Ensemble-based Randomized Classifier for Data-driven Fault Diagnosis of IGBT in Traction Converters

Abstract

Insulated gate bipolar transistor (IGBT) is one of the most common but fragile devices in induction motor drives. Fault diagnosis methods for IGBT are developed to increase the system reliability. This paper proposes a diagnosis method for IGBT open-circuit fault based on ensemble-based randomized data-driven method. A novel learning technology named random vector functional network (RVFL) is adopted to train and learn the historical IGBT fault dataset. An ensemble structure is designed to increase the diagnosis accuracy. Finally, time window is defined to illustrate the relationship between diagnostic accuracy and decision time. It is shown that more input information of the classifier, the higher diagnostic accuracy can be achieved, but with more time. By comparing other data-driven methods, the proposed ensemble-based method has a higher diagnostic accuracy within the same decision time.