Modelagem via redes neurais artificiais do processo de tratamento de efluentes têxteis, utilizando processos oxidativos avançados

Abstract

Advanced oxidative processes are excellent alternatives to the treatment of effluents, such as industrial effluents, mainly due to their capacity to form a transient compound with a high oxidation potential, the hydroxyl radical. For the optimization of the treatment, it is common to model via Artificial Neural Networks (ANN), which consist of computer systems that have a high capacity for storing and processing information for later reproduction. Thus, the objective of the present work was to compare the degradation of methylene blue dye from a synthetic textile effluent (ETS) through the photolysis of hydrogen peroxide with ultraviolet radiation (H2O2 / UV) and photo-Fenton (H2O2 / UV / Fe^2+ ), on a bench scale. In addition, it was sought to perform the modeling of the data obtained by means of RNA, using a backpropagation algorithm. For this purpose, a factorial 2^4 experimental design was used, in which the influence of the dye concentration, peroxide concentration (for the H2O2 / UV process) and peroxide and iron concentration ratio (for the photo-Fenton process) were evaluated, pH of the ETS and time of incidence of UV radiation, in the efficiency of removal of methylene blue. For the subsequent modeling of the data, the MATLAB® software was used. It was observed that the photo-Fenton process is quite efficient for the discoloration of the ETS when compared to the H2O2 / UV process, which reached an additional 98% in 5 minutes of UV radiation. It was found that all four factors were statistically significant in the treatment for both processes and that the dye degradation kinetics in the best condition of the experimental design was 1.25 and 1.56, for the H2O2 / UV and photo-Fenton processes , respectively. As for the data modeling via RNA, it was possible to observe that the models for both processes adjusted satisfactorily to the experimental data, reaching a determination coefficient above 0.96 for a network with only 5 neurons in the single hidden layer.