Knowledge-infused hierarchical causal inference framework for decoding sludge settleability to quantitatively regulate multi-process wastewater treatment

Sludge settleability characterized by the sludge volume index (SVI) is essential in assessing the operational stability of wastewater treatment systems. However, previous studies primarily focus on correlations between features and SVI in single wastewater treatment processes but lack revealing SVI causal relationships among multi-process differences. To address this, this study proposes a framework combining knowledge-infused feature engineering and hierarchical directed causal informatics linking strategies based on monitoring data of a full-scale multi-process wastewater treatment plant (WWTP) (400,000 m3/d). Datasets from three parallel processes (Oxidation Ditch, OD; Anaerobic-Anoxic-Oxic, A2O; A2O-Membrane Bioreactor, A2O_MBR) and feature stratification guide the modeling of machine learning (ML) and structural causal model (SCM). The framework delivers excellent fitting and generalization, with the integrated ML model reaching an average R2 of 0.83. Specific causal mechanisms for different processes were identified as follows. Temperature (T) exerts the most significant negative average treatment effects (ATEs) on SVI in A2O (−0.4372), followed by OD (−0.3327) and A2O_MBR (−0.2574). A2O_MBR exhibits the strongest disturbance resistance that can be recommended for regions with extreme or highly variable climates. Quantitative differentiated operational strategies are developed. At low T, OD controls biochemical oxygen demand loading (BOD_load) to less than 0.06 kg (BOD)/kg (MLSS) or external sludge recirculation ratio (SRR_ex) less than 0.75 to prevent bulking. At high T, A2O maintains mixed liquor suspended solids (MLSS) above 4800 mg/L or pH_in between 7.0 and 7.3 to mitigate aging. These insights enable the development of differentiated, causality-aware operational strategies to advance stable and sustainable wastewater treatment.