Online Proactive Health Monitoring Methods Using a Neural Network for DC Link Capacitors in an AC/DC/AC Converter

As capacitors age, the likelihood of failure increases. Electrolytic capacitors typically fail when their electrolyte layers evaporate, resulting in an open-circuit fault which is minimally dangerous to other circuit components in the system. However, in recent years, film capacitors have become a common choice for DC-link capacitors in rectifier and inverter circuits, owing to their superior durability and lifetime to electrolytic capacitors. The drawback of film capacitors is that their failures are typically short-circuit faults which occur when the capacitor dielectric wears through. Short-circuit faults create high current stresses on neighboring circuit components such as MOSFETs and IGBTs, which can in turn propagate component failures throughout the entire system. Therefore, capacitor health monitoring is of utmost importance for DC-link capacitors. In this thesis, the current state of capacitor health monitoring is analyzed. Two novel capacitor health monitoring techniques which utilize an artificial neural network (ANN) are then proposed to estimate the current capacitance of a DC-link capacitor. The present capacitance is compared to the initial capacitance to gauge the health of the capacitor. The first method uses the RMS values of the DC-link voltage ripple and load current. The second method utilizes the magnitude of the line frequency harmonics present in the DC-link voltage. The two methods are verified via simulation. Finally, the potential for further research and improvements are discussed.