC) 2000, volume 1, pages 508513, Beijing, China, 5-18 August 2000 2000. IEEE. [97] Siemens Brothers & Co. Ltd. and Francis Lydall. Improvements in polyphase induction motors. British Patent No.: 16839, July 1902. [98] Gordon R. Slemon. Modelling of induction machines for electric drives. In Conference record of the IEEE Industry Applications Society Annual Meeting, volume 1, pages 111115. IEEE, October 7-12 1989. [99] Gordon R. Slemon. Electrical Machines and Drives. Addison-Wesley, 1992. [100] G.R. Slemon and A. Straughen. Electrical Machines. Addison-Wesley, 1980. [101] Jean-Jaques E. Slotine and Weiping Li. Applied Nonlinear Control. Prentice-Hall, 1991. [102] B. H. Smith. The theory and performance of a twin stator induction machine. IEEE Transactions on Power Apparatus and Systems, 85(2):123131, February 1966. [103] B. H. Smith. Synchronous behaviour of doubly-fed twin stator induction machine. IEEE Transactions on Power Apparatus and Systems, 86(10):12271236, 1967.
BIBLIOGRAPHY
285
[104] R. Spe, A. K. Wallace, and H. K. Lauw. Performance simulation of brushless doubly-fed adjustable speed drives. In Conference record of the IEEE Industry Applications Society Annual Meeting, San Diego, CA, 1989. IEEE. [105] Robert F. Stengel. Optimal Control and Estimation. Dover Publications Inc., 1994. [106] Z. Bo Tang. Adaptive partitioned search to global optimization. IEEE Transactions on Automatic Control, 39(11):22352244, November 1994. [107] The Mathworks, editor. Matlab Function Reference, volume 1. The Mathworks Inc., 7.0 edition, 2004. http://www.mathworks.com/access/helpdesk/. [108] The Mathworks, editor. Using Simulink. The Mathworks Inc., 2004.
http://www.mathworks.com/access/helpdesk/. [109] A. K. Wallace, P. Rochelle, and R. Spe. Rotor modelling and development for brushless doubly-fed machines. In Conference record of the International Conference on Electrical Machines, volume 1, Cambridge,MA, August 12-15 1990. ICEM. [110] A. K. Wallace, R. Spe, and H. K. Lauw. Dynamic modelling of brushless doubly-fed machines. In Conference record of the IEEE Industry Applications Society Annual Meeting, San Diego, CA, 1989. IEEE. [111] Samuel S. Waters and Ronald D. Willoughby. Modeling of induction motors for system studies. IEEE Trans. on Industry Applications, IA-19(5):875878, September 1983. [112] S. Williamson. Field analysis for rotating induction machines and its relationship to the equivalent-circuit method. IEE Proc. Pt. B - Elec. Power App., 127(2):8390, March 1980. [113] S. Williamson. Power factor improvement in cage-rotor induction motors. IEE Proceedings B Electrical Power Applications, 130(2):121129, 1983. [114] S. Williamson and A. C. Ferreira. Generalised theory of the brushless doubly-fed machine. part 2: Model verication and performance. IEE Proceedings - Electric Power Applications, 144(2):123129, 1997. [115] S. Williamson, A. C. Ferreira, and A. K. Wallace. Generalised theory of the brushless doublyfed machine. part 1: Analysis. IEE Proceedings - Electric Power Applications, 144(2):111 122, 1997. [116] S. Williamson and E. R. Laithwaite. Generalised harmonic analysis for the steady-state performance of sinusoidally excited cage induction motors. IEE Proceedings - B Electric Power Applications, 132(3):157163, 1985.
286
BIBLIOGRAPHY
