Fault Identification in DC-DC Converters Using Support Vector Machines with Power Spectrum-Based Features

The capability to detect and identify faults within power electronic converters can be crucial for many applications. This paper presents a supervised machine learning approach to accurately and reliably diagnose faults in a dc-dc converter system. The proposed algorithm employs support vector machine classification, and utilizes features related to the power spectrum of the converter input current for identifying the converter condition. Simulation results are presented that show the statistical performance of the proposed method under random variations.