Three-dimensional convolutional neural network for leak detection and localization in smart water distribution...

Area of Science:

• Water resource management
• Artificial intelligence in infrastructure
• Network analysis

Background:

• Advanced Metering Infrastructure (AMI) offers potential for identifying water distribution network (WDN) leaks.
• Current leak detection and localization methods for AMI systems are limited.
• Realistic leak detection requires advanced analytical tools.

Purpose of the Study:

• To propose and evaluate a deep learning (DL) model for leak detection and localization in WDNs using AMI data.
• To assess the benefits of using a 3D Convolutional Neural Network (CNN) for analyzing pressure data from WDNs.
• To compare the performance of the DL model against conventional optimization-based methods.

Main Methods:

• Development of a 3D CNN deep learning model to process spatio-temporal pressure data.
• Testing the 3D CNN model on a real WDN in Austin with simulated realistic leaks.
• Performance evaluation using metrics such as detection probability, false alarm rate, and localization accuracy.

Main Results:

• The 3D CNN model demonstrated superior performance compared to an optimization-based model in leak detection and localization.
• Deep learning offers advantages over traditional methods for WDN leak identification.
• The model's adaptability and sensitivity to hydraulic simulation errors require further investigation.

Conclusions:

• Deep learning, specifically 3D CNNs, presents a promising approach for enhancing leak detection in WDNs using AMI data.
• Further research is needed to address challenges related to model adaptability, simulation errors, and retraining for network changes.
• The findings support the integration of advanced DL techniques for more efficient water management and infrastructure maintenance.