engleski

Artificial neural network and mathematical modeling of drying of apples treated with high intensity ultrasound

INTRODUCTION Empirical mathematical models of drying produce very accurate results with minimal error. However, this is usually limited to specific values for input parameters of the experiment. Artificial neural network (ANN) is one way to predict output parameters based on relatively large set of input parameters. In this paper various mathematical models will be fitted to experimental data, and compared to ANN. It will show the ability of ANN to describe drying behaviour of apples treated with high intensity ultrasound METHODOLOGY High intensity ultrasound was used as pretreatment of apple slices with various thickness (2, 4 and 6 mm) using 20, 40, 60, 80 and 100 % of amplitude during 4, 6 and 8 minutes. After treatment, slices were dried in the infrared dryer (at 50, 60 and 70 oC) until no difference between readings was observed. ANN model was modeled and compared to most often used mathematical regression models for drying of food materials. Conducted statistical analysis resulted in mean square error and correlation coefficients, which were used as parameters for the selection of the model with best fitting to experimental data. RESULTS AND DISCUSSION All regression mathematical models were excellent in prediction of drying curve, and Page model was the one with best fitting to experimental data. Any change in the values for input parameters leads to decrease of mean square error, thus often making model inadequate. ANN performed excellent, independently of changes in the input parameters. CONCLUSIONS ANN appears to be the best way to model experimental data, especially with larger number of input parameters. While all mathematical models were adequate in fitting, with acceptable error, Page’s model shown smallest error and therefore appear to be best model for just one set of input parameters.

ultrasound; apple; drying; artificial neural network

nije evidentirano