Research On Monitoring Of Fermentation Process Based Nonlinear PLS Method

Fermentation process is one of the important processes in the modern industrial process, and has been widely used in higher value-added application, such as Pharmaceuticals, fine chemicals and deep processing of agricultural products. Fermentation process is a classical batch process, which exhibits properties including nonlinearity, fast-speed dynamic respond, distinct time-variant feature, and multiple phases, etc. The online process monitor is considered as a necessary to the security of process and quality of products, since the mechanism of fermentation process is complex and sensitive to the environment.According to strong nonlinearity and multi-stage characteristics of fermentation process, research the application problem of Statistical process monitoring method based on Multi-way Partial Least Square (MPLS). The main research included in this paper is as follows:(1) Expand the Neural Networks PLS, a Multi-way Neural Networks Partial Least Square (MNNPLS) method is proposed. The method is more accurate description the dynamic properties of the process, and is used on penicillin fermentation process monitoring. Through the monitoring results we can see that the real time is higher.(2) A novel multi-stage PLS method is proposed. The method using ISODATA dynamic clustering algorithm,process data was automatically divided into several operation stages, then the Dynamic Time Warping (DTW) algorithm is used to synchronize the time length of all stages of data, and depicts the dynamic characteristics of each stage by different mechanism. The method applied to penicillin fermentation process, the entire process was automatically divided into four operation stages conform to the actual working conditon. Monitoring results shows that the accuracy and timeliness have distinctly improved compared to MPLS and MNNPLS method.(3) MPLS, MNNPLS and multi-phases methods are each used to realize on-line monitoring and quality prediction for the actual fermentation process of Coenzyme Q10.The results show that multi-phases and MNNPLS methods are better than the traditional MPLS methods in fault detection and quality prediction, and the multi-stage method is the best.