X. Wang, S. Lu, Z. Zhao, B. Li
University of Connecticut,
United States
Keywords: sensor-driven ASM1 modelling, dynamic digitalization, mechanistic autonomy
Summary:
1. Introduction Aeration accounts for 50–70 % of total energy use in wastewater treatment plants, especially in biological nutrient removal (BNR) systems where oxygen regulates nitrification and organic oxidation. Yet most plants remain constrained by limited digitalization and absence of autonomous control (Figure 1.1). Electrochemical sensors drift under ionic interference, fouling, and polarization, while recalibration disrupts data continuity, preventing the Activated Sludge Model No. 1 (ASM1) from tracking real-time microbial kinetics. To overcome these barriers, we developed an AI-calibrated Miniature Electrochemical Sensor Array (MESA) integrated with a dynamic ASM1 digital twin, enabling continuous, high-fidelity sensing and self-adaptive, energy-efficient BNR control through real-time digitalization and mechanistic autonomy (Figure 1.2). 2. Results and Discussion We evaluated the AI-integrated MESA–ASM1 digital twin in a 1-L aerobic reactor treating municipal wastewater from the University of Connecticut BNR system. The reactor was operated at 25 °C under nitrifying, fully aerated conditions. Four types of MESA sensors (NH₄⁺, pH, conductivity, and temperature) continuously monitored key parameters under steady operation and NH₄⁺ shocks (25 mg N L⁻¹) over 50-day period. Generative AI Enables Digitalization. We applied a variational autoencoder (VAE) to auto-calibrate MESA data, reconstructing true ion concentrations from latent correlations among multi-sensor signals. The Generative AI (GenAI) model autonomously corrected data drifting caused by ionic interference and fouling in wastewater, maintaining >95 % accuracy over 50 days and reducing calibration frequency tenfold (Figure 2.1). This confirmed that autonomous data integrity, instead of manual calibration, is the foundation of true digitalization in reactive wastewater environments. Dynamic ASM1 Enables Autonomous Control. GenAI-stabilized sensor data were assimilated into dynamic ASM1, forming a real-time digital twin that continuously recalculated heterotrophic organic carbon oxidation and autotrophic nitrification rates in response to evolving MESA inputs. Under NH₄⁺ shocks, the model instantly detected elevated nitrification demand, increasing oxygen supply accordingly (Figure 2.2). Aeration followed modeled oxygen demand (>0.85 mg O₂ L⁻¹ h⁻¹ = ON; <0.00 = OFF), proving that biochemical kinetics, rather than preset timing, can guide oxygen regulation. Mechanistic Autonomy Improves Efficiency and Resilience. Compared with continuous aeration currently implanted in treatment plants to ensure treatment efficiency, the dynamic sensor–ASM1 control achieved the same removal of organic carbon (≈459 mg L⁻¹) and NH₄⁺ (≈444 mg N L⁻¹) while reducing oxygen use by 56.3% and energy by 45.7% (Figure 2.3-2.4). Under NH₄⁺ shocks, aeration energy rose briefly then stabilized, demonstrating resilient adaptation to influent variability. Oxygen was supplied only during active biochemical oxidation, minimizing energy waste during substrate depletion. 3. Significance and Impact This study introduces two integrated innovations to transform BNR into a truly digitalized, autonomous process. Specifically, the GenAI auto-calibrated MESA continuously corrects sensor data, ensuring accurate, high-fidelity in situ data for real-time modeling. These stable data streams feed a dynamic ASM1 digital twin, translating microbial kinetics into self-regulating control. By linking AI-based digitalization with mechanistic autonomy, the system halves energy use while preserving treatment efficiency, establishing a scalable, intelligent, and resilient water infrastructure.