A predictive neural network for biomass and substrate concentration estimation applied to the fermentation of Bifidobacterium longum ATCC15707

Abstract

This work presents the results of a predictive neural network model coupled with a mass balance equation applied to a Bifidobacterium longdum culture. The model can estimate the concentration of biomass and substrate for 17 hours from a single measurement at the beginning of the process. The data for the neural network training was obtained from experiments, in which values of current biomass, substrate and time were acquired. A Fourier filter was applied to the data to reduce high frequency variations attributed to experimental error. Results shows that the model obtained can estimate the growing behavior of the microorganisms and substrate consumption. These estimations can be used to reduce the amount of labor-intensive measurements of biomass and substrate concentration required to automate the process.