多阶段青霉素生产过程的故障检测研究  

Fault Detection for Multiphase Penicillin Fermentation Process

作　　者：张晓玲 丁腾飞 曹玉苹[2] ZHANG Xiaoling;DING Tengfei;CAO Yuping(School of Intelligent Manufacturing and Control Engineering,Shandong Institute of Petroleum and Chemical Technology,Dongying 257061,China;College of Control Science and Engineering,China University of Petroleum(East China),Qingdao 266580,China)

机构地区：[1]山东石油化工学院智能制造与控制工程学院,山东东营257061 [2]中国石油大学(华东)控制科学与工程学院,山东青岛266580

出　　处：《控制工程》2024年第5期850-857,共8页Control Engineering of China

基　　金：山东省高等学校科技计划项目(J18KA359);大学生创新创业训练计划项目(S202113386045)。

摘　　要：考虑到青霉素发酵是复杂的多阶段动态批处理过程,提出多阶段多向可预测成分分析(multiphase multiway forecastable component analysis,MP-MForeCA)的故障检测方法。该方法基于K均值和密度峰值聚类算法设计阶段划分方法,将多批次的青霉素发酵过程数据划分成不同子阶段,在子阶段三维数据上分别应用多向展开的可预测成分分析(multiway forecastable component analysis,MForeCA)算法,提取各阶段中具有可预测特性的可预测成分,捕捉各阶段的动态变化;最后利用可预测成分和残差分别构建F^(2)和SPE统计量来监控模型。通过与MForeCA和多向主元分析(multiway principal component analysis,MPCA)的实验对比发现,MP-MForeCA方法不管是对青霉素生产中的突变故障还是缓慢变化故障都有更好的检测效果,包括提高检测率和降低误报率,体现了多阶段建模规则的MP-MForeCA方法的优越性。Considering a penicillin fermentation process is a highly complex and batch dynamic process with multiple phases,a multiphase multiway forecastable component analysis(MP-MForeCA)method is presented for penicillin fermentation process fault detection.In the proposed fault detection strategy,a new phase partition scheme based on the combination of K-means and density peaks clustering algorithm is designed to separate the batch fermentation process into multiple phases.In each sub-phase three-way dataset,the multiway forecastable component analysis approach is applied to obtaining the forecastable components for capturing the sub-phase dynamic characteristics.Finally,F^(2) and SPE statistics are calculated by utilizing forecastable components and residuals for monitoring modeling respectively.The experiment results on penicillin cultivation process show that the MP-MForeCA method with multiple monitoring models generates a better fault detection result and has monitoring superiority compared to MForeCA and multiway principal component analysis(MPCA).The fault detection rate is increased and the false alarms are lowered under the implementation of the MP-MForeCA scheme.

关 键 词：可预测成分分析 青霉素发酵 多阶段 F^(2)统计量 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]