Model-based fault diagnosis for voltage source inverters

This thesis introduces a fault detection and isolation (FDI) method for diagnosis faulty semiconductor switches of a pulsewidth modulated (PWM) voltage source inverter (VSI). This method analyses the pole voltage signals in a time-free domain called voltage space. For a healthy inverter, the projection of the state transitions in the voltage space results in a cubic pattern. It shows that each switch fault changes the voltage space pattern (VSP) uniquely that allows isolating the faulty switch. The mathematical model of the inverter can predict these patterns. Monitoring the transition vectors in the voltage space and comparing to the fault signatures can diagnose the faults. An experimental set-up has been designed and implemented to simulate the switch faults in a three-phase inverter. Simulations and experimental results validate the efficiency and consistency of the VSP-based FDI. The fault detection time is within only one PWM carrier period which is significantly faster than current-based conventional FDI methods. It shows that the VSP-based FDI method can overcome the FDI problem of multilevel multiphase inverters with only one voltage detector per phase leg. It is sufficiency fast to be integrated with a fault tolerant control to maintain maximum efficiency of the inverter in the presence of a faulty switch.DOCTOR OF PHILOSOPHY (EEE