🌊 SubseaAI — Real-Time Pipeline Leak Detection (<15s) AI system that detects and localizes subsea oil & gas pipeline leaks in seconds using multi-sensor fusion. 🚨 Problem Subsea pipelines represent a $5.8T global infrastructure. Leaks cause $10B+ annual losses and severe environmental damage. Current detection: Manual ROV inspection Takes hours to days 👉 Critical issue: Detection is too slow 💡 Solution SubseaAI continuously monitors pipelines and detects leaks in <15 seconds. It combines 4 sensor modalities: Thermal (FBG + DTS fiber optics) Acoustic (hydrophones) Pressure Then applies AI fusion + physics models to: Detect leaks Estimate location Predict severity ⚙️ How It Works (Simple) Sensors stream real-time data Physics models simulate leak behavior AI detects anomalies across all signals Fusion engine confirms leak probability System outputs: Leak detected Location (km) Severity level 🧠 Key Innovation Multi-Modal AI Fusion Uses: entity["scientific_concept","Dempster–Shafer theory","evidence fusion method"] for combining sensor evidence entity["scientific_concept","Kalman filter","state estimation algorithm"] for leak tracking entity["scientific_concept","Isolation Forest","anomaly detection algorithm"] for anomaly detection 👉 Result: Higher accuracy than single-sensor systems 🧪 Live Demo ✔ Inject leak at any pipeline location ✔ System detects anomaly in real time ✔ Dashboard shows: Probability of leak Node location Alert severity 🏗️ Architecture Physical Layer (Simulated): 100 km pipeline 100 nodes 64 FBG sensors/node DTS fiber (temperature profile) Hydrophones + pressure sensors Software Layer: FastAPI backend (real-time API + WebSocket) AI detection engine Digital twin prediction model Live dashboard (heatmap + alerts) 🔬 Technical Highlights Thermal plume modeling (Gaussian dispersion) Acoustic leak signature (1/r² propagation) Pressure drop modeling (Bernoulli equation) Real-time anomaly detection Kalman-based localization 🌍 Real-World Impact Prevents $100M+ loss per major leak Reduces detection time from hours → seconds Applicable to: Offshore oil & gas Industrial pipelines Smart infrastructure monitoring ⚠️ Note This is a high-fidelity simulation of a real deployment system, including: Sensor physics Edge processing logic AI detection pipeline Designed to be deployable with real hardware.
🔬 Physics & Mathematical Modeling SubseaAI is not a generic AI model — it is a physics-informed system where detection is grounded in real-world fluid, acoustic, and thermal behavior. 🌡️ 1. Thermal Plume Modeling (Leak Heat Propagation) Leakage of hot hydrocarbons creates a buoyant thermal jet in cold seawater. We model this using Gaussian plume dispersion: Where: � = leak mass flow rate � = plume spread (depends on ocean current + distance) � = distance from leak 👉 This allows: Detection of thermal gradients across FBG sensors Spatial heat profiling via DTS fiber 🔊 2. Acoustic Leak Signature (Underwater Physics) Leaks generate turbulent broadband acoustic noise (0.1–50 kHz). Modeled using inverse-square propagation: Where: � = distance to leak � = leak rate We extract: Broadband energy Cavitation indicators Time Difference of Arrival (TDOA) 👉 Used for triangulating leak location ⚙️ 3. Pressure Drop Modeling (Fluid Dynamics) Pipeline pressure follows Bernoulli-based flow loss: Where: � = pipe diameter � = leak rate Includes: Pump pulsation noise Transient pressure decay 👉 Detects rupture-level events quickly 🧵 4. Fiber Optic Sensing Physics FBG (Fiber Bragg Grating) Temperature from wavelength shift: With aging compensation: 100 ppm/year drift correction applied DTS (Distributed Temperature Sensing) Based on Raman scattering: 👉 Provides continuous temperature profile along pipeline 🧠 5. AI Fusion (Mathematical Decision Layer) We combine all sensor evidence using Dempster–Shafer theory: Where: � = thermal, acoustic, pressure signals � = reliability weights � = sigmoid normalization 👉 Handles uncertainty across sensors 📍 6. Leak Localization (Kalman Tracking) We estimate leak position using Kalman filter: Combines: Acoustic triangulation Thermal centroid State update: 👉 Produces smooth, real-time location estimates 🤖 7. Anomaly Detection (Unsupervised Learning) Using Isolation Forest: Trained on normal pipeline behavior Detects deviations without labeled leak data 👉 Enables real-world adaptability ⚡ Implementation Highlights Real-time pipeline: 100 nodes over 100 km Sensor simulation at 10 Hz streaming Async processing with FastAPI + WebSockets SQLite WAL for persistent event logging Modular architecture: Physics → Sensors → AI → Alerts
Log in or sign up for Devpost to join the conversation.