[117] Stephen Williamson and Michael Boger. Impact of inter-bar current on the performance of the brushless doubly fed motor. IEEE Trans. Industry Applications, 35(2):453460, March/April 1999. [118] G. D. Wood, P. J. Goddard, and K. Glover. Approximation of linear parameter-varying systems. In Proc. 35th Conference on Decision and Control (CDC), pages 406411, Kobe, Japan, December 1996. IEEE CSS. [119] Giles David Wood. Control of Parameter-Dependent Mechanical Systems. PhD thesis, University of Cambridge, 1995. [120] Fen Wu, Xin Hua Yang, Andy Packard, and Greg Becker. Induced L 2 -norm control for LPV systems with bounded parameter variation rates. Int. Journal of Robust and Nonlinear Control, 6(9-10):983998, November 1996. [121] Dante C. Youla and Jr. Joseph J. Bongiorno. A oquet theory of the general linear rotating machine. IEEE Trans. Circuits and Systems, CAS-27(1):1519, January 1980. [122] Hugh D. Young. University Physics. Addison Wesley, 8 edition, 1992. [123] Anatoly A. Zhigljavsky. Theory of Global Random Search. Kluwer Academic Publishers, 1991. [124] D. Zhou and R. Spe. Field orientated control development for brushless doubly-fed machines. In Conference record of the Industry Applications Society (IAS) Annual Meeting 1996, volume 1, pages 304310. IEEE, 6-10 October 1996. Taipai, Taiwan. [125] D. Zhou, R. Spe, and G. C. Alexander. Experimental evaluation of a rotor ux oriented control algorithm for brushless doubly-fed machines. IEEE Transactions on Power Electronics, 12(1):7278, 1997. [126] D. Zhou, R. Spe, G. C. Alexander, and A. K. Wallace. A simplied method for dynamic control of brushless doubly-fed machines. In Conference record of the 22nd Industrial Electronics, Control and Instrumentation Conference (IECON) 1996, volume 2, pages 946951. IEEE, 5-10 August 1996. Taipai, Taiwan. [127] D. Zhou, R. Spe, and A. K. Wallace. Laboratory control implementations for doubly-fed machines. In Proceedings of the IECON 93, volume 2, pages 11811186, IEEE Conf. on Industrial Electronics, Control, and Instrumentation, 1993. 15-19 Nov 1993. [128] D. Zhou, R. Spe, and A. K. Wallace. Model reference adaptive speed control for doubly-fed machines. In Proceedings of International Conference on Industrial Electronics, Control and Instrumentation (IECON), volume 2. IEEE, 1993.
BIBLIOGRAPHY
287
[129] Dongsheng Zhou, R. Spe, and G. C. Alexander. Experimental evaluation of a rotor ux orientated control algorithm for brushless doubly-fed machines. In Conference record of the 27th Power Electronics Specialists Conference (PESC) 1996, volume 1, pages 913919. IEEE, 23-27 June 1996. Baveno, Italy. [130] Dongsheng Zhou and Ren Spe. Synchronous frame model and decoupled control development for doubly-fed machines. In Conference record of the 25th Power Electronics Specialists Conference (PESC) 1994, volume 2, pages 12291236. IEEE, 20-23 June 1994. Taipei, Taiwan. [131] Kemin Zhou, John C. Doyle, and Keith Glover. Robust and Optimal Control. Prentice-Hall, 1996.
Author Index
Abdi-Jalebi, E. 141, 210 Adams, Richard J. 182, 183 Alexander, G. C. 4, 6, 12, 13, 87, 98100, 156, 168, 181, 187, 211 Alexander, Gerald C. 4, 13, 36, 137, 140 Apkarian, Pierre 182, 183, 216 Atkinson, D. J. 7, 211 Beck, Carolyn 70, 209 Becker, G. 72, 73 Becker, Greg 183 Benchaib, A. 197, 211 Bendotti, Pascale 70 Bhowmik, Shibashis 10, 155 Birch, T. S. 257, 262 Boger, M. 4, 10, 12, 35, 49, 52, 54, 65, 78, 85, 155, 268 Boger, M. S. 4, 12, 49, 52, 64, 78, 85 Boger, Michael 4, 49, 80, 121, 122 Brasseld, W. R. 5, 187, 195 Broadway, A. R. W. 3, 9, 12, 35, 36, 39, 83, 87, 115, 119, 120, 122124, 126, 127, 135, 268 Brosan, G. S. 257 Brune, Chris S. 10, 268 Burbridge, L. 3, 9, 12, 35, 36, 39, 83, 87, 119, 120, 122124, 126, 127, 135, 268 Burington, Richard S. 92 Butler, O. I. 257, 262 Carlson, A. Bruce 101, 140 Carlson, R. 268 Carter, F. W. 262, 265 Castaldi, Paolo 138, 211 Charles Kingsley, Jr. 16 Chilali, Mahmoud 71, 72 Clever, Bryan 4 Cook, C. D. 3, 5, 156, 170 Cook, Peter A. 156, 166 Creedy, F. 2, 3, 35, 36, 87 DAngelo, Henry 5 de Doncker, Rik M. 10 de Kock, J. A. 138 de Souza Filho, Eurico Bezerra 138 Deicke, M. 10 Doyle, John 70, 209 Doyle, John C. 53, 6870, 72, 73, 214, 218 Draper, A. 28, 97, 257259 Duan, Guang-Ren 73 Enslin, Johan H. R. 10, 155 Ferreira, A. C. 4, 812, 21, 41, 87, 115, 121, 168, 208210, 268 Fitzgerald, A.E. 16 Flack, T. J. 87, 141, 149, 178, 182, 187, 197, 210 Gahinet, Pascal 71, 72, 216 Gantmacher, F. R. 213 Geri, Walter 138, 211 Geyer, Tobias 211 Gibbs, William. J. 265 Glover, K. 70, 71, 78 Glover, Keith 53, 6870, 72, 73, 214, 218 Goddard, P. J. 70, 71, 78
