Fault classification based on variable‐weighted dynamic sparse stacked autoencoder for industrial processes

Abstract

AbstractEffective process monitoring and fault diagnosis are of great significance to the safe operation of industrial processes as production scale increases and production systems become more and more complex. Identifying fault types in complex industries by fault classification can help workers to determine the source of the fault as soon as possible, which is crucial to timely recover work. In this paper, a fault classification method based on a variable‐weighted dynamic sparse stacked autoencoder is proposed. First, considering the dynamic characteristics of process data, the input data are processed dynamically by a sliding window. Then, in the pre‐training stage, the weight of each variable is calculated by Fisher discriminant analysis for the reconstruction of the loss function. The sparse term is added to the loss function so that it can learn the effective representation of data in a harsh environment. Finally, the proposed method is applied to the Tennessee‐Eastman benchmark to evaluate the classification performance. The result shows the superiority of the proposed method.