Advanced Modeling of Food Convective Drying: A Comparison Between Artificial Neural Networks and Hybrid Approaches

In the present paper, three different approaches are proposed to model the convective drying of food. The performance of thin-layer, pure neural network and hybrid neural model is compared in a wide range of operating conditions, with two different vegetables, available either as cylinders or as slabs with different characteristic dimensions. It was found that the thin-layer model was adequate to describe food drying behavior, but it could be applied only as a fitting procedure. Pure neural models gave accurate predictions in some situations, but exhibited poor performance when tested outside the range of operating conditions exploited during their development. Finally, it was shown that hybrid neural models, formulated as a combination of both theoretical and neural network models, are capable of offering the most accurate predictions of system behavior with average relative errors never exceeding 10%, even in operating conditions unexploited during the definition of the neural part of the model. The results obtained proved that the hybrid neural paradigm is a novel and efficient modeling technique that could be used successfully in food processing, thus allowing drying process optimization to be achieved, and efficient and fast on-line controllers to be implemented.