Abstract:Tennessee Eastman process (TEP) data have the characteristics of high latitude and high coupling,so it is difficult to extract the data characteristics.In order to further improve the recognition rate of fault monitoring in the process industry system,1D-DenseNet was combined with deep separable convolution (DSC),the efficient feature extraction ability of DenseNet was used,and DSC was used to reduce the calculation parameters,so as to improve the diagnosis efficiency.To provide fault monitoring mode based on DSC-DenseNet,the data were normalized and random seeds were added to avoid over-fitting.Then the processed results were used as the input of DSC-DenseNet for feature extraction,and the output results were transmitted to the full connection layer for fault classification.Finally,the accuracy test was carried out on the TEP dataset.The results show that the method based on DSC-DenseNet can be used to effectively distinguish the fault types,and the accuracy of fault classification reaches 98.8%.It is proved that DSC-DenseNet has better fault identification effect than traditional DenseNet.