Fault Diagnosis of Synchronous Generator Based on Adaptive Chirp Mode Decomposition Permutation Entropy and Deep Learning

Abstract

Generators are subject to extreme environmental conditions that can cause critical areas to gradually break down, potentially leading to catastrophic failures. This paper proposes a hybrid method for generator fault diagnosis. Firstly, adaptive chirp mode decomposition (ACMD) is applied to decompose the vibration signal into five intrinsic mode function (IMF) components. Then, the permutation entropy (PE) of each IMF is calculated to construct the feature vector. The deep learning part of the proposed method uses convolutional neural network (CNN) as a classifier to recognize different faults. Finally, the visualization result using t-Distributed Stochastic Neighbor Embedding (t-SNE) is presented. The result of classification suggests that the method proposed in this paper realizes fault diagnosis with the accuracy of 98%, which has a higher recognition rate than other methods mentioned in this paper.