Abstract
In this work the Artificial Neural Network (ANN) and the Perturbed Hard Sphere Chain (PHSC) equation of state (EoS) have been utilized to estimate the osmotic coefficient, activity coefficient, and water activity of aqueous sugar solutions containing glucose, fructose, fucose, xylose, maltose, mannitol, mannose, sorbitol, xylitol, galactose, lactose, ribose, arabinose, and sucrose. The PHSC model parameters have been adjusted using the osmotic coefficient experimental data. Then, the water activity and sugar activity coefficient were predicted. In the case of the ANN approach, six variables containing critical temperature (T(c)), critical volume (V(c)), molality, temperature, melting temperature (T(m)), and melting enthalpy (∆H(m)) of sugars have been considered as input layer. As well, 32 neurons are considered in one hidden layer. The Group Contribution (GC) method was utilized to estimate the critical properties of sugars. The training correlating coefficient (R(2)), and the Mean Square Error (MSE) have been obtained 0.999 and 2.06 × 10(-6), respectively. The average relative deviation (ARD) value of osmotic coefficient, water activity, and sugar activity coefficient using the PHSC EoS and the ANN + GC model have been obtained 0.43%, 0.12%, 0.66%, and 2.1%, 0.89%,1.65%, respectively. The model's performance has been evaluated using the prediction of sugar solubilities in water. The results show that the ANN + GC and PHSC model can predict the solubility data satisfactory. The ANN + GC method can be used to predict the thermodynamic properties of a new aqueous sugar solution using the molecular structure in the absence of experimental data.