Phase-based Batch Process Monitoring

In recent years, with the continuous development of process industry, the requirement of modern social for products with various varieties and high quality has become more and more urgent. Batch process has been widely concerned because of its ability to produce low-volume and high value-added products and has been widely used in medicine, polymer reactions, dyes, metal processing, biological products and other fine chemical production. As an important production method in modern industry, the safety and reliability of the batch production process as well as the quality of its final products have become the focus of attention. When establishing the monitoring model, multivariate statistical analysis methods only require the process data collected under normal working condition, and they have significant advantages in dealing with highly coupled high-dimensional data, so they are becoming more and more attractive for researchers. The application of multivariate statistical analysis methods in statistical modeling and on-line monitoring for batch process has become a widely studied topic. Different from the continuous operation, the correlations among variables and process dynamics in batch process are more complicated. A batch operation is divided into several phases according to the change of process dynamics and data correlation structures. Therefore, it is more challenging to conduct multivariate statistical analysis in multiphase batch process. It should focus on the latent process dynamics of each phase while monitoring the whole operation status.In this thesis, the research status in the application of multivariate statistical analysis for batch process is introduced, at the same time the multiphase property of batch process is studied deeply. Then a new method for phase-based batch process on-line monitoring and fault diagnosis is proposed to solve practical production problems. The main work of this paper is as follows:(1) Introduce the multivariate statistical analysis techniques for batch process, and study the basic principles and data preprocessing methods of them intensively. Find out the disadvantages of every common data preprocessing method through the comparisons between them. Then a regularized batch wise unfolding method is proposed, which based on the batch wise unfolding while taking the time-lag into account. So this new method can avoid the requirement of large amount of data when modeling and maintain the benefits of batch wise unfolding.(2) For the phase transition characteristic of multiphase batch process, the concept of fuzzy phase is introduced before conducting K-means phase division algorithm, and a phase based soft phase division method is proposed, which divides the batch into different phases and transition regions between neighboring phases according to the changes of process operation characteristics along time. Before the on-line monitoring and statistical modeling for multiphase batch process, data is preprocessed by regularized batch wise unfolding algorithm firstly, and then the statistical models for each phase and transition region are built based on the soft phase division.