Intermittent Fault Diagnosis Method Based On Graph Convolutional Neural Network

Intermittent faults are widespread in various systems, which often greatly reduce the mean time to failure of the system, lead to the replacement of normal components, and increase unnecessary maintenance and support costs. The research of intermittent fault diagnosis methods is of great significance for improving equipment reliability, reducing system maintenance and support costs, and improving the level of intelligent manufacturing. At present, various machine learning methods have received more and more attention in the field of fault diagnosis, but the research on intermittent fault diagnosis based on graph convolutional neural network is less carried out. Compared with convolutional neural networks, graph convolutional neural networks are good at processing graph data which is more widely existing. Based on this, this paper decides to construct an intermittent fault diagnosis model based on graph convolutional neural network for intermittent fault localization. This paper proposes a method to convert the intermittent fault voltage waveform data into graph data. The original waveform is decomposed into six sub-waveforms by wavelet decomposition, and the edge connection is defined by the Pearson correlation coefficient between the sub-waveforms. On this basis, six different graph convolutional network models are built, the network is trained and optimized, and the accuracy of 90.38% is obtained on the test set.