ARTIFICIAL NEURAL NETWORKS FOR MODELING THE DRYING PROCESS DYNAMICS OF SCHINUS TEREBINTHIFOLIA RADDI FRUIT

摘要

In this study, topologies of Artificial Neural Networks (ANNs) for predicting the drying kinetics of Schinus terebinthifolia Raddi fruit (pink peppercorns) were investigated. Pink peppercorn, also called poivre rose in French, is among the most sophisticated condiments in international cuisine. The experiments were performed in a pilot thin-layer dryer (fixed bed dryer) at different air temperatures (40 to 70ºC) and drying air velocity (0.4 to 0.8 m/s). The database was expanded using data interpolation by cubic splines. In the modeling were used feedforward ANN formed by three layers. The activation function for hidden neurons was the hyperbolic tangent. For the ANN training, the Levenberg-Marquardt algorithm with Bayesian regularization was used. The coefficient of determination (R~2) and the root-mean-square error (RMSE) were used to compare the performance of the ANNs. The influences of database, input neurons, and activation function were also investigated. The results show that the ANN successfully represented the drying kinetics of S. terebinthifolia Raddi fruits.