Application of EMD Combined with Deep Learning and Knowledge Graph in Bearing Fault

This paper proposes a bearing fault diagnosis method using empirical mode decomposition (EMD), deep learning, and a combination of knowledge graphs to analyze faults from multiple perspectives. Specifically, EMD is used to decompose the vibration signal of the bearing, and the intrinsic mode function (IMF) is input into the layered one-dimensional convolutional neural networks (1D-CNN), which is fed into a Bi-LSTM for the further time series feature extraction after multi-scale convolution and pyramidal pooling module. The classified fault labels connect the graph and the deep learning network, involving detailed fault information. The proposed multi-scale convolutional neural networks (1D-EMCNN) based on EMD, pyramid pooling module, and Bi-LSTM has several advantages: first of all, EMD overcomes the problem of non-adaptive basis functions. For a segment of unknown signal, 1D-EMCNN can automatically decompose according to some fixed modes without any pre-analysis and research to obtain multi-scale observation results. Secondly, inputting the intrinsic mode function into the 1D-EMCNN can effectively extract fault information from multiple directions and avoid the problem of manual feature extraction that requires a lot of prior knowledge and information loss. Finally, the 1D-EMCNN is combined with the knowledge graph to obtain bearing fault-related information, which is used to utilize the search function of the knowledge graph. Experiments show that 1D-EMCNN fault reaches an accuracy rate of fault diagnosis up to 99% under various loads, an accuracy rate of 92% under noise conditions, and indicates detailed information about faults through knowledge graphs.