zerzery

1. LeakTestbed (Aghashahi et al.)

• What it is: A fully labeled dataset from a physical laboratory-scale water distribution testbed, often referred to as LeakTestbed or “Dataset of Leak Simulations in Experimental Testbed Water Network.

• It contains 280 signals acquired on a 47 m PVC testbed with multiple sensor types: accelerometers, hydrophones, and dynamic pressure sensors, under four leak types (orifice, longitudinal crack, circumferential crack, gasket) plus no‑leak, two topologies (looped/branched), and several background demand/noise conditions.

Why it’s top‑tier for our work

• Published as a Data in Brief article with DOI and detailed documentation, making it easy to justify and cite in thesis and papers.

• Excellent for signal‑level leak detection and localization, sensor fusion (acoustic + pressure + vibration), and benchmarking supervised/unsupervised ML under controlled but realistic conditions.

2. GraphLeak (Tomazini et al.)

• What it is: A realistic simulated WDN leak dataset designed explicitly for ML, provided via the GraphLeak project on GitHub (“WDN‑Models‑and‑Data‑Sets”) and documented in the paper _“GraphLeak: A Realistic Dataset for Analyzing Leaks in Water Distribution Systems.”

• The dataset provides tabular CSV data where each column corresponds to sensor variables such as pressure, flow, volume, and leak labels/localization, generated from EPANET‑based models under diverse consumption patterns and leak scenarios.

Why it’s top‑tier for our work

• Designed for machine learning and especially deep learning / GNNs on water networks, with clear structure, labels, and reproducible simulation scripts.

• Ideal if we want to connect graph‑based models, distributed detection, or network‑wide leak localization to your smart‑grid / CPS angle, since it naturally represents WDNs as graphs.

3. (KIOS Leakage Diagnosis Benchmark)

• What it is: LeakDB is a benchmark leakage dataset for water distribution networks created by KIOS Research, distributed via GitHub.[github]​

• It consists of a large number of realistic leakage scenarios on different WDNs, with a MATLAB‑based scoring framework to evaluate leak diagnosis algorithms; the reference publication is “LeakDB: A benchmark dataset for leakage diagnosis in water distribution networks” (WDSA/CCWI 2018).[github]​

Why it’s top‑tier for our work

• Focuses on scenario‑based leakage diagnosis across multiple networks and operating conditions, which is very useful for algorithm comparison and benchmarking.[github]​

• The provided scoring and example scripts make it straightforward to compare your methods (e.g., your edge‑computing / distributed detection architecture) against existing approaches in a standardized way.[github]​

Quick comparison for choosing

Dataset Real vs simulated Main data type / features Main tasks it supports LeakTestbed Real lab testbed High‑frequency accelerometer, hydrophone, dynamic pressure signals with leak labels data.mendeleywaterfutures.githubdata.mendeley Signal‑level leak detection & localization, sensor fusion, robustness to noise data.mendeleywaterfutures.githubcolab GraphLeak Simulated WDN Tabular pressure, flow, volume, leak presence and location in graph‑structured networks githubgithubsba.org Network‑wide leak detection/localization, graph/ML models, scalable simulations githubgithubsba.org LeakDB Simulated WDN Many realistic leak scenarios on different networks, with evaluation scripts github​ Benchmarking diagnosis algorithms, scenario‑based evaluation, method comparison [github]​