Studies On Batch Process Monitoring Based On Multi-phase MAR-PCA

Batch process, one of the most common process industries, has been widely usedin the production and preparation of pharmaceutical, food, dyes, spice, andbiochemical products for its characteristic of flexibility. In recent years, because of theincreasing demand for rich varieties, different specification, and high quality products,batch process with small batch production, high value-added gains comprehensiveappliance in industrial production. Furthermore, to improve the security andmaintainability of batch process and the quality of products, process monitoringsystem is imminently needed to monitor the process.In order to build the PCA based monitoring model, the information in historicaldata is mined. From the perspective of solving practical problems, the characteristicsof multi-phase and dynamic of batches process are deep studied which can besummarized as follows:(1) Phase division method based on improved AP clustering.For batch process, to divide precisely operation phases of batch process using APclustering method, and the preferences which are the inputs of AP clustering isdetermined based on Silhouette index. This method can get the precise division resultwithout a priori knowledge of the process. Meanwhile, we can achieve the effect oflocal linearization phases.(2) Proposed of MAR-PCA monitoring methodIn batch process, Most of the process variables present a dynamic characteristicwhich is caused by the existence of delay characteristic, closed-loop control systemand disturbance. Because variables measured are highly correlated, data obtainedduring the process are information poor.To monitor the process of dynamic characteristics, MAR-PCA method isproposed. Firstly, MAR model is build based on historical data. Then the residuals ofMAR are used to build PCA monitoring model. The validity of proposed method istested by Monte Carlo numerical experiment.(3) Research on multi-phase based MAR-PCA methodIn practical industry process, both multi-phase and dynamic should beconsidered. So, the multi-phase based MAR-PCA method is researched. AP clusteringis firstly applied to divide operation phases of batch process. Then, in each phase, MAR-PCA method is applied to eliminate the characteristic of dynamic. When onlinemonitoring, to overcome the defect of AR model-based fault detection, normalhistorical data used for modeling is introduced to calculate AR scores. A case studyfrom a simulated fed-batch penicillin cultivation process indicates the efficacy ofapproach.(4) Research on E. coli fermentation process online monitoringA GUI monitoring interface based on Matlab is developed to apply the proposedmethod in this article to biochemical-pharmaceutical factory in Beijing whosefermentation production is interleukin-2. Through the contrast experimenteffectiveness of proposed method is verified.