An expert system applied to the control of an industrial-scale bioreactor

Control of a fed-batch industrial-scale fermenter requires a large amount of “finger-tip feeling” on the part of operators, despite reasonably sophisticated instrumentation. The reason for this is that there are no applicable deterministic models giving aid to the control system in deciding the best operation strategies. This is a typical case of a heuristic experimental process, treating inexact knowledge that is excluded from deterministic and statistic modelling a priori. Essential parts of this problem can be solved by means of an expert system shell, capable of dealing with uncertain information, using the linguistic shell application. Knowledge was extracted from process records on fermentation batches producing baker's yeast on an industrial scale. A strategy for heuristically optimized yield and growth rate can be predicted using knowledge gained in previous runs and stored in the knowledge base of the expert system. The expert system, once built up, can also be used for experimental simulation of the effects of fundamental decision parameters: temperature, production rate, ethanol concentration and specific growth rate. The use of expert system consultations can be generally recommended as an efficient aid in fermentation control.