Journal of Food Processing & Technology

Computational performance of two convergence methods applied to the estimation of viscosities for essential oils industry

4^th International Conference and Exhibition on Food Processing & Technology

August 10-12, 2015 London, UK

Cintia B Gonçalves, Gustavo V Von Atzingen, Priscila M Florido, Christianne E C Rodrigues and Camila N Pinto

Posters-Accepted Abstracts: J Food Process Technol

Abstract:

Essential oils are mainly composed of terpenes, which are unstable to heat, and oxygenate compounds, responsible for their
characteristic aroma. Deterpenation of essential oils results in a better quality product and can be performed using solvent
extraction. In this process, viscosity is important as it affects the loss of energy by friction and the mechanisms of heat and mass
transfer. As viscosity can be affected by a large amount of variables, the use of appropriate methods for estimating it becomes a useful
tool. The aim of this work was to study the influence of the convergence method on the modeling procedure to calculate viscosities of
phases arose from the deterpenation of essential oils, using the UNIFAC-VISCO model. Interaction group parameters were obtained
by correlating the model with experimental data of systems containing terpenes oxygenates compounds and alcoholic solutions. The
matrix of parameters obtained was used to predict viscosities of similar mixtures that were not included in the modeling. The model
was programmed using MATLAB® platform and gradient descent as convergence method. Results were compared to those obtained
in our previous work* using a genetic algorithm. Averages relative deviations obtained in this work for modeling and prediction (3.06
% and 4.97 %, respectively) were higher than those obtained in our previous work (1.70 % and 3.56 %, respectively). The modeling
results show that by using the gradient descent presented a lower predictive capacity. Furthermore, interaction parameters were
obtained one hundred times slower than those ones using genetic algorithm.