Built an IoT-powered water utility management platform with ML leak detection that reduced non-revenue water losses from 28% to 11% and cut leak detection time from 45 days to 4 hours.
The Hidden Crisis in Water Distribution
Water distribution infrastructure in many municipalities was built decades ago and has deteriorated significantly, creating a pervasive problem of non-revenue water—treated water that enters the distribution system but never reaches a paying customer due to leaks, pipe bursts, metering inaccuracies, and unauthorized consumption. AquaFlow Municipal Water served a metropolitan area of 800,000 residents through 3,200 miles of distribution mains, and their non-revenue water rate stood at 28%—meaning more than a quarter of all treated water was being lost before reaching customers. The utility relied on manual meter reading conducted bimonthly, which meant consumption anomalies that might indicate leaks were detected weeks after they began. Leak detection was primarily reactive, depending on visible surface water, customer complaints, or catastrophic pipe failures that disrupted service. The utility employed a small team of leak detection specialists who used acoustic listening equipment to survey pipes manually, but at their pace they could only cover the entire network once every three years. Pressure management was rudimentary, with fixed pressure zones that delivered excessive pressure during low-demand periods, accelerating pipe deterioration and increasing leak volumes. Customer service was overwhelmed with billing disputes because bimonthly meter readings could not distinguish between gradual leaks on customer properties and sudden usage spikes. The utility faced increasing pressure from state regulators to reduce water losses and from ratepayers who questioned rising costs while visible infrastructure problems persisted. Climate conditions in the desert Southwest made water conservation particularly critical, adding urgency to the need for a comprehensive monitoring and management solution.
- 28% non-revenue water rate representing over 1.2 billion gallons of treated water lost annually
- Manual bimonthly meter readings meant leak detection averaged 45 days from onset
- Acoustic leak detection team could only survey the full 3,200-mile network once every 3 years
- Fixed pressure zones delivered excessive pressure during low-demand periods, accelerating deterioration
- Customer billing disputes from bimonthly readings could not identify gradual leaks vs usage changes
- Desert climate made every gallon of water conservation critical for sustainability
Intelligent Water Network
We designed AquaFlow as a comprehensive IoT platform connecting thousands of sensors throughout the water distribution network to a cloud-based analytics engine. Acoustic leak detection sensors were deployed at strategic intervals along distribution mains, listening continuously for the distinct sound signatures of water escaping from pipes. These sensors transmit data to AWS IoT Core, where a TensorFlow machine learning model trained on thousands of confirmed leak recordings classifies sounds in real time, distinguishing actual leaks from normal hydraulic noise with 94% accuracy. Smart meters installed at customer connections provide 15-minute interval consumption data transmitted via LoRaWAN, replacing bimonthly manual readings and enabling immediate detection of abnormal usage patterns. Pressure sensors throughout the network feed into a dynamic pressure management system that adjusts zone pressures based on real-time demand, reducing excess pressure during low-demand periods to minimize pipe stress and leak volumes. All sensor data flows through Apache Kafka into TimescaleDB, a time-series database optimized for the massive volumes of IoT telemetry generated across the network. The analytics platform applies predictive maintenance algorithms that correlate pipe age, material, soil conditions, pressure history, and leak frequency to prioritize capital replacement programs for maximum impact. A customer self-service portal provides real-time usage dashboards, leak alerts, conservation tips, and the ability to set usage budgets with automated notifications. The operations dashboard gives utility engineers a network-wide view with drill-down capability to individual pipe segments, overlaid with leak probability heat maps and maintenance work order management.
- Acoustic leak detection sensors deployed across 3,200 miles of distribution mains with 94% accuracy
- Smart meters providing 15-minute interval data via LoRaWAN replacing bimonthly manual readings
- Dynamic pressure management adjusting zone pressures based on real-time demand patterns
- TensorFlow ML model distinguishing leak signatures from normal hydraulic noise
- Predictive maintenance algorithms prioritizing capital replacement based on multi-factor risk scoring
- Customer self-service portal with real-time usage dashboards and automated leak alerts
- TimescaleDB handling billions of time-series data points from network-wide IoT sensors
Our Approach
Every Drop Counts
The impact of AquaFlow's deployment transformed the utility's operational efficiency and water conservation outcomes. Non-revenue water decreased from 28% to 11% within the first 18 months—a reduction that translated to saving over 1.2 billion gallons of treated water annually. Leak detection time plummeted from an average of 45 days to just 4 hours, with the acoustic sensor network identifying leaks almost immediately after they begin and automatically generating prioritized work orders for repair crews. Dynamic pressure management reduced average network pressure by 15% during low-demand periods, which both decreased leak volumes and extended pipe lifespan. Customer billing disputes decreased by 72% as 15-minute interval data provided transparent, granular usage information that could clearly identify the timing and nature of consumption changes. The predictive maintenance model identified 23 high-risk pipe segments that were replaced proactively, preventing an estimated 15 main breaks that would have disrupted service to thousands of residents.
Return on Investment
Technologies Used
Integrations
AquaFlow has given us visibility into our distribution network that we never thought possible. We went from losing a quarter of our water to operating at near best-in-class efficiency.
Project Gallery
Lessons Learned
- Sensor placement optimization using hydraulic modeling was critical—random deployment would have required 3x more sensors for equivalent coverage
- The ML model needed location-specific tuning because pipe material and soil conditions significantly affect acoustic signatures
- Customer portal adoption increased 4x when we added proactive leak alerts that helped customers find property-side leaks before they became costly
Summary
Advenno deployed an IoT-powered water utility management platform with ML leak detection, smart metering, and dynamic pressure management that reduced non-revenue water from 28% to 11% for a municipal utility serving 800,000 residents.
Key Takeaways
- Acoustic ML leak detection reduced detection time from 45 days to 4 hours
- Non-revenue water dropped from 28% to 11%, saving 1.2 billion gallons annually
- Dynamic pressure management reduced average network pressure 15% during low-demand periods
- Smart meters with 15-minute interval data cut billing disputes by 72%
- Predictive maintenance prevented 15 main breaks through proactive pipe replacement
Frequently Asked Questions
Key Terms
- Non-Revenue Water
- Treated water that enters a distribution system but is never billed to customers, lost through leaks, pipe bursts, metering errors, or unauthorized consumption.
- Acoustic Leak Detection
- A technology that uses sensitive microphones or hydrophones to listen for the distinctive sound frequencies generated by water escaping from pressurized pipes through cracks or joints.
Facts & Statistics
Sources & Citations
Ready to Reduce Water Loss?
Deploy IoT monitoring and ML-powered leak detection to transform your water distribution efficiency and customer service.
Discuss Your Project
