Combining experimental and modelling approaches towards full-scale implementation for membrane bioreactor optimisation

Membrane bioreactors (MBRs) are a mature and widely applied technology for municipal wastewater treatment, offering high effluent quality and compact plant configurations. Nevertheless, their operation remains constrained by membrane fouling and by the high energy consumption associated with air-scouring. This doctoral thesis addresses these challenges by developing an integrated approach that combines experimental membrane characterisation, advanced control strategies and data-driven modelling to optimise full-scale MBR operation. The first part of the research focuses on improving the experimental assessment of filtration performance. The conventional flux-step test (FST), commonly used to estimate the critical flux, is shown to be insufficient to fully describe filtration stability, as it does not account for the interaction between permeate flux and aeration intensity. To overcome this limitation, an aeration-step test (AST) is proposed to quantify the specific aeration demand per membrane area and to identify aeration thresholds required to control reversible fouling. Experimental results obtained with different hollow-fibre membrane types demonstrate that membranes with similar critical flux values may require substantially different aeration strategies, highlighting the need for integrated flux–aeration criteria. The second part of the thesis develops and validates a dual-phase fuzzy logic control system for full-scale MBRs. The proposed control architecture uses online measurements of transmembrane pressure and fouling rates at different time scales to dynamically regulate air-scouring intensity and, under severe fouling conditions, supervise permeate production. Full-scale validation over an extended operational period shows that the controlled filtration line achieves lower fouling rates and reduced aeration demand compared to conventionally operated lines, while maintaining stable process performance. Finally, the thesis explores the use of data-driven models to support predictive membrane maintenance. Machine learning techniques are applied to predict transmembrane pressure evolution using routinely available operational data, incorporating uncertainty analysis to improve model robustness. Overall, the results demonstrate that combining advanced experimental methods, intelligent control systems and data-driven modelling provides an effective pathway towards more energy-efficient, robust and autonomous operation of full-scale membrane bioreactors.