A Novel Prognostics Scheme for Nonisolated DC-DC Converters Using Voltage Features

Abstract

DC-DC converters have been widely used in various industrial systems. Accurate performance prognostics of their remaining useful performance (RUP) can effectively avoid the occurrence of faults. This paper proposes a performance prediction method based on the output signal of the circuit, which only needs to monitor the output response of the circuit without monitoring the nodes inside the circuit. The RUP can be predicted by continuous monitoring of the output voltage. Particle filter algorithm, as one of the algorithms often employed in prediction, is found to have the biggest problem of particle degradation, which will reduce the diversity of particles and then affect the final prediction accuracy. This method monitors circuit degradation by collecting historical degradation data. In addition, the kernel smoothing algorithm is integrated into the particle filter algorithm to ensure that the particle variance unchanged with the circuit performance prediction model during the recursive propagation process. The model can be updated after obtaining new measurements. The analysis of two DC-DC converter circuits shows that the proposed prognostics scheme has good prediction accuracy for nonisolated DC-DC converters.