Kieran Levin with advisor A. Dominguez-Garcia
Power electronics for automobile drive systems and solar systems may fail before the critical systems that depend on them. To help mitigate the effects of power electronics-caused system failures, a fault detection filter has been implemented to detect both hard and soft faults in the power supply. This filter allows reduced system operation or maintenance by determining how much load the supply can power while staying within specifications.
A software solver has been developed for the observer-based fault detection system implemented
last year. The system is built around a low-cost digital signal processor connected to a dc-dc switching power converter. A detection filter was developed that requires a smaller number of measurements of the system
state—only twice per cycle—to get accuracies similar to the previous approach which requires
over ten samples per cycle. Furthermore the detection filter uses a forward-looking model which takes current measurements and predicts what the system state will be at the next sample. This is then compared to the
sample and generates an error signal which is fed into a signature detection filter.
The signature detection filter looks at the error signals generated independently for each converter state to detect and identify faults. This allows the detection filter to detect specific component failures in each section of the converter using a limited number of measurements. The detection filter can now detect
and identify faults in each active and passive component in the converter, including high resistance
in parts and connections, component drift, and failed components. Hundreds of runs of the converter operation over its entire output operating range have been gathered to qualify observer performance at detecting
faults over its operating range and are currently under analysis.
This research is supported by Grainger Fellowships.