Development of an alcohol fermentation control system based on biosensor measurements interpreted by neural networks

The lack of cheap and robust continuous analytical instruments — capable of on line measurement of the substance concentrations present in a bioreactor — has stimulated the development of instruments called biosensors. The use of such sensors brings difficulties, because of their nonlinear response to different concentrations. Additionally, the sampling system frequently utilized, based on flow injection analysis (FIA) technology, provides discrete signals of the measured concentration, related to peaks occurring at different times. The objective of this work is to anticipate and suggest solutions to the difficulties arising from the use of these biosensors for monitoring and control of fermentation processes. For this aim, two studies were accomplished. The first one verified the capability of artificial neural networks to interpret biosensor signals. Experimental glucose and sucrose biosensor data were employed with this purpose. The second one investigated, by simulation, the use of a biosensor — modeled using typical experimental data, interpreted by neural networks — in the control of an alcohol fed-batch fermentation. It was concluded that the nets are adequate tools to correlate biosensor signal to the measured concentrations. Alternatives to the use of conventional feedback controllers in the presence of these sensors are also proposed.