A Robust Hybrid IoT Framework for Post-Disaster Water Leakage Detection under Uncertain Sensing Conditions

Abstract

Water distribution networks are highly vulnerable during natural disasters, where pipe damage and hydraulic instability can lead to hidden leakages, service disruption, and increased flood risk. Conventional IoT-based leakage detection systems perform well under stable conditions but deteriorate when exposed to disaster-induced noise, missing data, and irregular pressure-flow behaviour. This paper presents a modular IoT-driven framework specifically designed for post-disaster environments. The framework augments benchmark leakage datasets with synthetic anomalies, such as pressure spikes, flow surges, sensor noise, and data outages, to emulate realistic flood- and earthquake-induced instability. A hybrid detection model combining rule-based logic with an unsupervised anomaly-detection module is then applied to extract and evaluate key hydraulic features. Experimental results from simulated disaster scenarios demonstrate that the hybrid model achieves higher detection rates, reduced false-alarm occurrences, and improved temporal stability compared to threshold-only detection. These findings indicate the effectiveness of the proposed approach for rapid leakage identification and early flood-risk mitigation in disaster-affected water networks.