Control of the penicillin production using fuzzy neural networks

Addresses the control of a penicillin fermentation pilot plant using internal model control (IMC) strategies with modules based on a FasArt neuro-fuzzy system. FasArt features fast, stable learning and shows good MIMO identification, which makes it suitable for development of the modules in IMC. Experiments have been done on training FasArt on real data and applying the controller to the pilot plant, and these show that the trend of reference is captured, thus allowing high penicillin production. Other experiments have been aimed at the development of soft sensors of important variables using FasArt. Biomass, viscosity and penicillin production predictors are very accurate, and reveal that FasArt modules could be employed for fault detection, control with constraints or predictive control.