Kenneth ChengMathematical modeling

Editorial project overview

Resilient Water-Network Leak Localisation

Can pressure sensors selected for lower-tail leak-signature separation retain localisation performance when measurements are noisy and one sensor may fail?

Exploratory studyUpdated 13 Jul 2026
Technical figure from Resilient Water-Network Leak Localisation
A reviewed research figure synchronized from ScienceProject. The original aspect ratio and labels are preserved. Source and generation notes.

Overview

This study treats pressure-sensor placement as a fault-tolerant design problem. Instead of asking only whether leak signatures are distinguishable when all sensors work, it scores each layout after every possible one-sensor removal.

Why the problem matters

A leak-localisation pipeline can depend on a small number of pressure sensors. A layout chosen for the intact system may become ambiguous during communications loss, maintenance, power interruption, or sensor rejection. Redundancy therefore needs to be represented in placement, not added only after a failure occurs.

Model, data, and assumptions

The pilot uses a 36-node synthetic meshed network with one fixed-head reservoir and 35 candidate leak zones. A damped graph Laplacian supplies hydraulic-inspired pressure signatures. Seven sensors are chosen by random placement, node degree, nominal greedy separation, or the proposed dropout-aware greedy objective.

Leak magnitude varies from 0.65 to 1.35. The benchmark adds five relative noise levels and three probabilities of losing one sensor. Localisation fits a non-negative magnitude to every candidate signature and ranks the residuals.

Validation and findings

Every method-condition pair contains 1,225 trials. Common random seeds make the strategy comparison paired at each condition. The designated stress test shows a modest aggregate top-1 improvement and a larger worst-zone improvement over the nominal ablation. At 12% noise and 40% dropout, however, the two greedy methods are effectively level; extreme-noise fairness remains unresolved.

Reproduction

The technical repository contains the source, tests, complete metric table, approved vector figures, long-form technical write-up, and a project-local reproduction guide. A natural follow-up would be to replace the graph proxy with EPANET benchmark hydraulics and add demand and topology uncertainty.

Key findings

  • Under the designated 8% noise and 40% single-sensor-dropout stress test, the proposed layout reached 88.8% top-1 accuracy versus 85.9% for its nominal ablation.
  • Worst-zone accuracy in that stress test was 65.7% for the proposed layout and 51.4% for the nominal ablation.
  • The advantage was not uniform across all noise levels, so the current evidence supports benchmark extension rather than a field-performance claim.

Limitations

  • The graph-Laplacian response is a controlled proxy rather than an EPANET or field-calibrated hydraulic model.
  • Only one unavailable sensor, independent Gaussian noise, and one leak at a time are represented.
  • Aggregate Wilson intervals do not replace a paired significance analysis or account for every within-zone dependence.

Technical record

Detailed source, calculations, generated figures, and reproduction instructions remain in ScienceProject. Open the technical project.

Version history

2026-07-13 — Curated overview reviewed against repository evidence.