Document Type

Article

Publication Date

4-20-2022

Abstract

Data-driven methods have been widely used in process monitoring and fault detection areas. Despite that, most of these data-driven approaches are extremely specific, where the developed models can only be applied in the given process with the selected variables. The lack of generalization potentially limits its applications. This work mainly focuses on developing a model with better generalization ability and transferability for process data analytics. We proposed a transitional invariant feature extraction method using convolutional neural networks, which can learn features regardless of the size or sequence of the input. Such an approach allows the transfer learning so that the trained model can be reused on different variables or different processes. To verify our approach, a pre-trained model from the Tennessee Eastman process data can be applied to the monitoring tasks with additional variables and can even be adapted to a totally different process. The results illustrate the superior generalization ability of the proposed method.

Publication Source (Journal or Book title)

Industrial and Engineering Chemistry Research

First Page

5202

Last Page

5214

Share

COinS