Modelagem e estimação de parâmetros da adsorção de corantes em carvão ativado

Abstract

The adsorption of dyes on activated carbon has been extensively studied in the past three decades. However, much due to the difficult to implement numerical procedures, these studies have been carried out without the necessary statistical background to make reliable conclusions. The main objective of this work was to perform the early stages of modeling and parameter estimation applied to the process of dyes adsorption on activated carbon, collaborating to the systematization of a computer routines set for the study of such process. The methodology has focused especially on the evaluation of the experimental error influence and linearization of the equilibrium models influence in the values of parameters and its confidence regions. The main routines used are part of a package called ESTIMA (PEQ/COPPE/UFRJ) which uses, basically, a stochastic method (Particle Swarm) together with a Gauss-Newton method in order to minimize the objective function. The parametric uncertainties have been compared by the elliptical approximation, the evaluation of several points obtained during minimization procedure, which are smaller than a determined limit, and an adapted method (SBA) from the Bootstrap method to build the parameters confidence region. Also, a methodology for model discrimination has been used and the effects of model linearization and proper characterization of experimental error have been studied. The realization of several replicates and subsequent error characterization have been shown of extremely importance to the study as the search for an alternative to obtain a more accurate estimate of the error in experimental conditions that have not been performed yet. It has been possible to experimentally show that the error on the amount adsorbed is indeed dependent of the error in the equilibrium concentration of adsorbate solution, a fact that, despite being known a priori, had not been yet demonstrated in other works. The comparison between linear and nonlinear methods also demonstrated how the linearization process negatively influences the correct data analysis and parameter determination. Through de analysis of the probabilities distribution of the equilibrium concentrations of the adsorbate solution (Ce), it has been shown that confidence region methods present a better performance when the dependent variable is transformed to the Log-Normal distribution, using the operation Neperian logarithm of Ce.