Online Monitoring Framework for Pressure Transient Detection in Water Distribution Networks

pressure transients;water distribution networks;condition monitoring;internet of things;smart infrastructure;statistical process control;anomaly detection

Feng, Ninaadvisor:Narasimhan, Sriram ; affiliation1:Faculty of Engineering ; Narasimhan, Sriram ;

University of Waterloo

关键词: pressure transients;    water distribution networks;    Master Thesis;    internet of things;    condition monitoring;    anomaly detection;    statistical process control;    smart infrastructure;   

Others  :  https://uwspace.uwaterloo.ca/bitstream/10012/14549/1/Feng_Nina.pdf

瑞士|英语

来源: UWSPACE Waterloo Institutional Repository

【 摘 要 】

Access to potable drinking water is a necessity and basic human right. Most North Americans obtain treated water through water distribution networks, an essential part of municipal infrastructure that is subject to decay and degradation. Amongst the factors influencing pipe failure are events that trigger abrupt pressure changes, or transients, which can cause pipe breakages in the short term, and general fatigue in the long term. The ability to quantify these transients as they occur is important for effective asset management, and for preventing and mitigating the occurrence of failure. Current practices take a largely reactive approach to event detection, and few systems capable of real-time transient detection have ever been implemented.This research addresses the need for an online monitoring framework aimed towards understanding pressure transient effects and behaviour. The proposed system uses an Internet of Things approach, combining pressure sensors with Raspberry Pi computers, as well as open-source tools that transmit and display the data. The data analysis combines computationally inexpensive methods in order to achieve an accurate decision-making tool for both transient detection and abnormal transient risk identification. The techniques used include different filtering and detrending methods, feature extraction for dimensionality reduction, three-sigma statistical process control, and classification using voting methods. The process also includes a second process, based on statistical process control and trained using transient data identified in the original process, in order to assign a risk for a transient to cause damage, as well as identify transients that are particularly severe.Data was collected from a unique laboratory water distribution network as well as a field installation in Guelph, Ontario. The results showed that the framework achieves real-time transient identification with reasonable detection and error rates. Further analysis illustrated the effect of factors such as transient source location, active flow in the pipes, and transient type, on transient propagation and detection. The performance of the framework proves the concept of IoT-based systems for pressure monitoring and event detection in municipal water infrastructure.