Research And Application Of Statistical Monitoring Algorithm Based On Time-varying MPCA

Penicillin fermentation process is a typical representative of the biochemical reactionsystem, which has the feature of nonlinear strong, complex biochemical mechanism, poorreproducibility, and detecting biological parameters difficultly. Batch processes are fairlymore complex compared with continuous processes, while the quality of the product is verysensitive to the uncertain factor such as condition and external environmental conditions.Therefore, to improve the maintainability of fermentation control system effectively, ensurethe security of the production process and improve the final product quality and economicefficiency. It’s urgent need to carry out performance monitoring and fault diagnosis on itsproduction process.With the high-rapid development of computer control technology, a great amount ofprocess data is collected and stored, which brought new development fortune to multivariatestatistical analysis method based on data-driven. In the batch process, the traditionalmultivariate statistical process monitoring method is based on MPCA (Multi-way PrincipalComponent Analysis) and MPLS (Multi-way Partial Least Square). Combined with thecharacteristics of the fermentation process, process monitoring method based on theconventional MPCA were different degrees of improvement. The improved algorithm isapplied to penicillin fermentation process, the results show that the improved algorithm issuperior than the MPCA.(1)Traditional MPCA algorithm is introduced in a batch process monitoring. CombinedPensim simulation platform, the effects of the monitoring of penicillin fermentation process forthe algorithm is not ideal.(2)According to the problem of strong nonlinear when MPCA algorithm monitors thecomplex batch process, we propose a modeling method of Sub-period MPCA. Firstly, itdivides the batch process into several phases with FCM (Fuzzy c-Mean), then the offlinemodeling method is adopted to the whole batch production process to establish a phasedmodel. Finally, a phased MPCA algorithm based on FCM is applied the online monitoring ofpenicillin fermentation process. The simulation results show that the improved algorithmreduces errors and improves the monitoring system performance.(3)There is a common problem between traditional and improved MPCA algorithm: itneed to change three dimensional data into two dimensional data in the process of modeling,which leads to tedious calculation. Here, we propose a new fault detection method-two-dimensional principal component analysis algorithm (2DPCA). Firstly, The algorithm isdirectly related to the two-dimensional data matrix of the processing of each batch processingFinally,2DPCA was used to model with the covariance average of all the batches, whichconducted the data monitoring for a new batch. Through the research and application of thepenicillin fermentation, the algorithm is superior to the traditional MPCA algorithm in theprocess monitoring effect.