WaterShield Alerts

Inspiration

Water leaks and pipe bursts cause significant damage annually and waste huge amounts of clean water in residential and other buildings. In many systems such as homes, campuses, and cities, leaks go undetected until damage is already severe. We wanted to build a system that could proactively detect anomalies in water flow and alert users before small leaks turn into major failures.

What it does

WaterShield Alerts is an automated leak and burst detection system for piping networks.

It continuously monitors water flow using two flow rate sensors placed at different points in a pipe system. By comparing the flow rates in real time, the system detects inconsistencies that indicate leaks or bursts.

An AI model trained on water flow data helps determine what qualifies as “normal” vs. “anomalous” behavior. When the difference in flow rates exceeds a specified (learned) threshold, the system flags it as:

• Leak (small but persistent discrepancy)
• Burst (sudden, large discrepancy)

The system can trigger send text alerts to the user's phone instantly, enabling quick intervention and minimizing damage and water loss.

How we built it

We combined hardware sensing with data-driven modeling:

Sensors: Two flow rate sensors capture upstream and downstream water flow. Data Processing: Sensor data is streamed and compared in real time. AI Model: We trained a model on simulated/collected pipe flow data to classify normal vs. abnormal conditions and dynamically determine thresholds. Detection Logic: The system calculates the difference between the two sensor readings and uses the model to classify events. Alert System: When anomalies are detected, alerts are generated for the user/system operator. Automated valve: When a burst is detected, a motor is run to close the valve upstream. The user can press a button to unlock the valve once they assess and repair the burst themselves.

Challenges we ran into

One of the biggest challenges was distinguishing between normal fluctuations (pressure changes, changes in flow rate due to variable pipe shape) and actual leaks. The results from the model training combined with trial-and-error on the thresholds with the actual product led us to consistent constraints.

Additionally, we encountered issues in enabling the text alerts, as none of the team members had used a tool like Photon Spectrum before. The logic for triggering burst alerts (only at the first burst, to avoid redundancy) goes through arduino code on ESP32 WROOM -> python serial bridge code -> leak-alerts js (text alert) -> dashboard

Accomplishments that we're proud of

We’re proud that we built a system that goes beyond basic threshold detection by incorporating AI-informed decisions. We found that the incorporation of modular design, text alerts, the automated valve makes the technology conducive to scaling and easy retrofitting.

What we learned

We learned how powerful combining physical systems with machine learning can be. On the technical side, we gained experience with sensor integration, real-time data handling, and training models on time-series data. We also learned that system reliability depends just as much on synchronization with the actual sensor output as it does on model accuracy.

What's next for WaterShield Alerts

• Improve model sophistication with additional sensors (pressure, temp for wintertime)
• Scale up the pipe, potentially reconfigure how the water flow sensors are placed
• Create varying pipe sizes for retrofitting with different pipe systems (modularity with larger pipe networks)

Built With

• arduino
• esp32
• javascript
• photon
• python