Prediction of Freezing Times And Weight Losses of Unpackaged Foods Using Artificial Intelligence Techniques

CAMPAÑONE L; ODDONE S.; SEGURA J.A; SALVADORI V.O.; MASCHERONI, R.H.

Brasil

Congreso; Enpromer 2005; 2005

Abstract. Artificial neural networks (ANN) allow to model complex systems in a relatively simple manner, in which system intrinsic nonlinearities are dealt by means of the learning and training of the network. In this way, ANN may be used to estimate or predict process times without the need of a mathematical model or a prediction equation associated to the physical problem. Topologically a neural network is an structure of interconnected nodes organized into layers and weighed through weight factors; each node’s output feeds into all nodes in the subsequent layer. The final factor governing a node’s output is the transfer function. During the training of the network, weight factors were adjusted until the calculated response pattern – for a given input – reflects the desired relationships. Freezing of unpackaged foods implies simultaneous heat and mass transfer. Freezing times and weight (water) losses are directly related to food characteristics and operating conditions. A very wide range of possibilities arise due to various food shapes, sizes and compositions and different temperatures, speeds and relative humidities of cooling air, feasible to be found under industrial freezing conditions. Numerical prediction methods are difficult to develop and use, but they provide a whole and accurate coverage of all possible conditions. The same order of accuracy can be attained by using an Artificial Neural Network developed ad-hoc and fed with the predicted freezing time and weight loss data from the numerical method. This work followed that way using as infeed variables food size, Biot number, relative humidity and initial, final and air temperatures. A very good matching to numerical data was obtained. The average absolute error was of the same order than that of the best prediction models.