Improving control of membrane fouling on membrane bioreactors: A data-driven approach

Abstract

Despite the widespread application and plentiful benefits of membrane bioreactors (MBR), this technology is constrained by membrane fouling, which is regarded as the most serious drawback of process efficiency, especially for treating complex industrial wastewater. In order to control membrane fouling, a large set of variables that are often strongly correlated must be investigated and therefore data mining techniques can be used to extract relevant information from monitoring data sets. This work investigates the relations between different analytic and operating variables (mixed liquor volatile suspended solids, sludge filterability, pH, chemical oxygen demand of the feed, temperature, sequential days without cleaning (SDWC) and membrane permeability) of a pilot-scale MBR treating an oil refinery wastewater by applying Principal Components Analysis (PCA) and Multivariate Statistical Process Control (MSPC). The analyses were performed in R based on a five-year monitoring database. PCA identified sludge filterability, temperature and SDWC as the most influential variables on membrane permeability and it was effective in predicting MBR performance (R2 = 0.71 and Q2 = 0.78), enabling to detect atypical observations and thus identify operating faults. Besides, T2 and Q control charts were able to detect 100% and 96%, respectively, of the low membrane permeability operation (lower than 100 L h−1 m−2 bar−1) and also to alert preventively about permeability decrease. Therefore, the multivariate control charts could be applied as tools for supporting the decision-making regarding fouling control, e.g., guiding when to perform chemical cleanings and/or dose membrane permeability improvers.