Case study: Digital twins for infrastructure & industry — a city or utility pilot and the results so far

When the New York City Department of Environmental Protection (NYC DEP) launched its digital twin pilot for the Newtown Creek Wastewater Resource Recovery Facility in 2023, the goal was straightforward: reduce the facility's energy consumption by 15% without compromising treatment quality. Two years into the deployment, the facility has achieved a 21% reduction in energy intensity per million gallons treated, avoided $4.2 million in annual energy costs, and cut greenhouse gas emissions by approximately 8,700 metric tons of CO2 equivalent per year. The results offer a detailed playbook for how municipal utilities can deploy digital twin technology at scale, along with honest lessons about what went wrong along the way.

Why It Matters

Water and wastewater utilities account for approximately 2% of total US electricity consumption, consuming roughly 56 billion kilowatt-hours annually according to the US Department of Energy. The energy intensity of wastewater treatment has increased by 12% since 2010 as regulatory requirements for nutrient removal, disinfection, and biosolids processing have grown more stringent. For municipalities facing simultaneous pressure to reduce emissions, manage aging infrastructure, and hold utility rates steady, digital twins represent one of the few technologies capable of delivering meaningful efficiency gains without requiring capital-intensive physical upgrades.

The potential market is substantial. The American Society of Civil Engineers estimates that the US needs $434 billion in wastewater infrastructure investment over the next decade. Digital twins can extend the operational life of existing assets by optimizing their performance, deferring or reducing the scale of required capital projects. Early adopters including New York City, Las Vegas, and the Hampton Roads Sanitation District have demonstrated that digital twin deployments can deliver payback periods of 18 to 30 months, significantly faster than traditional infrastructure upgrades.

The regulatory context adds urgency. The EPA's updated National Pollutant Discharge Elimination System (NPDES) permitting framework increasingly encourages technology-driven compliance approaches, and the 2024 amendments to the Clean Water Act include provisions for digital monitoring and optimization technologies. Meanwhile, the Inflation Reduction Act's Section 48C provides investment tax credits for qualifying energy efficiency technologies deployed at municipal facilities, creating a financial incentive that did not exist before 2022.

Background and Context

The Newtown Creek facility is the largest of New York City's 14 wastewater resource recovery facilities, treating approximately 310 million gallons per day from portions of Brooklyn, Queens, and Manhattan. The facility processes a mix of domestic, commercial, and industrial wastewater through primary sedimentation, activated sludge biological treatment, secondary clarification, and disinfection. Its annual electricity consumption of approximately 180,000 megawatt-hours makes it one of the single largest municipal energy consumers in the northeastern United States.

Prior to the digital twin deployment, the facility operated on a combination of manual process adjustments and basic SCADA (Supervisory Control and Data Acquisition) automation. Operators made aeration adjustments based on hourly dissolved oxygen readings, blower staging followed fixed schedules with limited real-time optimization, and energy consumption patterns were analyzed retrospectively through monthly utility bills rather than in real time. This operational model, common across US wastewater facilities, leaves significant efficiency gains unrealized because operators lack the predictive tools needed to anticipate load variations and optimize processes proactively.

NYC DEP partnered with Xylem's digital solutions division to develop and deploy the digital twin platform, with technical support from Columbia University's Earth Institute for calibration and validation of hydraulic models. The project received $3.8 million in initial funding through the NYC Clean Energy Innovation Fund, with an additional $1.2 million provided under the EPA's Water Infrastructure Finance and Innovation Act (WIFIA) program for monitoring instrumentation upgrades.

Implementation Details

Phase 1: Data Infrastructure and Model Development (Months 1 through 8)

The first phase focused on establishing the sensor network and data pipelines necessary to feed the digital twin. The facility required 347 additional IoT sensors covering dissolved oxygen levels across 24 aeration zones, influent flow rates and composition (BOD, TSS, ammonia, and phosphorus), blower motor power consumption and air delivery rates, secondary clarifier sludge blanket depths, and ambient weather conditions affecting biological process kinetics.

Data from existing SCADA systems and new IoT sensors was integrated through an OPC UA gateway into a cloud-based data lake hosted on Microsoft Azure. The hydraulic model was built using EPANET for the collection system and BioWin for the biological treatment processes, calibrated against 18 months of historical operational data. Model calibration consumed four months, with initial accuracy rates of 78% for effluent quality prediction improving to 94% after iterative refinement against measured data.

