Fault diagnosis method based on a hybrid convolutional neural network

This paper proposes a fault diagnosis method based on slow feature analysis and a hybrid convolutional neural network. First, the slowly changing features of raw data are extracted by the slow feature analysis technique, which transforms the time series data into slow feature components that vary slowly but still carry significant dynamic information. Then, in order to better distinguish the feature differences in industrial process fault data, the slow feature data is transformed into 2D images based on the Gramian angular field graph transformation method. Next, a hybrid convolutional neural network model for fault classification is proposed, which can simultaneously extract features from 1D time-series data and 2D image features. Finally, experimental simulations are carried out on the Case Western Reserve University (CWRU) dataset and Benchmark Simulation Model 1 (BSM1) dataset, respectively. Simulation results show that the proposed method has better classification accuracy compared with some other existing methods.