Model-based identification and control on baker's yeast fed-batch fermentation

Abstract

A theoretical and experimental study is reported concerning the control of baker's yeast fed-batch fermentation. The overall approach follows the methodology proposed by Bastin & Dochain (1990) which exploits the structure of a general nonlinear deterministic model for bioreactors. Software-sensors have been developed and employed for the on-line estimation of 'non-measured' state variables (Feyo de Azevedo et al., 1992). An adaptive linearizing control cheme, based as well on the general model structure, utilises the observed state to compute the control action. The strategy was tested theoretically through a 'real-time process simulator' (Pimenta & Feyo de Azevedo, 1993) acting as the baker's yeast fermentation, linked to a 'control computer' via serial RS232C communications. Experimental work was carried out in parallel in a 5 litre laboratory rig (B. Braun Biostat MD5) locally controlled by a 'digital control unit' and supervised by a control computer. The main results to be reported address the overall performance of the control scheme, particularly in what concerns its sensitivity to model uncertainty, the noises in measurements and to slow and fast changing process parameters. As well it is relevant to emphasise that the 'process simulator' employed has proved to be an important tool for studies on process control and for experiment planning.