Fault Diagnosis Based on RseNet-LSTM for Industrial Process

Aiming at the problems that conventional data-driven diagnosis methods are difficult to adaptively extract effective features from industrial process data, and do not make full use of the time series characteristics of process data, in this paper, a fault diagnosis method based on residual convolutional neural networks and long short-term memory networks (ResNet-LSTM) is developed. Firstly, the local spatial features of process data are captured by the deep residual convolution network. Then, the time series characteristics of process data are extracted by LSTM. Finally, the output of the fault category is performed through the softmax classifier. This method can extract features adaptively, more fully extract features in time series fault data, and effectively reduce the difficulty of deep neural network training. The benchmark Tennessee-Eastman (TE) process is used to validate performance of the proposed method. The ResNet-LSTM model is compared with the CNN, LSTM, ResNets, CNN-LSTM models, and the experiment results show that the ResNet-LSTM method achieves better performance.