COMPARISON OF ARTIFICIAL NEURAL NETWORK AND MATHEMATICAL MODELS FOR DRYING OF APPLE SLICES PRE-TREATED WITH HIGH INTENSITY ULTRASOUND

摘要

In this paper, an artificial neural network model was compared to the traditional regression models for drying food materials. High intensity ultrasound with amplitudes set to 25%, 50%, 75% and 100% of maximal was used for the treatment of apple slices of different thicknesses. After 7 min of treatment, samples were dried in the infrared drier at two different temperatures. The four most frequently used regression models for drying available in the literature were fitted based on experimental data, and their usability was tested on different experimental sets. For the creation of back-propagation neural network, 3 input parameters were used (amplitude of ultrasound, sample thickness and drying temperature) together with one output (moisture content). After training and the validation of networks, statistical analysis was conducted, based on the mean square error and correlation coefficient, the best network was selected. After the assessment of networks and statistical results, neural networks showed excellent fitting to experimental data, independently of the input parameters obtained in experiments. This is opposed to standard regression models, which had excellent fit to just one set of experimental data, and show inadequate fit even with small-introduced changes in one or more input parameter.