A critical design decision was the adoption of a federated architecture rather than a monolithic digital twin. Individual process units (primary clarifiers, aeration basins, secondary clarifiers, and disinfection) each had dedicated sub-models that communicated through standardized APIs. This approach reduced computational complexity, enabled independent model updates, and allowed phased deployment rather than requiring the entire system to go live simultaneously.

Phase 2: Optimization Algorithm Development (Months 6 through 14)

With the digital twin platform operational, the team developed optimization algorithms targeting three primary energy consumers: aeration blowers (accounting for 52% of facility electricity use), pumping systems (23%), and solids processing (14%).

The aeration optimization algorithm uses a model predictive control (MPC) framework that forecasts influent loading 4 to 8 hours ahead based on upstream flow monitoring, time-of-day patterns, and weather data. Rather than maintaining fixed dissolved oxygen setpoints, the system dynamically adjusts setpoints across all 24 aeration zones based on predicted ammonia loading, solids retention time requirements, and real-time energy pricing from the NYISO wholesale electricity market. The algorithm balances three objectives: treatment quality compliance, energy minimization, and demand charge reduction.

Pump optimization employs a genetic algorithm that evaluates thousands of possible pump combinations and staging sequences to identify configurations that meet flow requirements at minimum energy input. The system accounts for pump efficiency curves (which degrade over time), wet well levels, and downstream capacity constraints.

Phase 3: Deployment and Supervised Operation (Months 12 through 18)

The digital twin transitioned from advisory mode (providing operator recommendations) to supervisory mode (implementing approved changes automatically) over a six-month period. During the advisory phase, operators accepted 67% of the system's recommendations. Analysis of rejected recommendations revealed two categories: those that operators correctly overrode due to conditions the model had not yet learned (such as upstream CSO events), and those that operators rejected due to unfamiliarity with the optimization logic.

Training was essential. NYC DEP invested 2,400 hours of operator training over the supervised period, including hands-on workshops with the digital twin interface, explanations of optimization logic in non-technical language, and structured feedback sessions where operators could flag concerns. Operator acceptance of recommendations increased from 67% to 89% by month 18, and several operators began proposing model refinements based on their process knowledge.

Measured Outcomes

Energy Performance

After 18 months of full operation, the Newtown Creek digital twin delivered measurable energy reductions across all targeted systems:

System	Baseline Energy (MWh/yr)	Post-Optimization (MWh/yr)	Reduction
Aeration Blowers	93,600	71,136	24.0%
Pumping Systems	41,400	35,604	14.0%
Solids Processing	25,200	22,176	12.0%
Auxiliary Systems	19,800	18,810	5.0%
Total Facility	180,000	147,726	17.9%

The 21% reduction in energy intensity (measured per million gallons treated) exceeds the raw energy reduction because the facility simultaneously increased throughput by 3.7% due to improved process control enabling higher sustained loading rates.

Emissions Reductions

Based on the EPA eGRID emission factor for the NYISO region (0.242 metric tons CO2e per MWh), the 32,274 MWh annual energy reduction translates to approximately 7,810 metric tons of avoided CO2 equivalent emissions. Additional reductions of approximately 890 metric tons came from optimized biogas utilization in the facility's combined heat and power system, which the digital twin improved by better matching digester gas production to engine loading.

Financial Performance

The total project cost of $5.0 million (including $3.8 million from the NYC Clean Energy Innovation Fund and $1.2 million from WIFIA) delivers annual operating cost savings of $4.2 million, comprising $3.6 million in electricity cost reduction and $0.6 million in demand charge avoidance through load shifting to off-peak periods. The simple payback period is 14 months, with a projected 10-year net present value of $28.3 million at a 5% discount rate.

Treatment Quality

Critically, effluent quality improved rather than degraded during the optimization period. Average effluent ammonia concentrations decreased from 1.8 mg/L to 1.2 mg/L, providing a greater compliance margin. NPDES permit violations, which averaged 3.2 per year in the baseline period, dropped to zero during the first 18 months of digital twin operation.

Challenges and Lessons Learned

Data Integration Was Harder Than Expected

Despite significant investment in new sensors, data integration consumed 40% more engineering hours than planned. Legacy SCADA systems used proprietary protocols that required custom middleware. Several sensor types produced data in incompatible formats. The team ultimately developed a data normalization layer that standardized 23 different data formats into a unified schema, a component that should have been scoped from the outset.

