Please use this identifier to cite or link to this item:
Title: A machine learning-based model for real-time leak pinpointing in buildings using accelerometers
Authors: El-Zahab, Samer
Al-Sakkaf, Abobakr
Mohammed Abdelkader, Eslam
Zayed, Tarek
Affiliations: Faculty of Engineering 
Keywords: Accelerometers
Leak detection
Leak pinpointing
Neural networks
Pressurized water networks
Regression analysis
Vibration signals
Issue Date: 2023-04
Publisher: SAGE
Part of: JVC/Journal of Vibration and Control
Modern water networks from municipal network to building networks are plagued with the threat of leaks. Leaks create a significant amount of loss of resources. Pressurized water pipelines are more susceptible due to the high pressure at which water travels. Multiple researchers have tried to utilize a variety of static (devices that are left in the network) and dynamic (devices that are mobilized to the suspected location) leak detection techniques to ensure the early detection and pinpointing of leaks in water transportation networks. The main goal is to provide quick and efficient tools that can identify and pinpoint leaks in buildings while being cost-effective. This article proposes a small-scale experimental static real-time monitoring system that can identify leaks and their location with high accuracy by measuring vibration signals via wireless accelerometers. The experiment utilizes one-inch and two-inch Polyvinyl Chloride (PVC) and iron pipelines, which are commonly used in residential buildings. Since the proposed system is static, the wireless accelerometers are placed on the exterior walls of the pipelines. The vibration signals, derived from each accelerometer, were calculated and analyzed. A leak is identified when a spike in the signal is detected. Once a leak was identified, the model would move to determine the source of the signal, that is, the leak location. The developed models proved to be capable of accurately pinpointing leaks within an accuracy of 25 cm. The main techniques that were used in model development were regression analysis and backpropagation of artificial neural networks models.
ISSN: 10775463
DOI: 10.1177/10775463211066247
Ezproxy URL: Link to full text
Type: Journal Article
Appears in Collections:Department of Civil and Environmental Engineering

Show full item record


checked on Mar 25, 2023

Record view(s)

checked on Nov 28, 2023

Google ScholarTM


Dimensions Altmetric

Dimensions Altmetric

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.