Automation of an industrial pilot bioreactor based on model predictive control with nonlinear parameter estimation

This paper presents an approach for modelling and control of fed-batch yeast growth which requires extensive calculation without the need of off-line measurements and model training. The main aim of the presented approach is to provide a methodology with high flexibility towards a varying process. The process model represents essential system behaviour. Based on this model, an adapted predictive control algorithm provides the future manipulated variable by analytical calculation, avoiding numerical optimisation methods. Hereby calculation time is significantly reduced. Thus all the software for data-acquisition, supervision and control can be run on one PC for several bioreactors at same time. In order to adapt the controller automatically to operating conditions and the respective yeast strain, model parameters are identified on-line by a non-linear estimation algorithm. The controller software was implemented together with additional supervisory routines on an industrial DCS. Experimental performance of this approach is shown for the control of ethanol production during fed-batch growth of Saccharomyces cerevisiae.