A critical systematic review of machine learning based models for water leak detection using vibroacoustic technology

Detecting leaks in water pipes is a critical task for water utilities, as it helps prevent water loss and damage to infrastructure. Machine learning (ML) techniques have shown potential in detecting water pipe leaks, but there are gaps and limitations in the ML modeling process that need to be addressed. This study presents a comprehensive evaluation of ML-based techniques for leak detection in water pipes, discussing the general steps to develop an ML-based model, including data collection, data processing, and model development. The study reviews a diverse array of ML-based methods, from traditional ML techniques to advanced deep learning approaches, and introduces their applicability and limitations. Finally, current research limitations and gaps are summarized following the reviews, and potential future directions are suggested, including overcoming data scarcity, enhancing current model performance and interpretability, assessing the potential of various ML-based approaches, and improving the practicability of water management for relevant researchers, engineers, and managers. This study provides a reference for academics and industrial practitioners interested in ML-based water leak detection models, especially in signal processing techniques, model selection and development, and practical applications. The findings contribute to understanding water pipe leak detection, promoting the development of water leak detection technologies, advancing the current water leak management, enhancing the decision-making process, and benefiting water utilities, academics, and end users.