Feature learning of nonlinear process fault detection detection based on SRB-SCAE

In recent years, fault detection and diagnosis methods based on deep learning have attracted more and more attention. In this paper, a fault monitoring method for sparse convolution autoencoder based on selective residual blocks(SRB-SCAE) is proposed. First, the sparse AE can obtain the sparse representation of the data, and the convolution AE can be used to extract the temporal features between the processed data. Secondly, the selective residual block can increase the attention to jump connection to improve the training accuracy. In addition, the reconstruction error between input and output is used as a statistic and the control limit is determined by kernel density estimation. Finally, the feasibility and effectiveness of this method are verified in Tennessee-Eastman benchmark process.