Modeling, analysis, state and parameter estimation of faulty induction machines

Abstract: This thesis covers a broad range of subjects pertaining to induction machines, faulty as well as healthy ones. Firstly, a completely new and unconventional methodology for deriving simplest possible, yet sufficiently accurate, mathematical models for induction machines with general electrical faults is presented. Examples of common faults covered are inter-turn short circuit and increased resistance in one or more stator phases. Some common rotor faults treated in the thesis are broken rotor bars and broken end rings. Another fault that is encompassed by the modeling approach is diminished air-gap in front of one or more stator and/or rotor phases. Although straightforward to consider, it is unlikely in practice that more than one fault would occur at a time. Therefore, the modeling technique is explicitly used for arriving at two different models, where it is assumed that there is an electrical fault in either one stator phase or in one rotor phase. Interestingly, complexity of the resulting models is comparable to that of conventional models for healthy induction machines. The two models are validated against data collected from real-life induction machines with faults on purpose imposed upon them, with excellent agreement between the simulated and measured signals. Secondly, system properties of the two derived models, such as dissipativity, observability and controllability, are analyzed. It is shown that neither a stator nor a rotor fault alters these system attributes, except for very special cases. Furthermore, when a healthy induction machine is subject to a constant load and a perfectly balanced stator voltage source, it is proven that the rotor speed cannot be stabilized if the voltage has a spectral content comprising more than two distinct frequencies. Thirdly, the two developed models are used for state and parameter estimation in induction machines. Two kinds of observers are designed. The first one aims at robust stator flux estimation for healthy induction machines. It is shown that exponential convergence of the estimation error can be guaranteed, at a rate independent of the rotor speed. Therefore, the design possesses robustness against disturbance in rotor speed measurements, however with an upper bound on attainable performance. The second observer methodology provides a generalized linear-quadratic optimal framework for linear complex-valued time-varying systems, where an arbitrary, possibly time-varying, guaranteed convergence rate can be assigned. The methodology is thereafter used in a new parameter estimation scheme, both for healthy induction machines and those subject to stator faults, in order to estimate all electrical parameters on-line, under very mild assumptions. Specifically, the scheme can be used for fault detection. Simulations validating the methodology are provided.