State Estimation in Biotechnological Processes Using a Software-Sensor Combining Full-Horizon Observer and Neural Networks

This paper presents an innovative method for the online determination of biomass in fermentation, using a combination of model-based full-horizon observer and Neural Network. The performance of the Neural Network depends highly on correct initial conditions of the unknown process values. Unfortunately, in biological processes it is impossible to guarantee reproducible initial conditions. On the contrary, the variations in the innoculated amount of bacteria oscillate between 30% in lab scale and more than 100% in industrial applications. To reduce the effect of these variations to the Neural Network, we herein propose the use of an optimization based estimator to determine the initial conditions of the process values online in early process stages in order to improve the estimation results of the Neural Network.