To examine applicability of a fuzzy neural network (FNN) to decision of moromi temperature in ginjo-sake brewing process, two kinds of ginjo-sake were made and compared. Temperature of the one moromi was calculated automatically by the FNN, while that of the other was manually managed by the toji, a sake-brew master. Baumé alcohol concentration and temperature from 25 kinds of ginjo moromi made in 17 sake breweries in Aichi prefecture from 1989 to 1991 were used to construct the FNN model for decision of ginjo moromi temperature.
Each sake brewing employed 100 kg total Wakamizu rice polished to 50% and sokujomoto made by using a ginjo-sake yeast, S. cerevisiae FIA-2 strain. Temperatures during the first 11 days changedsimilarly for the both moromi. However, temperatures from 12 to 29 days in the case of the FNN control were 0.5-1.5°C lower than those of the manual control. Althoughthere were some differences in the concentrations of flavor components (e. g., 4.20 ppm iso-amylacetate in the sake by the FNN control, 3.82 ppm by the manual control), the seven panelists judged that ginjo flavor of the two kinds of sake was similar. The concentrations of chemical components, physicalproperties and sensory evaluation had almost the same values in these sake, suggesting that ginjo-sake can be made under the FNN control with almost the same quality as that made under the manual control of the toji.