Studies On PCA Methods Towards Monitoring Of Multi-phase Batch Process

Principal Component Analysis (PCA) is a data-driven method of multivariate statistical analysis. Because PCA does not depends on inner mechanism and takes account of interactions among variables, it becomes one of the promising fields to realize the process monitoring in multi-phase batch process.The principle of PCA and its limitations in multi-phase batch process are further investigated in this paper, then an improved PCA method is presented. Using data from normal and abnormal batch, some experiments on penicillin fermentation process are performed, whose results show that this approach is more effective and feasible than conventional methods.This paper initially analyzes characteristics of batch process and goes into the PCA based monitoring methods, followed by a stage-base sub-PCA modeling scheme by means of K-means clustering method to build different phases. To overcome the limitations, this paper proposes an improved PCA monitoring method which merge outlier detection and process variables pattern fitting algorithm. Thereafter a scheme of complete monitoring algorithm is introduced. The improved method is proven to be more consistent with the application environments of PCA as well as able to enhance the performance of monitoring effectively. Eventually, a desirable result is obtained through an application example on penicillin fermentation process. A basic application software system in VB6.0 is developed to realize on-line process monitoring towards the penicillin fermentation process.The research shows that the proposed approach is capable of overcoming flaws of stage-based sub-PCA modeling method as well as depending less on demands of data from process and prior knowledge. Simulation experiments well demonstrate the validity of this contribution. Therefore, more applications are expected in the future.