Research on Failure Prediction Using DBN and LSTM Neural Network

Because the vibration signal data has a high dimension, ordinary fault prediction methods are very complicated for data preprocessing. A fault diagnosis method combining DBN and LSTM was proposed in this paper. This method uses DBN's excellent feature extraction capability to extract feature vectors of time-series fault data, and uses LSTM network to perform fault prediction on the extracted feature vectors, because LSTM network has good adaptability to sequential data. The author used the bearing data of Case Western Reserve University to build a training set and conduct experiments. Experimental results show that this method can reduce the difficulty of data preprocessing and increase the accuracy of fault prediction.