Leakgeek
Case Study
LeakGeek™ is our proprietary small leak detection algorithm that integrates our Data Quality Method (DQM), Automated Data Labeling and Small Leak Detection tools
Designed to detect leaks smaller than 5% nominal flow and approximately 4-8 hours faster, LeakGeek reduces false alarms caused by poor data quality. Existing computational pipeline monitoring (CPM) systems are reliable but struggle to detect small leaks as they may create a high quantity of false positives forcing operators to shut down operations. By mitigating data quality risk, we’ve developed a scalable AI-powered solution for accurate and reliable small leak detection that has proven to reduce false positive risk by 90%+.
Summary
Our proprietary “LeakGeek” solution is an advanced leak detection system that blends machine learning with deep domain expertise to identify and prioritize small leaks in liquid pipelines. Its core strength is a proactive focus on data quality—featuring continuous SCADA system audits, automated data labeling for context, and a domain knowledge engine for precise and accurate leak detection. AI is integrated throughout the workflow as a supporting tool to enhance accuracy and reliability.
Problem / Challenge
Conventional leak detection systems typically use fixed thresholds, which can result in frequent false alarms—causing operator fatigue and inefficient use of resources. To test our solution, we worked with a leading midstream company to implement the dynamic, machine learning-based solution that learns and adapts to normal flow behavior. This approach significantly reduces false positives while improving the accuracy of real leak detection.
Results
* 90% Reduction in risk of False Positives: we’ve successfully reduced false alarms by 90% translating into significantly fewer operator interventions and a more focused, reliable alerting system.
* 4-8 Hour Leak Detection Speed Improvement: Our system now identifies leaks up to 4-8 hours faster than previous methods. This rapid response time is critical for minimizing potential economic losses and environmental impact.
* Broad Applicability: The algorithm’s effectiveness has been demonstrated across diverse pipeline geometries and profiles, including both gathering and transmission lines.
* Real-Time Validation & Continuous Learning: The integrated validation process, with its constant monitoring of quality checks, data behavior, and availability, provides a layer of reassurance that alarms detected are genuine anomalies, not simply noise.