Data-driven application of MEMS-based accelerometers for leak detection in water distribution networks

Area of Science:

• Environmental Science
• Engineering
• Data Science

Background:

• Global water scarcity is a critical issue, exacerbated by significant leaks in water distribution networks (WDNs).
• These leaks result in substantial financial losses and environmental damage, intensifying the need for effective leak detection technologies.

Purpose of the Study:

• To investigate the efficacy of cost-effective Micro-Electro-Mechanical Systems (MEMS)-based accelerometers for detecting leaks in WDNs.
• To develop and validate machine learning models for accurate leak classification in both metal and non-metal pipes.

Main Methods:

• Experiments were conducted over ten months on real WDNs (metal and non-metal pipes).
• Acceleration signals were analyzed using a monitoring algorithm to extract features for leak and no-leak cases.
• Four machine learning models (KNN, Decision Tree, Random Forest, Adaboost) were developed and trained for leak classification.

Main Results:

• The Random Forest model demonstrated superior performance, achieving 100% accuracy for metal pipes and 94.93% for non-metal pipes.
• Machine learning models were validated on unseen data, confirming their high effectiveness in leak detection.
• The study established real-world, network-based machine learning models for WDN leak detection.

Conclusions:

• MEMS-based accelerometers are a viable and cost-effective technology for detecting leaks in water distribution networks.
• Machine learning models, particularly Random Forest, significantly enhance the accuracy and reliability of leak detection systems.
• This research addresses a critical gap in WDN leak detection, offering practical solutions to mitigate water loss and scarcity.