A Novel LSTM-1DCNN-based Deep Network for Fault Diagnosis in Chemical Process

Fault diagnosis in chemical process can provide equipment maintenance personnel with a basis for decision making and ensure the safe operation of production processes. Deep learning has shown unique strengths and promising for fault diagnosis in chemical process. In practical production, the occurrence of a fault can be uniquely determined by its temporal and spatial information. However, most deep learning models based on a single network are difficult to guarantee the accuracy meet the requirements due to the limited capabilities of feature extraction caused by the single network structure. To deal with these problems, a novel deep network model combining with long short-term memory network (LSTM) and one-dimensional convolutional neural network (1D-CNN) is proposed in this paper to improve fault diagnosis performance for chemical process, named DLCM in short, which uses LSTM layers to obtain the temporal features and 1D-CNN layers to extract the spatial features sequentially. Finally, the fault identification outputs are acquired by feature fusion through the fully connected layers. A benchmark chemical process is used to validate the effectiveness of the proposed DLCM. The results show that the DLCM achieved 99.23% classification accuracy, and the diagnosis results are better than other traditional methods based on the single network and combined network.