Investigation of zero-valent iron (ZVI)/H2O continuous processes using multivariate analysis and artificial neural networks

Abstract

Multivariate statistical techniques and artificial neural networks (ANNs) were used for the analysis, interpretation, and modeling of the results obtained in the study of zero-valent iron (ZVI) reactive beds designed for contaminant removal. A wide range of operating conditions was evaluated through more than 120 rapid small-scale column tests (RSSCT). The production of Fe(II) and Fe(III) species, dissolved oxygen consumption, and pH variation along the reactive bed were used as response variables for evaluating the process performance. Due to the complexity of the system, and the difficulty in defining and fitting kinetic parameters, ANN models were used to simulate the system without the need for kinetic expressions. Therefore the latter were used for assessing the system behavior within the investigated experimental domain and for evaluating the relative importance of the operating factors. In addition, the application of the multivariate techniques cluster analysis (CA) and principal component analysis (PCA) revealed underlying relationships among the response variables. Moreover, although multiple physicochemical processes are involved, the results obtained through PCA indicate that the main trends can be rationalized by considering a few key reactions only. The strategy of analyzing RSSCT results with different numerical techniques provides valuable knowledge for designing real-scale ZVI-based treatments aimed at the efficient elimination of a wide range of contaminants in the aqueous phase.