Summary
An East African pipeline operator required a cost-effective leak detection system for a 680 km crude oil export pipeline crossing remote terrain with variable elevations. Hypotenuse deployed a hybrid acoustic-pressure system with AI-driven signal classification, achieving < 0.5% volume leak detection and a < 6-minute alert-to-location response.
Background & Context
The pipeline traverses three national parks, two river crossings, and a mountainous section with 1,200 m elevation change. Manual patrol is conducted bi-monthly. The previous system — a basic SCADA mass-balance algorithm — had a minimum detectable leak size of 5% throughput and a detection lag of 6–24 hours. Three significant spill events had occurred in 4 years, resulting in $22M in environmental remediation costs.
Leak Physics & Detection Principles
Hydrocarbon leaks generate three detectable physical signatures:
1. Negative Pressure Wave (NPW): A leak event initiates a pressure drop that propagates in both directions from the leak point at the speed of sound in the fluid (~1,200 m/s). Cross-correlating arrival times at two sensor stations localises the leak to within ±150 m.
2. Sustained Pressure Imbalance: A steady-state leak creates a persistent differential between the expected and measured pressure profile — detectable by the hydraulic gradient deviation algorithm.
3. Acoustic Emission: Fluid escaping through a defect generates broadband acoustic energy (2–20 kHz) propagating along the pipe wall and through the soil.
Sensor Deployment
AI Detection Model
A convolutional neural network processes DAS acoustic spectrograms in rolling 30-second windows. The network is trained to distinguish leak signatures from noise classes including: pump vibration harmonics, vehicle crossings, animal activity, and rain impingement. The NPW algorithm runs in parallel, with the AI layer filtering out false positives from pressure transients caused by valve operations.
Leak location is computed by the time-difference-of-arrival (TDOA) algorithm when at least two acoustic sensors register a common event within the theoretical propagation window.
Key Outcomes & Results
Minimum Detectable Leak
0.3% throughput — 16× improvement over SCADA mass-balance
Alert-to-Location Time
Average 5.8 minutes from event to crew dispatch with GPS coordinates
False Positive Rate
1.2 per week — reduced to 0.3/week after model fine-tuning in month 4
Third-Party Interference
47 unauthorised excavation events detected in 24 months
Environmental Savings
Zero reportable spill events in 30 months since deployment
0.3%
Min detectable leak
5.8min
Alert-to-locate time
680km
Pipeline covered
47
Intrusions detected
Deployment Snapshot
Location
East Africa
Pipeline Length
680 km
Product
Crude oil
Operating Pressure
82 bar
Sensor Count
680 km DAS + 104 discrete