Fault Diagnosis Of Batch Process Based On Sub-phase MPCA-SVM

Due to the advantages of high additional value, small batch, multi species, product seriation, etc., the technique of batch process has been widely used in many fields, which are closely related to people’s life, such as the industry of biopharmaceutical, food, dyestuff, spice, etc.. In order to ensure the quality of final product, and decrease the loss of production process, building a safe and steady system is the first thing to deal with, thus it is particularly important to conduct a research on on-line fault diagnosis of batch process. But batch process has the character of mulit-phases, strong-nonlinearity, etc., which pose great challenges for the study of on-line fault diagnosis for batch process.On the basis of the multi-phase feature of batch process and the data information among batch process samples, this article studies a fault diagnosis method regarding SVM as the core which can be summarized as follows:(1) Phase division method of batch process based on improved MPCAMulti-phase characteristic is the inherent feature of batch process, the phase division method based on MPCA just chose the angle information of load matrix as the similarity criterion, while ignoring the influence of the distance information of load matrix, so it would cause the inaccurate of phase division easily. Therefore, this paper constructs a new similarity criterion formula of load matrix combining the distance information of load matrix with the angle information of load matrix, which can increase the accurate of phase division. Finally, this method integrated with SVM is applied to fed-batch penicillin fermentation simulation platform, which is compared with traditional SVM, and the results show the method’s availability.(2) Research on an improved binary SVM algorithmThe multi-classification of traditional binary SVM decided the hierarchy of classification random without taking the separation of specific object data into consideration, while the hierarchy of classification plays a decisive role for the whole classification method. So, this paper designs a formula of separability measurement for batch processes according to a principle that the same class is relatively clustered and the different classes have a relatively for distance is easier to classify, at last an improved binary SVM method is achieved and is verified by the same simulation platform as the content(1).(3) Research on a fault diagnosis method for batch process based on sub-phase MPCA-SVMIn the previous sections(1) and(2), the off-line fault diagnosis methods were discussed, but for the batch process on-line ones have more actual applied value. In order to achieve the on-line fault diagnosis, it is necessary to import an algorithm of process monitoring.Hence, this paper proposes an on-line fault diagnosis method for batch process based on sub-phase MPCA-SVM, which uses the improved MPCA phase division method to part the phase of batch process, and then builds each MPCA monitoring model which is used to monitor the current process and the improved binary SVM fault diagnosis model which dignoses the fault monitored in each phase respectively. A case study from a simulated fed-batch penicillin fermentation process indicates the method proposed has higher accuracy than the SVM diagnosis method without phase division.(4) Research on E. coli fermentation process fault diagnosisSub-phase MPCA-SVM fault diagnosis method proposed in this paper is applied to an actual biochemical- pharmaceutical factory in Beijing whose fermentation production is interleukin-2. On the basis of taking multi-phase characteristic and data separability of batch process into consideration, the results represent this method increases the accuracy of fault diagnosis obviously, thus can reduce the frequency of accident, so it has significant meaning for people.