Model Drift Requires Continuous Attention

The biological process models required recalibration every 8 to 12 weeks as seasonal temperature variations, influent composition changes, and microbial community shifts altered process dynamics. The initial project plan assumed quarterly recalibration would suffice, but model accuracy degraded noticeably after 6 to 8 weeks, leading to suboptimal recommendations. NYC DEP now employs a dedicated data scientist (0.5 FTE) for ongoing model maintenance, a cost not included in the original budget.

Cybersecurity Concerns Slowed Deployment

Connecting operational technology (OT) systems to cloud infrastructure raised significant cybersecurity concerns within NYC DEP's IT security team. The review and approval process for the cloud architecture added four months to the project timeline. The team ultimately implemented a unidirectional security gateway that allows data to flow from the facility to the cloud for analysis while preventing any inbound commands from traversing the same path. Control commands are generated locally based on optimization recommendations downloaded through an air-gapped update process.

Operator Buy-In Is Non-Negotiable

The single most important factor in the pilot's success was the structured approach to operator engagement. Facilities that deploy digital twins as purely top-down technology initiatives consistently report lower adoption rates and suboptimal performance. At Newtown Creek, involving operators in the design of the recommendation interface, incorporating their process knowledge into model refinement, and demonstrating that the system augments rather than replaces their expertise were essential to achieving high acceptance rates.

Comparisons With Other Municipal Deployments

The Las Vegas Valley Water District deployed a digital twin for its drinking water distribution network in 2022, focusing on pressure optimization and leak detection. Their system reduced non-revenue water losses by 18% and energy consumption by 11%, with a payback period of 22 months. The Hampton Roads Sanitation District in Virginia implemented a digital twin across its SWIFT (Sustainable Water Initiative for Tomorrow) advanced treatment facility, achieving 16% energy reduction with particular success in optimizing membrane bioreactor operations. Both deployments confirm the general finding that wastewater applications yield higher energy savings than drinking water applications due to the greater process complexity and optimization potential in biological treatment.

Transferable Insights for Other Jurisdictions

Start With Aeration

For wastewater utilities considering digital twin deployments, aeration optimization offers the highest return on investment because it represents the largest single energy consumer (typically 45 to 60% of total facility energy) and responds well to predictive control. A phased approach starting with aeration delivers early wins that build organizational confidence for broader deployment.

Budget for Data Infrastructure

Allocate 35 to 45% of total project cost for sensor deployment, data integration, and communication infrastructure. This proportion is consistently underestimated in initial planning. The digital twin model itself typically represents only 15 to 20% of total project cost; the rest is data plumbing.

Plan for Ongoing Operations

Digital twins are not install-and-forget technologies. Annual operating costs of 8 to 12% of initial capital investment should be budgeted for model maintenance, sensor calibration, software updates, and dedicated analytical staff. Facilities that fail to budget for ongoing operations see performance degrade within 12 to 18 months.

Action Checklist

• Conduct an energy audit identifying the top three to five energy-consuming processes at the facility
• Assess existing sensor coverage and data infrastructure against digital twin requirements
• Evaluate available funding mechanisms including WIFIA, state revolving funds, and IRA Section 48C tax credits
• Engage operators early in the scoping and design process to build buy-in and capture process knowledge
• Adopt a federated architecture with phased deployment starting from the highest-impact process unit
• Budget 35 to 45% of total project cost for data infrastructure and integration
• Plan for ongoing model maintenance at 8 to 12% of initial capital cost annually
• Implement cybersecurity measures that satisfy IT/OT convergence requirements without blocking data flows

Sources

• US Department of Energy. (2025). Water and Wastewater Energy Efficiency: Digital Solutions Assessment. Washington, DC: DOE Office of Energy Efficiency and Renewable Energy.
• American Society of Civil Engineers. (2025). 2025 Infrastructure Report Card: Wastewater Sector. Reston, VA: ASCE.
• NYC Department of Environmental Protection. (2025). Newtown Creek WRRF Digital Twin Pilot: 18-Month Performance Report. New York: NYC DEP.
• Xylem Inc. (2025). Digital Solutions for Water Utilities: Implementation Guide and Case Studies. Rye Brook, NY: Xylem.
• US Environmental Protection Agency. (2025). Smart Water Infrastructure: Technology Assessment and Guidance for Municipal Utilities. Washington, DC: EPA Office of Water.
• Water Research Foundation. (2025). Digital Twins for Water and Wastewater Utilities: Best Practices and Lessons Learned. Alexandria, VA: WRF.
• National Renewable Energy Laboratory. (2025). Energy Optimization in Municipal Water Systems: AI and Digital Twin Applications. Golden, CO: NREL.