Key Components Incipient Fault Diagnosis of NPC Inverter Based on Multi-domains Feature Extraction and Deep Belief Network

Abstract

Aiming at the problems that distinctions of fault characters are not obvious and fault types are complex when occurring on key components IGBT and DC-Link capacitor of three-level neutral point clamped (NPC) inverter. This paper introduces a new fault diagnosis method based on multi-domains feature extraction and deep belief network. Firstly, different fault types of three line voltage data on AC side are sampled. Then, intrinsic mode functions (IMFs) component of voltage signal in frequency domain are obtained by variational mode decomposition(VMD), wavelet packet energies and their energy entropies of IMFs in time-frequency domain and statistical parameters in time domain of original fault signals are calculated. Frequency-time frequency-energy domains and time domain features are constructed. Finally, these fusion feature vectors were input into the deep belief network (DBN) to train fault diagnosis model, the prime multi-domains feature extraction and DBN model is used for fault classification of key components in three-level NPC inverter.