Desenvolvimento de metodologias para monitoramento de biorreatores com membranas no tratamento de efluente de refinaria de petróleo

Abstract

Membrane bioreactors (MBR) have been widely used for industrial effluent treatment due to their greater efficiency in pollutants removal, enabling effluent reuse. However, membrane fouling is still a limiting factor in this technology applicability. Understanding and optimizing such a complex system as MBR requires a thorough study that adds the individual and synergistic contribution of various phenomena involved, made possible by the use of statistical tools. In this context, the objective of this work was to evaluate and optimize the employment of MBR using statistical tools. For this purpose, MBR pilot plants from oil refinery effluents treatment were employed. Using Artificial Neural Networks, a sensitivity analysis study was performed to investigate the analytical and operational variables effects on membrane permeability. After the identification and validation of a predictive permeability neural model, sensitivity analysis methods were applied to quantify and classify the variables influence. Comprehensive analysis showed that suspended solids and days between cleanings had the greatest effect on permeability. Subsequently, a specific analysis revealed distinct dynamics in MBR operation, considering different solids concentrations. By using Principal Component Analysis, it was possible to verify the relationships between the membrane permeability and some process variables, as well as to observe the influence of parameters over the monitoring period and the possibility of process control through verification of observations outside the confidence ellipses. By applying multivariate statistical process control, the possibility of predicting membrane permeability loss was verified. With the construction of multivariate control charts T2 and Q it was possible to detect 45% MBR operation err, which would decrease membrane permeability. Once the points of failure were detected, the fault identification methodology was able to verify the parameters that most influenced the control leakage, namely days between cleanings. The results may improve the understanding of process variables effects on the interest responses, achieving more efficient interventions. It may bring, therefore, from the verified statistical analyzes, clarifications and contributions to more rational decision making about BRM use and monitoring and industrial processes in general.