Research On Fault Detection And Diagnosis Method Of Fermentation Process Based On Improved Algorithms KECA And SVDD

There are many production stages in the fermentation process and the mechanism of the production process is complex.Its product quality is easily affected by uncertain factors.Monitoring and fault diagnosis of the fermentation process face many challenges.In this study,the fermentation process was taken as the research background,and the multivariate statistical method based on data driving was taken as the main method of the research.The research was carried out from the two directions of fault monitoring and diagnosis in fermentation process,and some improvement strategies were proposed for the traditional monitoring and diagnosis model based on the characteristics of fermentation process.The main research contents of this study are as follows:(1)This paper introduced the principle of kernel entropy component analysis(KECA)applied to multivariate statistical process monitoring and the calculation of Cauchy-Schwarz(CS)statistics based on angle structure.According to the three-dimensional data structure characteristics of fermentation process,three different unfolding modes of multiway KECA method were discussed.Then,the monitoring framework and monitoring steps of multiway KECA method in batch fermentation process were elaborated,which provides theoretical basis for subsequent improvement methods.(2)Aiming at the dynamic,time-varying and multi-stage problems in batch fermentation process,a new on-line fault monitoring model,moving window-kernel entropy component analysis(MW-KECA)monitoring model,for fermentation process was proposed by introducing a moving window strategy based on adjacent region modeling on the basis of the above KECA monitoring algorithm.The MW strategy based on samples of adjacent time points can track the changes of the process well and solve the dynamic and time-varying problems of each batches.At the same time,the MW-KECA monitoring retains KECA algorithm's excellent ability to deal with nonlinear data,its CS statistic also has good robustness.The MW-KECA monitoring method was compared with MW-KPCA and KECA methods in penicillin fermentation simulation experiments,the results show that MW-KECA can detect faults more accurately,which proves the effectiveness of the monitoring method proposed in this study.(3)Aiming at the problems of insufficient generalization ability and low recognition rate of fault diagnosis model in fermentation process,a fault diagnosis model based on support vector data description(SVDD)with fault variable contribution data as input parameters was proposed.The variable contribution data is equivalent to the feature data extracted from the original data,which can highlight the similar information of the same kind and the difference information of different kinds in the data.Through the feature extraction,the recognition effect of SVDD fault diagnosis model is optimized.In penicillin fermentation simulation experiments,compared with SVDD method based on original data modeling,SVDD method based on variable contribution improved the generalization ability and accuracy of the diagnostic model,which proves the feasibility of the fault diagnosis method proposed in this study.