化工学报 ›› 2022, Vol. 73 ›› Issue (3): 1300-1314.DOI: 10.11949/0438-1157.20211294

• 过程系统工程 • 上一篇    下一篇

基于联合典型变量矩阵的多阶段发酵过程质量相关故障监测

高学金1,2(),何紫鹤1,2,高慧慧1,2,齐咏生3   

  1. 1.北京工业大学信息学部,北京 100124
    2.计算智能与智能系统北京市重点实验室,北京 100124
    3.内蒙古工业大学电力学院,内蒙古 呼和浩特 010051
  • 收稿日期:2021-09-06 修回日期:2021-10-26 出版日期:2022-03-15 发布日期:2022-03-14
  • 通讯作者: 高学金
  • 作者简介:高学金(1973—),男,博士,教授, gaoxuejin@bjut.edu.cn
  • 基金资助:
    国家自然科学基金项目(61803005);北京市自然科学基金项目(4192011);山东省重点研发计划项目(2018CXGC0608)

Quality-related fault monitoring of multi-phase fermentation process based on joint canonical variable matrix

Xuejin GAO1,2(),Zihe HE1,2,Huihui GAO1,2,Yongsheng QI3   

  1. 1.Department of Information, Beijing University of Technology, Beijing 100124, China
    2.Beijing Key Laboratory of Computational Intelligence and Intelligent System, Beijing 100124, China
    3.School of Electric Power, Inner Mongolia University of Technology, Hohhot 010051, Inner Mongolia, China
  • Received:2021-09-06 Revised:2021-10-26 Online:2022-03-15 Published:2022-03-14
  • Contact: Xuejin GAO

摘要:

为考虑发酵过程的质量变量和动态特征对于阶段划分的影响,提出了一种基于联合典型变量矩阵的多阶段发酵过程质量相关故障监测方法。首先,将历史三维数据沿批次方向展开,对每个时间片矩阵进行典型相关分析(canonical correlation analysis, CCA),得到融合过程变量和质量变量信息的联合典型变量矩阵,对其进行K均值聚类,实现基于静态特征的第1步划分;然后采用慢特征分析(slow feature analysis, SFA)算法提取表征过程动态性的慢特征,对其进行聚类实现第2步划分。最后综合分析两步划分结果,将生产过程划分为不同的稳定阶段和过渡阶段,并在划分的子阶段中分别建立CCA监测模型进行质量相关故障监测。该方法通过静态和动态特征的变化实现两步划分,准确区分强动态变化与阶段切换,有效提高质量相关的故障监测模型精度。青霉素仿真平台与大肠杆菌实际生产数据验证了所提方法的可行性和有效性。

关键词: 发酵, 间歇式, 多阶段, 联合典型变量矩阵, 典型相关分析, 故障监测, 算法

Abstract:

In order to consider the influence of the quality variables and dynamic characteristics of the fermentation process on the stage division, a multi-phase fermentation process quality-related fault monitoring method based on the joint canonical variable matrix is?proposed. Firstly, the 3D data were unfolded along the batch direction. Canonical correlation analysis (CCA) was performed on each time slice matrix to obtain the joint canonical variable matrices, which were equipped with process variables and quality variables, and K-mean algorithm was used to realize first partition. Then slow feature analysis (SFA) was used to extract the slow characteristics of the process dynamics, and the K-mean algorithm was used for the second partition. Finally, the production process was divided into different stable phases and transition phases through the comprehensive analysis of the two-step division results. The CCA monitoring model was established in each phase after partition for quality-related fault detection. According to the changes of static and dynamic characteristics, this method can accurately distinguish the strong dynamics and the phase boundaries by a two-step partition, also e?ectively improve the accuracy of quality-related fault monitoring. The feasibility and effectiveness of the proposed algorithm were illustrated by a penicillin simulation platform and an industrial application of E. coli fermentation, respectively.

Key words: fermentation, batch-wise, multi-phase, joint canonical variable matrix, canonical correlation analysis, fault monitoring, algorithm

中图分类号: