Adsorption Isotherms for Red Onion Slices Using Empirical and Neural Network Models

Abstract

Moisture sorption isotherms of red onion slices were determined at 30, 40, 50, and 60°C using the standard gravimetric static method over a range of relative humidity from 0.11 to 0.83. The experimental sorption curves were fitted by seven empirical equations: modified Henderson, modified Chung–Pfost, modified Halsey, modified Oswin, modified Smith, modified BET, and GAB. Also three types of Artificial neural network models: linear, multilayer perceptron, and radial basis function were tested and developed to predict the equilibrium moisture content of onion slices and the selected models were trained by using related algorithms. The modified Oswin model was found acceptable for predicting adsorption moisture isotherms and fitting to the experimental data, based on the coefficient of determination (R2= 0.991), mean relative percent error (MRE=15.019), and standard error of estimation (SEE=1.371). Besides, multilayer perceptron model with four layers (2: 17: 14: 1) was selected as the best artificial neural network for estimation of onion slices’ equilibrium moisture content by considering R2= 0.993 and good performance. The net isosteric heat of adsorption of moisture was determined by applying the Clausius–Clapeyron equation to the sorption isotherms at different temperatures. The net isosteric heat of adsorption of red onion slices varied between 1.46 and 4.96 kJ/mol at moisture content varying between 2% and 52% (d.b.).