FastDetect: Design of a performance monitoring methodology based on data analytics for large-scale water distribution systems in industrial settings
Hashim, Hafiz
Hashim, Hafiz
Loading...
Identifiers
http://hdl.handle.net/10379/17108
https://doi.org/10.13025/15748
https://doi.org/10.13025/15748
Repository DOI
Publication Date
2022-03-25
Type
Thesis
Downloads
Citation
Abstract
Large non-residential buildings can contain complex and often inefficient water distribution systems. As requirements for water increase due to water scarcity and industrialization, it has become increasingly important to effectively detect and diagnose faults in the water distribution systems in large buildings. For many of these faults, if the water supply is not impacted, water loss can go unnoticed for long periods. This can lead to unnecessary increases in water usage and associated energy losses arising from water pumping, treating, and heating. The majority of fault detection and diagnosis (FDD) studies in the water sector are limited to municipal water supplies and leakage detection. The application of FDD in building water networks remains largely unexplored. While considerable attention has been given to data driven methods that analyse and control energy systems in buildings, the same cannot be said for building water systems. This is a relatively complex and challenging research area, as the non-stationarity (variations in statistical properties over time) of water usage in non-residential buildings makes it challenging to distinguish between routine (normal) and non-routine (anomalous) water uses. As a result, methodologies which support enhanced efficiency in building water consumption are somewhat underdeveloped, particularly in industrial settings. In such scenarios, FDD methods that leverage multivariate statistical process control with, for example, principal component analysis can be successfully used to identify faults. This research leverages two case-studies, to develop and demonstrate a methodology based on component analysis (PCA) and a multi-class support vector machine (SVM) to detect and classify faults for non-residential building water networks. In the absence of a process model (which is typical for such water distribution systems), PCA is proposed as a data-driven fault detection technique for building water distribution systems for the first time herein. PCA detection indices (T2-statistics and Q-statistics) were employed to detect faults in the incoming data, and a multi-class SVM was trained for fault classification. However, even with these methodologies, the non-stationarity of water uses can lead to false alarms being generated. Historically, issues such as false alarm prevalence has led to a relatively low industry uptake of FDD systems. Thus, to efficiently detect and diagnose faults in non-residential building water distribution systems in a manner which is practical, false alarms should be controlled through false alarm moderation approaches so that building managers/operators only need to focus on true faults. This research utilises two statistical non-parametric false alarm moderation approaches (namely window-based, and trial-based) that generate a second control limit for T2-statistics and Q-statistics. The implementation of these false alarm moderation approaches was combined with PCA to detect true faults. Using both the window-based and trial-based approaches false alarms were reduced greatly, and the overall performance and reliability of the overall approach was improved. The PCA model with the window-based approach was shown to be particularly effective in reducing false alarms during fault detection process. Despite the relatively limited training data available from the case-studies (which can often reflect the situation in many buildings), meaningful faults were detected. The designed FDD methodology (FastDetect) proved successful in discriminating between various types of faults in the case-studies. The effectiveness of FastDetect was compared to a conventional univariate threshold technique through comparison of their respective performance in the detection of faults that occurred in the case-study sites. FastDetect demonstrated promising capabilities when compared to the conventional, in-situ, fault detection systems. The multi-class SVM also allowed these faults to be classified, providing a greater level of information to building managers, which may avoid unnecessary emergency shutdown in industrial applications. Finally, a comprehensive investigation of the energy and economic impact of water use in the case-study site 1 was conducted. This assessment was then used to develop a cost-benefit analysis for the implementation of FastDetect in industrial settings. Given the links between water and energy systems, faults in a water distribution system can impact overall building energy use and associated carbon footprint. Thus, addressing water and associated energy losses resulting from faults and inefficiencies in water distribution systems can positively impact upon the overall sustainability of non-residential buildings, reducing life-cycle carbon footprint.
Funder
Publisher
NUI Galway