IJCAS

Based on complementary ensemble empirical mode decomposition and deep neural network (CEEMDDNN), a novel diagnosis method is proposed to discover the open-circuit faults of insulated gate bipolar transistors (IGBTs) in single-phase pulse width modulation (PWM) AC-DC rectifier, an important part of traction power supply system of high-speed railway. By virtue of the combination of signal processing and deep learning schemes, the multi-scale feature information for IGBT’s open-circuit fault diagnosis are extracted by regarding the input current of rectifier as the original signal. More clearly, CEEMD is adopted to decompose the original signal to a series of intrinsic mode function components (IMFs). Then, the correlation coefficient algorithm is used to evaluate the correlation between each of the IMFs and the original signal. Further, those IMFs that have been verified to be highly correlated with the original signal are selected as the input eigenvectors of DNN to train the IGBTs fault diagnosis network. Experimental results show that the proposed CEEMD-DNN algorithm is superior to pure DNN, empirical mode decomposition (EMD)-DNN and ensemble empirical mode decomposition (EEMD)-DNN in the sense that within the similar time-consuming situations the fault diagnosis accuracy has been improved from 73.0%, 86.0%, and 89.9% to almost 99.0%. This is mainly owing to the remarkable decomposing performance of CEEMD for complex mixed signal, and the strong feature extraction and learning abilities of DNN in pattern recognition.

Keywords: Complementary ensemble empirical mode decomposition, deep neural network, fault diagnosis, IGBTs, intrinsic mode function components.