Fault Diagnosis Method Research Of Penicillin Batch Process

With the quick development of the biological technology, batchprocesses have played important role in process industry. Penicillinfermentation process is a typical batch process, which is a non-linear,time-varying, multi-batch and other characteristic. In order to improveproduction stability and safety of fermentation process， an efficientmonitoring system must be structured to detect the abnormal events of theprocess．Considering the characteristic of the fermentation process andmultivariate statistical monitoring of the situation, penicillin batch processmonitoring systems is made significant improvements, experimental resultsproof the effectiveness of the proposed method. Main researches include:(1)The statistical monitoring of batch processes of the status quo athome and abroad is described. Then characteristics, operation, process andfactors of fermentation process are introduced; Pensim2.0experimentalplatform is also described.(2)Based on a fast independent component analysis (FastICA) andrecursive least squares support vector machine (RLSSVM) monitoring method is proposed. The maximization of non-Gaussian criterion isadopted to decompose the independent non-Gaussian components from themonitoring variables. RLSSVM monitor independent variables on theextraction of the monitor model, failure to detect the moment. Simulationresults show that classification accuracy has been significantly improved,with fewer false alarms and missed faults than RLSSVM training model.(3)Batch process monitoring method based on lifting wavelettransform and multi-way principal component analysis is proposed. Liftingwavelet transform solve the noise of process data, then use MPCAmodeling. The simulation is performed by means of statistical method suchas HotellingT2、 SPE and contribution plots based on Pensim2.0. Thesimulation results confirm that, the method describes the behavior of batchprocess more precisely.