文献类型：期刊文献

英文题名：Plant-Wide Distributed Adaptive Probability Density Analysis Method for Incipient Fault Detection

作者：Tao, Yang[1];Chen, Min[1];Shi, Hongbo[1];Song, Bing[1];Tan, Shuai[1];Ding, Weichao[1]

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

年份：2025

卷号：74

外文期刊名：IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

收录：;EI(收录号：20251118045288);WOS:【SCI-EXPANDED(收录号:WOS:001449658500026)】；

基金：This work was supported in part by the National Natural Science Foundation of China under Grant 62473156,Grant 62473154, Grant 62473155, and Grant 62273147; in part by Shanghai Chenguang Project under Grant 21CGA37; and in part by National Natural Science Foundation of Shanghai under Grant 19ZR1473200.

语种：英文

外文关键词：Adaptation models; Feature extraction; Fault detection; Data mining; Process monitoring; Data models; Probability; Heuristic algorithms; Analytical models; Correlation; Distributed adaptive probability density analysis; incipient fault detection; multivariate statistics; plant-wide; process monitoring

摘要：In plant-wide process, the incipient fault may be ignored due to its small amplitude. In order to realize the incipient fault detection for the large-scale time-varying process, a novel distributed adaptive probability density analysis method is introduced in this article. First, a data-driven process decomposition and subblock division strategy is proposed, and the adaptive modeling sample set is constructed based on the online data in each subblock, which converges the information related to the current operating state. Afterward, the key latent variable which contributes to the online variation is selected for the process modeling, and a probability density based monitoring indicator is constructed for the incipient fault detection, which can effectively identify the unusual data distribution in the online testing samples. After the subblock modeling and monitoring, the Bayesian fusion strategy is introduced for the subblock decision integration. Finally, two case studies are used to illustrate the advantages of the proposed method.