Dissertation
Dissertation > Industrial Technology > Chemical Industry > Pharmaceutical chemical industry > Antibiotics manufacture > Penicillin and its derivatives

Batch Process Modeling and Fault Detecting Based on Support Vector Machine

Author LiangQin
Tutor JiaMingXing
School Northeastern University
Course Control Theory and Control Engineering
Keywords Support Vector Machine Batch Process Penicillin Fermentation FaultMonitoring
CLC TQ465.1
Type Master's thesis
Year 2012
Downloads 4
Quotes 0
Download Dissertation

The batch process production is widespread used in the field of chemicals and pharmaceuticals that related to national economy and people’s lives. In order to improve the safety and maintainability of the batch process, it is urgent need to establish process monitoring system to monitor the production process. However, for these processes, fault monitoring methods based on multivariate statistical and mechanism model are both exist some defects. Therefore, this article proposes a new statistical process monitoring algorithm based on support vector machine (SVM) for basic characteristics of batch processes, to improve the ability of process monitoring.Firstly, the article introduces development status of multivariate statistical monitoring, basic concepts of process monitoring and characteristics of batch process, stating the background and significance of the SVM for batch process. Secondly, the article states shortcoming of two multivariate statistics arithmetic used extensively in modeling and proposes monitoring method for batch process based on support vector machine. The basic idea of the method is first to classify the process variables of batch process that is to determine the input variables that affect a number of different output variables. Then it creates multiple output models and each model training error follows a normal distribution. Finally, calculate error control limits to conduct fault detection. In order to improve the sensitivity of the process monitoring, aiming at the problems that single output variable error of detection samples may be relatively small and fault diction effect is not obvious, this article gives the calculation method of SPE statistic obeying chi-square distribution of multiple output errors and calculate SPE control limit. When the process occur abnormalities that are correlation between the process variables change or some process variables occur mutation, fault can be detected by monitoring the SPE values and has good detection result.Finally, using monitoring method for batch process based on SVM, this article constructs modeling and monitoring for penicillin fermentation process based on Pensim simulation platform. The simulation results show the proposed method can well reflect nonlinear characteristic during fermentation, reduce the computational complexity of online monitoring, has good monitoring capability and prove the effectiveness of the algorithm.

Related Dissertations
More Dissertations