文献类型：期刊文献

英文题名：Deep discriminative feature learning based on classification-enhanced neural networks for visual process monitoring

作者：Wang, Wenjing[1];Yu, Zhenhua[1];Ding, Weichao[2];Jiang, Qingchao[1]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China

年份：2024

卷号：156

外文期刊名：JOURNAL OF THE TAIWAN INSTITUTE OF CHEMICAL ENGINEERS

收录：;EI(收录号：20240715539140);WOS:【SCI-EXPANDED(收录号:WOS:001176892200001)】；

基金：The authors gratefully acknowledge the support from the following foundations: National Natural Science Foundation of China under Grants 62322309 and 61973119, Shanghai Rising-Star Program under Grant 20QA1402600, Nature Science Foundation of Shanghai, China, under Grant 23ZR1414900, and Shanghai Pilot Program for Basic Research Under Grant 22TQ1400100-16.

语种：英文

外文关键词：Discriminative feature learning; Visual process monitoring; Deep neural network; T -stochastic neighbor embedding

摘要：Background: Process monitoring plays an important role in ensuring plant safety and product quality. Among various monitoring methods, visual process monitoring provides an intuitive indication of process status, and is gaining increasing attention. However, the feature learning, which is the core of the visual process monitoring, has not been well discussed. Methods: A Deep Discriminative Feature learning-based Supervised Neural Network (DFNN) is proposed for effective visual process monitoring. The DFNN is composed of an Extended Stacked Autoencoder (ESAE) and a Feedforward Neural Network (FNN). The ESAE augments system data into a novel feature subspace, serving as the FNN's input. The DFNN imposes class center and classification constraints on data to extract discriminative features. Concurrently, a t-Distributed Stochastic Neighbor Embedding-based Neural Network (t-SNE-based NN) maps these deep features into a 2D space, facilitating high-dimensional data visualization and intuitive presentation of operational status. Significant findings: Case studies on the Tennessee Eastman benchmark process and a wastewater treatment process are carried out. Comparison results show that the proposed method can provides more accurate fault classification results than some state-of-the-art methods.