BLENDING PROCESS OPTIMIZATION INTO SPECIAL FAT FORMULATION BY NEURAL NETWORKS

Computer programs are used to manage, supervise, and operate productio n lines of oil, margarine, butter, and mayonnaise in the fats and oils industry. Automation allows for lower-cost and better-quality product s. The present paper shows a multilayer perceptron-type, second-genera tion neural network that was built based on a desirable product solid profile and was designed to formulate fats from three ingredients (one refined oil and two hydrogenated soybean-based storks). This network operates with three sequential decision levels, technical, availabilit y and costs, to furnish up to nine possible formulations for the desir ed product, Upgrading verification was accomplished by soliciting to t he formulation network all 63 products used in the upgrading (the answ ers were evaluated by a panel of experts and considered satisfactory) and 17 commercial products. It was possible to formulate more than 50% of the products in the network with only the three bases available. T he results demonstrate the possibility of using neural networks as an alternative to the automation process for the special fats formulation process.