Effects of error propagation of uncertainties on pressure-based leakage localisation

Water distribution networks are increasingly equipped with measurement devices for real-time monitoring of hydraulic parameters, including permanent pressure sensors distributed in the network. Information from these devices can be utilised, among others, for pressure-based leakage localisation, which aims to identify a specific area of the network where a leak might have occurred, and assist refined on site leakage pinpointing. The effectiveness of leak localisation methods is thereby influenced by several factors and their associated uncertainties. For example, errors in available network data affect first the optimal sensor placement, second the hydraulic model calibration, and finally the accuracy of spatial localisation of a leakage. Yet, a systematic analysis including a quantification of the propagation of errors through the sub-processes included in pressure-based leakage localisation is still missing in literature.

The aim of this work is to combine different types of uncertainties in pressure-based leakage localisation to systematically investigate the effects of error propagation through the sub-processes. The following sub-processes are implemented in the error propagation analysis (the considered uncertainties of each subprocess are added in brackets): (1) creation of a hydraulic model or network graph based on GIS data (network topology, pipe diameters, pipe roughness, nodal demand), (2) selection of sensor placements (number of sensors), (3) model calibration during a leakage-free period (measurement errors), and (4) leakage localisation (measurement errors). In this work, both data-driven (i.e., graph-based state interpolation, differential pressure analysis) and model-based (i.e., sensitivity matrix, graph-based genetic algorithm) leak localisation methods are implemented for comparison. Both the L-Town benchmark network from the “Battle of the Leakage Detection and Isolation Methods” and a real-world WDN with engineered leakage events are utilised as demonstrative case studies. The leak localisation performance is evaluated by the pipeline distance between the assumed leakage location and the real leakage location.

The preliminary results show that model-based methods are substantially more accurate than data-driven methods under perfect conditions, i.e., perfectly calibrated hydraulic model and no measurement errors. However, model-based methods are also more affected by errors in the GIS data, as an accurate hydraulic model has a major influence on the accuracy. In the next steps, other uncertainties will be systematically added across all defined sub-processes in bandwidths defined by literature values, and their joint influence on the effectiveness of pressure-based leakage localisation will be analysed. These findings can then be used to optimise the quality of data collection strategies based on their relative importance, ultimately leading to an improvement in pressure-based leak localisation in science and practice.

FUNDING

This publication was produced as part of the “FOUND” project. This project is funded by the Federal Ministry of Agriculture, Forestry, Regions and Water Management (BML) (Austria) (Project C300198).