Project Summary:

This project integrates big data computational algorithms with physics-based modeling to develop the first general and physics-informed AI model for predicting and mitigating sewer overflows. The model will optimize control gate operations by determining the most effective sequence of decision variables to minimize overflow events.

The research combines multidisciplinary approaches—including laboratory-scale overflow experiments, computational fluid dynamics (CFD) modeling, reduced-order sewer modeling, hydrological modeling, and machine learning (ML)—to enhance our understanding of sewer overflow mitigation strategies.

Key outcomes of this work include: (1) a new sewer overflow prediction module for the open-source (Illinois Transient Model); (2) a new hydraulic engine based on ITM for integration with the widely used Storm Water Management Model (SWMM); and (3) a fully integrated, open-source, physics-based AI model for both predicting sewer overflows under fixed control settings and optimizing gate operations to reduce overflows system-wide.

Combined Sewer Overflows (CSOs) are a significant threat to public health and the environment, often resulting in contaminated drinking water, beach closures, and other disruptions. To address this challenge, the project successfully developed an ultrafast, integrated modeling framework that predicts the timing and location of CSOs in advance of heavy rainfall and identifies the optimal sequence of control gate operations to minimize overflows.

Traditional machine learning (ML) approaches for sewer system management rely on years of field data—often up to five years—to achieve reliable performance. However, such data are rarely available and cannot capture future climate scenarios. This project overcame that limitation by using high-fidelity numerical simulations, rather than field data, to train ML models—accelerating the development of predictive tools for real-time sewer system optimization.

The project achieved three major milestones: (1) the successful implementation and validation of ITM-OVER, a new sewer overflow module for the open-source Illinois Transient Model (ITM), enabling overflow simulation capabilities for the first time; (2) the development of a general, physics-based, open-source AI framework for predicting CSO location and volume under fixed control settings; and (3) the creation of an AI-powered decision support system that determines the optimal sequence of control gate operations to minimize CSOs.

To accomplish these goals, the team integrated laboratory-scale overflow experiments, Computational Fluid Dynamics (CFD) modeling, and reduced-order modeling with a PyTorch-based ML framework. These components were unified into an open-source platform named IMPACTO. The platform was successfully trained and validated using real-world data from two urban sewer systems, demonstrating high accuracy in predicting CSOs and recommending optimal control strategies—even under scenarios not included in the ML training phase.

In addition to its technical outcomes, the project made a strong educational impact. Through an established outreach program, hands-on learning modules on sewer overflow dynamics were delivered to middle-school students from underrepresented communities, promoting awareness and interest in engineering and environmental sciences.