Artificial intelligence and multivariate statistics for comprehensive assessment of filamentous bacteria in wastewater treatment plants experiencing sludge bulking

Abstract Sludge bulking is an operational hurdle that affects the solid–liquid separation in wastewater treatment plants (WWTPs) worldwide. In this study, filamentous bulking issues were investigated in seven WWTPs located in South Africa using artificial neural network (ANN) and multivariate statistics. The microbial community belonging to sludge bulking was determined using staining and microscopic methods, with a further confirmed identification of selected species via fluorescent in situ hybridization (FISH). Based on a filament index scale from 1 (None filament) to 7 (Excessive filament), the developed ANN could predict the sludge volume index (SVI) using the abundances of ten inputs of filamentous species. Eikelboom Type 0041 attained the highest impact on SVI, followed by Gordonia spp., Nostocoida limicola, and Thiothrix spp. Principal component analysis (PCA) combined with FISH images showed that most WWTPs experienced inadequate sludge settling properties; however, the application of an efficient aeration system (i.e., diffusion) in the three-stage Phoredox process improved the settling characteristics of bio-flocs. Operational conditions that caused filament overgrowth in each WWTP were also determined. The study outputs would provide a scientific basis to control the proliferation of filamentous bacteria in other WWTPs located in similar environmental conditions to South Africa.