📋 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

Oil & Gas Pipeline Leak Detection SCADA Upgrade Challenges • >12 false alarms/day • No sensor fusion • No hydraulic model Edge Server Python + EPANET API Model exec: 183ms @ 10Hz P Q T A Sensor Fusion Kalman Filter Leak Alert Logic ΔP/Δt > 12 psi/sec 3-sec window → 99.2% TP, 0.4% FP Hydraulic Model Integration

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