Batch process monitoring based on multiway principal component analysis
|School||Nanjing University of Technology and Engineering|
|Course||Control Theory and Control Engineering|
|Keywords||batch process Statistical process monitoring Multi-way Principal ComponentAnalysis multi-stage dynamic kernel feature samples|
Batch processes is a important method of production in modern process industry. They are widely used in specialty chemical industry, pharmaceutical industry and food industry. To improve the product quality and ensure the security of the control system, it’s necessary to carry out process monitoring.This thesis mainly studies fermentation process monitoring based on Multi-way Principal Component Analysis (MPCA) method, and some improvements of traditional MPCA method have been made according to the characteristics of batch processes, the main contents are as follows:(1)The theory of Multi-way Principal Component Analysis used in batch processes monitoring is explained. Two different unfold ways of MPCA method are discussed considering the properties of batch processes. The batch trajectory synchronization method is introduced to solve the problem of unequal length of batch process data.(2)In view of the multi-stage properties of fermentation process, a multi-stage MPCA method based on data similarity is proposed. The proposed method firstly clusters the similarity indicators between difference time-slice data matrices of batch process to divides the process into several stages, then MPCA models are builded for each stage. The new method is applied on the monitoring of the penicillin fermentation process, verifying the effectiveness of the method.(3)In view of the dynamic and highly nonlinear characteristics of batch processes,a batch dynamic kernel PCA(BDKPCA) method is studies. Feature samples (FS) theory is applied to BDKPCA to reduce the computational cost. FS is improved according to the data characteristics of batch processes and BDKPCA based on FS is proposed. The FS is performed to obtain the feature subsets with less samples firstly, then BDKPCA is conducted on the feature subsets. Monitoring results show that FS-BDKCPA can reduce computation cost and storage cost of BDKPCA effectively without influences on monitoring performance.