Learning the Multi-Grained Process Attributes for Industrial Fault Classification

The variability of abnormality not only stems from abnormal patterns, but also from process attributes, such as operation mode, material fluctuation. Leveraging these attributes can enhance diagnostic performance and be helpful at fault characteristic understanding. This paper proposes a novel fault diagnosis method which automatically ably captures the multi-grained attributes of the industrial process when learning fault features. Further, this paper also extends it into a semi-supervised manner. By incorporating limited prior knowledge, our model gives a more meaningful explanation of the diagnostic procedure. The experimental results of a multiphase flow process demonstrate the effectiveness of the proposed methods.