ANN inversion based soft-sensing of biochemical parameters in erythromycin fermentation

This paper presents a novel soft-sensing approach based on artificial neural network (ANN) inversion to estimate some crucial biochemical parameters in erythromycin fermentation, which usually can not be directly measurable by commercial sensors. Such direct-unmeasurable variables as mycelia concentration, sugar concentration and chemical potency, can be derived from other direct-measurable variables such as dissolved oxygen concentration, pH, and volume by using the proposed ANN inversion. The ANN inversion consists of a static ANN and several differentiators and acts as a soft-sensor. Experimental results show that the soft-sensing values are almost identical with the actual ones and the proposed method would be helpful for the real-time control of the biochemical fermentation.