IGBT Fault Diagnosis Based on Modal Time-frequency Diagram and Residual Parallel Attention Neural Network

Aiming at the limitation of the time-frequency analysis method for the open-circuit fault diagnosis of voltage-source-controlled static synchronous compensator(VSC-STATCOM), which has feature leakage, a fault diagnosis method using modal time-frequency diagrams to describe the open-circuit fault characteristics of IGBTs and combining with the residual-attention mechanism neural network is proposed. Firstly, the data set of 22 types of fault samples of VSC-STATCOM module under different operating conditions is obtained by Matlab/Simulink simulation. Secondly, the fault signals are decomposed into modal signals with the best number of modes using variational modal decomposition(VMD) and according to their centre frequencies. Finally, the modal signals are generated into modal time-frequency diagrams using synchronous extraction of wavelet transform(WSET) with pseudo-colour coding, and then combined with residual parallel attention mechanism neural network(Res-PAM) to identify the IGBT fault types. Taking VSC-STATCOM as the research object, based on training samples of 30 and test samples of 70, the fault recognition rate can reach 99.65%, and also 98.61% in 10 dB noise environment. The accuracy of the proposed method in this paper is also improved by comparing it with other time-frequency diagram methods.