📋 Case Study
Oil & Gas Pipeline Leak Detection SCADA Upgrade
False alarm rate > 12/day due to pressure transients, lack of sensor fusion, and no hydraulic model integration
🏗️ Project Overview
1,200 km pipeline with 22 RTU sites, upgrading from standalone RTU-based leak detection to centralized model-based SCADA analytics
🎯 Challenge
False alarm rate > 12/day due to pressure transients, lack of sensor fusion, and no hydraulic model integration
🔧 Design Approach
Deployed Python-based transient modeling engine (using EPANET API) on edge server; fused pressure, flow, temperature, and acoustic sensor streams using Kalman filtering
📐 Design Diagram
AI-generated project design illustration
📐 Key Calculations
Transient Detection Threshold
ΔP/Δt > 12 psi/sec over 3 sec window
Result: Tuned to 99.2% true positive, 0.4% false positive
Eliminated nuisance alarms from pigging operations
Edge Compute Load
Model execution time < 200ms @ 10Hz sample rate
Result: 183ms avg
Enabled real-time closed-loop leak mitigation
📊 Results
Leak detection time reduced from 8.2 min to 47 sec, false alarms dropped to 0.3/day, automatic valve isolation activated in < 90 sec💡 Lessons Learned
- •Physics-informed models outperform pure ML in low-data, high-consequence domains
- •Edge inference must be benchmarked under worst-case thermal throttling conditions
✅ Key Takeaways
- 1Physics-informed models outperform pure ML in low-data, high-consequence domains
- 2Edge inference must be benchmarked under worst-case thermal throttling conditions
📐 Prerequisites
Understand these before this topic
➡️ Next Step
Continue your engineering workflow
🔗 Engineering Applications
See how this applies across industries