Feature Fusion Based on Locally Linear Embedding in Fault Diagnosis of A Single-Phase PWM Rectifier

To advance the traction drive system reliability and the stability of overhead contact system, reduce the railway traffic accidents, ensure the safe operation of electric locomotives. For the single-phase Pulse Width Modulation (PWM) rectifier with high failure rate in electric locomotive traction drive system, a fault diagnosis method based on Locally Linear Embedding (LLE) algorithm in manifold learning theory is proposed for fusion of high dimensional features. The wavelet packet energy feature is extracted from the fault signal by wavelet packet transform, and as the high-dimensional feature (HDF). And then the low-dimensional feature which can be used to describe the diagnosis feature is obtained by using the LLE algorithm in the manifold learning algorithm to eliminate redundant information in HDF. In the simulation, it is verified that the final diagnosis result for the method is 95.33% when the load direct current (DC) voltage changes from 2000V to 2800V. The results show that this method has the advantages of good robustness, high fault diagnosis precision, simple implementation and no need for complicated parameter calculation.