A hybrid feedforward neural network model for the cephalosporin C production process

At present, direct on-line measurements of key bioprocess variables as biom ass, substrate and product concentrations is a difficult task. Many of the available hardware sensors are either expensive or lack reliability and rob ustness. To overcome this problem, indirect estimation techniques have been studied during the last decade. Inference algorithms rely either on phenom enological or on empirical models. Recently, hybrid models that combine the se two approaches have received great attention. In this work, a hybrid neu ral network algorithm was applied to a fermentative process. Mass balance e quations were coupled to a feedforward neural network (FNN). The FNN was us ed to estimate cellular growth and product formation rates, which are inser ted into the mass balance equations. On-line data of cephalosporin C fed-ba tch fermentation were used. The measured variables employed by the inferenc e algorithm were the contents of CO2 and O-2 in the effluent gas. The fairl y good results obtained encourage further studies to use this approach in t he development of process control algorithms.