A novel, repeated fed-batch ethanol production system from corn flour by fully recycling fermented supernatants is demonstrated. With recovery of ethanol by evaporation, consecutive removal of the insoluble and soluble inhibitory substances accumulated either completely or partially by filtration, the concentrations of the soluble inhibitors in the system could be maintained at their equilibriums. As a result, a sustained high concentration ethanol production and significant pollution control performance were obtained.

An on-line adaptive substrate feeding control system based on DO/pH easurements and artificial neural network pattern recognition (ANNPR)model for fed-batch cultivation processes proposed previously was successfully applied for the phytase production with recombinant Pichia pastoris. The control strategy could effectively improve the fermentation performance for both cultivation and induction phases. With the standard ANNPR model-based control strategy, the cultivation time before starting the methanol induction could be shortened for about 30%. By adequately increasing the feeding rate iterative step size of the ANNPR-based control during induction phase, methanol concentration could be automatically controlled within an optimal range, leading to an approximate three-fold stable increase in phytase activity compared with those obtained by the traditional DO-Stat method and the on-line methanol electrode-based on–off control strategy. The effectiveness, universal ability, as well as the operational simplicity of the proposed control system has been further verified in the recombinant P. pastoris fed-batch culture process.

It has been recognized that fuzzy control is quite effective in coping with the uncertain dynamics associated with bioprocesses. However, it is, in general, quite time-consuming to determine the IF-THEN type of rules and membership functions in conventional fuzzy control. In the present study, we consider adjusting the membership function on-line with the aid of neural networks, where the role of the neural networks is to recognize the patterns of changes in the DO concentration (oscillatory/ non-oscillatory), and ethanol concentration (increasing/decreasing/nearly constant) in baker's yeast fed-batch cultivation. This neuro-fuzzy control strategy retains the advantages of both neural nets and fuzzy control. The application to baker's yeast fermentation shows the power of this strategy as compared with conventional fuzzy control.

A novel on-line dual analyzer combined with an automatic sampling device was developed. When immobilized glucose oxidase and lactate oxidase are used in the analyzer, glucose and lactate concentration can be measured on-line. Lactate concentration did not influence the measurement of glucose concentration and vise versa. As an example of the application of the analyzer, on-off control of the glucose concentration in fed-batch lactic acid fermentation was carried out to increase the final lactate concentration as much as possible. When the glucose concentration was controlled at I g/l, the final lactate concentration was 57 g/l, corresponding to a 21% increase over that obtained when the glucose concentration was set at 20 g//. The experimental results show that if the glucose concentration is controlled at a low value, say about I g/l, then the accumulation of by-prod-ucts such as acetate can be repressed and the yield of lactate from glucose increases.

We previously proposed a neuro-fuzzy control strategy for bioreactors, and showed the power of this approach by computer simulation. However, pattern learning and pattern recognition by neural networks was quite time-consuming. This is because many training data must be generated somehow, and this has been considered to be an obstacle to practical applications. We made several experiments using E. coli. It was shown that pattern learning and recognition of the change in the dissolved oxygen concentration by neural networks can be easily made only with experimental data. This may be due to the fact that experimental data contain a variety of patterns in nature.