Recursive Mixture Factor Analyzer for Monitoring Multimode Time-Variant Industrial Processes

详细信息    查看全文

• 作者：Jinlin Zhu ; Zhiqiang Ge ; Zhihuan Song
• 刊名：Industrial & Engineering Chemistry Research
• 出版年：2016
• 出版时间：April 27, 2016
• 年：2016
• 卷：55
• 期：16
• 页码：4549-4561
• 全文大小：703K
• 年卷期：0
• ISSN：1520-5045

文摘

A critical issue for real-time monitoring of industrial processes with data-based methods is how to adjust the constructed model during monitoring procedure so as to adapt to the change of the process. Traditional latent variable models like factor analyzers are static models and are simply based on the single Gaussian assumption. Therefore, when one comes to multimode and time-variant process conditions, conventional strategies become cumbersome. In this work, a recursive mixture factor analyzer is proposed for multimode time-variant process modeling and monitoring. The developed model with a Bayesian mechanism can automatically select and update the Gaussian components during modeling. Furthermore, a corresponding monitoring mechanism is proposed so that the recursive model can also elegantly adjust parameters according to the newly incoming data and effectively employ more components for new operating modes. Feasibility and efficiency of the proposed method are illustrated through two case studies.