Reliable Diagnosis and Early Warning of Filamentous Bulking in Microaerobic Wastewater Treatment Systems under Small Data Constraints

Abstract

Abstract Filamentous bulking can destabilize microaerobic wastewater treatment systems, creating a need for diagnostic methods that distinguish bulking severity. Using 532-day operational records and morphology-graded sludge samples, this study investigated filamentous-bulking dynamics in a microaerobic wastewater treatment system. Elevated COD sludge loading was identified as a key operational condition associated with increased bulking risk. To support grade-level diagnosis, we developed a reliability-enhanced set-valued diagnosis (RESVD) framework. RESVD integrates statistically constrained training with set-valued inference to generate uncertainty-aware severity sets from multivariable sludge-state information. For reactive diagnosis of filamentous bulking, RESVD achieved an empirical inclusion rate of 90.8 ± 7.1% with an average set size of 1.52 ± 0.13. Coupling RESVD with a two-stage gated recurrent unit model for sludge-property prediction extended the diagnosis to a 5-day early-warning scenario; under forecast uncertainty, RESVD achieved an empirical inclusion rate (EIR) of 86.7 ± 9.2%, and 91% of Grade 3 severe-bulking cases were included in the set-valued outputs. These results show that linking routine operational dynamics, sludge-property trajectories, and morphology-based severity labels can provide conservative risk screening for filamentous bulking in data-limited microaerobic systems.