This project presents a low-cost, deployable, AI-integrated platform for monitoring fecal contamination in natural waters. By using inorganic water quality parameters (e.g., DO, PH, turbidity, conductivity) as surrogate inputs, the system predicts E. coli concentration and collects physical water samples for future validation. It integrates real-time sensing, wireless transmission, machine learning inference, data visualization, and automated water sampling, forming a complete perception–prediction–collection loop.
- Title: Integrated AI-driven Monitoring and Sampling System for Fecal Contamination in Urban Waters
- Core Idea: Replace expensive, time-delayed E. coli testing methods with an inorganic-indicator-based prediction model
- Application Scenario: Urban rivers (e.g., Thames), lakes, and distributed low-resource field sites
+------------------+ LoRa +-------------------+
| Floating Buoy |----------------->| LoRa Gateway |
| (Sensors + MCU) | | (MQTT Broker) |
+--------+---------+ +-------------------+
| |
| LoRa v MQTT
+---------v---------+ +---------------+
| Autonomous Sampler| | Cloud Server |
| (RTC + Pump + |<--------------------->| Data Storage |
| Bottle Rotation) | +------+--------+
+-------------------+ |
v
+-------------------------------+
| Web Dashboard & Mobile App |
| (Real-time + Historical) |
+-------------------------------+
- Turbidity, Dissolved Oxygen, PH, conductivity sensors
- Arduino-based circuit with solar power
- Periodic sampling + LoRa transmission
- Peristaltic pump + rotating vial mechanism
- RTC-controlled periodic water collection
- Time-synced with sensor data via LoRa
- LoRa gateway + MQTT + Python backend
- Database (SQLite)
- Web frontend with charts and map
- Mobile app with charts and map
- Trained neural network (FCN) on public water data (~23GB → 1MB preprocessed)
- Predicts log-scaled E. coli concentration
- RMSE ≈ 0.83 on test set
- Optional model variants: XGBoost, RF for comparison
- Fully solar-powered, suitable for field deployment
- Wireless transmission via LoRa (long range, low power)
- Real-time + historical data access
- Expandable to swarm systems or RL-based control
- ✅ First attempt to estimate fecal contamination using only inorganic indicators
- ✅ End-to-end system: sensing, prediction, visualization, sampling
- ✅ Open-source, low-cost, and designed for global scalability
- ✅ Modular and extensible: supports AI upgrades, edge deployment, or swarm coordination
- Integrate control strategies (e.g., RL-based adaptive sampling)
- Train with real E. coli lab data using automated sampler
- Edge deployment of ML model on MCU (TinyML)
- Multi-node deployment and anomaly correlation
Root/
├── 3D_Models/ # CAD models
├── Arduino_Code/ # Embedded firmware
├── Backend_code/ # Backend services, APIs
├── MLcode_Tommy_try/ # Experimental ML models by "Tommy"
├── ML_Main/ # Main ML pipeline: data, preprocess, training, model
├── Reports/ # Project documentation, research papers, final reports
└── Visualization/ # Frontend interfaces for data visualization and monitoring
├── APP/ # Mobile app interface "BlueGuardian"
└── Web/ # Web dashboard
Zhuohang Wu @ zhuohang2024@163.com
For technical inquiries, collaboration requests, or feedback, feel free to contact:
📌 Project Lead:
Zhuohang Wu (John) – zhuohang2024@163.com
👥 Team Members:
Chenghao Xin, Xueyin Fan
🧑🏫 Supervisors:
- Dr. Akin Delibasi – Department of Computer Science, UCL
- Dr. Valerio Modugno – Department of Computer Science, UCL
- Dr. Izzy Bishop – Department of Genetics, Evolution & Environment, UCL