289
290
AUTHOR INDEX
Gorti, B. V. 4, 6, 12, 87, 98100 Habetler, T. G. 5, 187, 195 Hamata, Vclav 21, 28, 29, 31, 34, 35, 44, 230, 231, 265 Hayden, J. T. 257 Heller, Bedich 21, 28, 29, 31, 34, 35, 44, 230, r 231, 265 Ho, Y. C. 139 Hopfensperger, B. 7, 211 Horn, Roger A. 74 Hunt, Louis J. 2, 7, 8, 35, 36, 87, 112, 120, 127, 134, 209 Isidori, A. 188191 Jacobina, Cursino Brando 138 Jevons, Morris 88 Johnson, Charles R. 74 Johnsson, Ulf 72 Jones, Charles V. 52 Joseph J. Bongiorno, Jr. 162 Kao, Chung-Yao 72 Kemp, Alson 49, 80, 122 Keyhani, Ali 138 Keyuan, Huang 6, 268 Khalil, Hassan K. 166 Kock, Brian 4 Kohrin, Toshio 4 Krause, Paul C. 51, 52 Kuo-Peng, P. 268 Kusko, Alexander 3, 34 Laithwaite, E. R. 21, 23, 25 Lakin, R. A. 7, 211 Lall, Sanjay 70 Laub, Alan J. 71, 72 Lauw, H. K. 4, 5, 10, 12, 15, 21, 44, 87, 268
Lawu, Zhou 6, 268 Li, R. 4, 5, 12, 13, 35, 49, 52, 54, 65, 78, 85, 87, 98, 100, 156, 178, 268 Li, Ruqi 46, 12, 13, 44, 49, 78, 87, 98 Li, Weiping 188, 189 Liao, Yuefeng 4, 12, 87 Lieen, Lai 6, 268 Lima, Antonio Marcus Nogueira 138 Lorenz, R. D. 197 Lotia, Mohammed Ali 268 Lydall, Francis 1, 120 Maciejowski, J. M. 87, 149, 178, 182, 187, 197 Majzoub, R.A. 272 Makrides, George 270 Matsumoto, M. 139 McMahon, R. A. 87, 141, 149, 178, 182, 187, 197, 210 Meadows, R. G. 17, 18 Megretski, Alexandre 72 Montanari, Marcello 138, 211 Morari, Manfred 211 Mller, S. 10 Nemirovski, Arkadi 71, 72 Nishimura, T. 139 Oliveira, A. M. 268 Packard, A. 72, 73 Packard, Andy 183 Papafotiou, Georgios 211 Patton, Ron J. 73 Pepyne, D. L. 139 Postlethwaite, I. 197, 211 Prempain, E. 197, 211 Proca, Amuliu Bogdan 138 Ramchandran, Ashok 4, 13, 137, 140
AUTHOR INDEX
291
Rantzer, Anders 72 Risley, N. S. 92 Roberts, P. C. 87, 141, 149, 178, 182, 187, 197, 210 Rochelle, P. 4, 12, 36, 40, 119, 122 Rugh, Wilson J. 182 Rncos, F. 268 Sadowski, N. 268 Sarma, Mulukutla S. 36 Say, M. G. 115 Scherer, Carsten 216 Shamma, Jeff S. 182 Shibata, Fukuo 4 Shoudao, Huang 6, 268 Siemens Brothers & Co. Ltd. 1, 120 Slemon, Gordon R. 32, 141, 231 Slemon, G.R. 101 Slotine, Jean-Jaques E. 188, 189 Smith, B. H. 2, 3, 5, 156, 170 Somuah, Clement 3, 34 Spe, R. 46, 10, 12, 13, 15, 21, 35, 36, 40, 44, 49, 52, 54, 65, 78, 85, 87, 98100, 119, 122, 155, 156, 168, 178, 181, 187, 195, 211, 268 Spe, Ren 46, 10, 12, 13, 44, 49, 78, 87, 98, 155, 156, 187, 211, 268 Stengel, Robert F. 197 Straughen, A. 101 Sudhoff, Scott D. 51, 52 Taka, Kouji 4 Tang, Z. Bo 139 Tavner, P. J. 87, 149 Tilli, Andrea 138, 211 Umans, Stephen D. 16 van der Merwe, F. S. 138 Vermeulen, H. J. 138
Wallace, A. 4, 5, 10, 12, 13, 35, 49, 52, 54, 65, 78, 85, 155, 156, 178, 268 Wallace, A. K. 46, 810, 12, 13, 15, 21, 36, 40, 41, 44, 87, 98, 100, 115, 119, 121, 122, 156, 168, 181, 187, 208, 209, 211, 268 Wallace, Alan 46, 12, 13, 44, 49, 78, 80, 87, 98, 122 Wallace, Alan K. 10, 268 Wang, X. 149 Wasynczuk, Oleg 51, 52 Waters, Samuel S. 138 Weiland, Siep 216 Wiedenbrg, Ernesto 49, 80, 122 Williamson, S. 4, 812, 21, 23, 25, 41, 87, 115, 121, 168, 208210, 231, 268 Williamson, Stephen 4, 121 Willoughby, Ronald D. 138 Wood, G. D. 70, 71, 78 Wood, Giles David 70, 72, 78 Wu, Fen 183 Yang, Xin Hua 183 Youla, Dante C. 162 Young, Hugh D. 16 Zhigljavsky, Anatoly A. 139 Zhou, D. 5, 6, 10, 13, 156, 168, 181, 187, 211, 268 Zhou, Dongsheng 6, 13, 156, 187, 211 Zhou, Kemin 53, 6870, 72, 73, 214, 218

