The control of product crystal size in an industrial crystallizer with an actual heat exchange area of 400 m
2 is discussed with regard to application of a neural network model consisting of three explanatory variables: steam flow rate, suspension density of crystal and frequency of circulation pump. The most suitable learning number for the neural network model obtained by the Leave-one-out Cross Validation method was 50,000, and the mean estimated error of product crystal size was about 30μm. From these results, it is believed that the neural network model is accurate enough for practical use, and is effective for designing operational conditions for manufacturing products with the desired crystal size in industrial crystallization. A practical constructing method for a neural network model is proposed.
View full abstract