Date of Award
Doctor of Philosophy (PhD)
Electrical and Computer Engineering
Yaz, Edwin E.
Schneider, Susan C.
Multiple model estimation is a versatile technique that has been applied in a large variety of adaptive estimation problems. In this technique, several estimators, each designed around a possible model of the system, are used in parallel, and the estimator that most closely models the true system is determined using Bayes’ Rule. In this dissertation, convergence properties of the multiple model estimation algorithm are investigated, including a proof of convergence and factors that influence convergence time. In addition, the multiple model estimation algorithm is applied in two parameter estimation problems.The first half of the dissertation focuses on properties of the multiple model estimation algorithm. A proof of convergence is developed which is compatible with any estimator that satisfies three conditions. The multiple model estimation algorithm is applied to the identification of the stator resistance of a permanent magnet motor, and the results are compared to those produced by the extended Kalman filter. This application demonstrates the advantage of preserving the linearity of a system in the parameter identification problem. In addition, simulations involving the identification of the fault rate of a sensor demonstrate that convergence time is related to both the signal-to-noise ratio of the system and the quantization of the parameter space.The second half of the dissertation incorporates these results to design a novel filter that can estimate the states of a system in the presence of intermittent actuator faults with an unknown fault rate. The proposed filter is a modification of the Kalman filter that incorporates intermittent actuator faults of a known fault rate. The filter is extended to the case of unknown actuator fault rates by using multiple model estimation to identify the unknown fault rate. The simulation of a DC motor with intermittent actuator faults demonstrates that the proposed filter simultaneously identifies the unknown fault rate and estimates the system states. In addition, a simple compensation control technique is proposed that can maintain the average system performance in the presence of intermittent actuator faults.