A Study of Brushless DFIMs

Uploaded by

Copyright:

Available Formats

A Study of Brushless DFIMs

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Study of Brushless DFIMs

Uploaded by

Copyright:

Available Formats

A Study of Brushless Doubly-Fed (Induction) Machines

C ONTRIBUTIONS IN M ACHINE A NALYSIS , D ESIGN AND C ONTROL

DISSERTATION SUBMITTED FOR THE DEGREE OF

Emmanuel College University of Cambridge

(revised January 2005)

To God, Ruth and my parents.

Approach of this work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Mutual Inductance of Machine Windings

2.5.3 2.6 2.6.1 2.6.2 2.7

Some properties of machine windings . . . . . . . . . . . . . . . . . . . . . . . . .

Effect of Slotting on Mutual inductance terms . . . . . . . . . . . . . . . . . . . . . BDFM Coupled-Circuit Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Simulation comparison of different BDFM model reduction techniques . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5 Possible Rotor Designs and Evaluation 5.1

5.3 5.4 5.5

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 137

6 BDFM Parameter Identication 6.1 6.2

Parameter Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 6.3.1

7 Modelling for Control of the BDFM 7.1 7.2

7.4.3 7.5 7.6 7.7 7.8 7.9

Linearization Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

Future work on linear model based control . . . . . . . . . . . . . . . . . . . . . . . 181

Towards A Practical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . 195

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 205

9 Conclusions and Future Work 9.1 9.2

B.3 Rotor 1: Nested-loop Rotor Design Details . . . . . . . . . . . . . . . . . . . . . . . 237

B.4 Rotor 2: New Double Layer Rotor Design Details . . . . . . . . . . . . . . . . . . . 239

B.8 Rotor 6: Wound Rotor Design Details . . . . . . . . . . . . . . . . . . . . . . . . . 249

Bibliography Author Index

Notation & Terminology

denotes that a set comprised of matrices of dimension n m, the elements

NOTATION & TERMINOLOGY

2-norm of a vector or vector signal. For x Rn , x

, which is known as the Jacobian of f when n = m. .. .

the n n identity matrix (also used to denote complex current)

NOTATION & TERMINOLOGY

the empty (null) set and set B

NOTATION & TERMINOLOGY

Electrical Machine Notation

NOTATION & TERMINOLOGY

K cos(t 2/3), and K cos(t 4/3) for some K , , .

Proportional-Integral-Derivative, a controller with transfer funcsee remark 2.6 + K d s.

NOTATION & TERMINOLOGY

1.1 Evolution of the BDFM

Lydalls work was to achieve self-cascaded operation by, in effect,

1.1 Evolution of the BDFM

1.1 Evolution of the BDFM

1.2 Description of the BDFM

1.2 Description of the BDFM

3 phase grid, 50Hz

Fractionally rated frequency converter

3 phase variable freq.

Figure 1.1: BDFM concept: a brushless, doubly-fed induction machine

Synchronous mode of operation

1.2 Description of the BDFM

Potential Applications for the BDFM

1.3 Approach of this work

1.3 Approach of this work

BDFM Model Development

1.3 Approach of this work

Analysis of stability of the BDFM and the development of control strategies