Formulation of fats using neural networks: commercial products and pilot-plant production

In the present work, trials were carried out to verify the range of application of a neural network, designed to formulate fats from 3 raw materials from soybean. For evaluation, 17 randomly selected commercial products, originally formulated with unknown raw materials, were used. Apart from the commercial products, 3 different table margarines were formulated using conventional methods and the neural network, and produced in pilot-plant scale. According to the results obtained, the neural network presented a success index of 64.7% in the formulation of commercial products. With respect to the products formulated and produced in pilot-plant scale, the results obtained using the network were similar to those obtained using the conventional formulation methods.