Model-based failure detection in induction motors using nonlinear filtering.

摘要

In this dissertation, we present a model-based failure detection method for induction motors. Our method addresses two failure modes of the squirrel-cage induction motors; namely, rotor bar failures and stator winding insulation problems. Under the assumption that the stator voltages, the stator currents and the rotor speed are measurable, the induction motor can be modeled as a linear time-varying system. We apply a nonlinear filtering algorithm for failure detection using a discrete-time representation for the system. Failure events are treated as jumps in the parameters of the linear model. These events are random processes and are modeled as a finite state Markov chain. The conditional probability for each operating condition given the observation sequence is obtained by a nonlinear recursive algorithm. Then, the failure of the induction motor can be isolated by the record of conditional probabilities.; To compensate for the variation of the rotor resistance due to thermal effect, we use the extended Kalman filter (EKF) to estimate the real-time value of the rotor resistance. Because speed sensor is a high-cost device, and may not be installed in some compact systems, we use a model reference adaptive system design technique to identify the rotor speed of the induction motor, thereby eliminating the need for a rotor speed measurement sensor in the failure detection scheme.