Authors: Amol Vitthalrao Paul, Professor Bangar Abhishek Popat, Professor S. B. Bhosale, Professor Karad V.A.
Abstract: Unaccounted-for water loss resulting from undetected underground pipe leaks poses a persistent operational and financial challenge for municipal distribution systems worldwide. In conventional utility practice, leaks are predominantly identified reactively—either when water eventually breaches the surface or following localized pressure drop complaints from consumers. In this paper, we develop a proactive, multi-sensor predictive leakage detection framework that pairs a low-power edge Internet of Things (IoT) architecture with a hybrid machine learning pipeline. Subsurface pipe segments are monitored using five complementary sensing modalities: hydraulic pressure, inline flow velocity, pipe-wall structural vibration via a MEMS accelerometer, soil moisture, and ambient temperature. Sensor readings are acquired, pre-filtered, and transmitted by an energy-efficient ESP32-based edge node using LoRa wireless communication and TLS-secured MQTT to a central analytics backend. A sliding-window feature extraction stage derives short-term statistical trends and feeds an unsupervised Isolation Forest model; the resulting anomaly metric is concatenated with temporal features to train a gradient boosting classifier that forecasts the likelihood of a leak initiating within a two-hour predictive lead window (H = 12 steps). We evaluate the proposed approach on a synthetic distribution network of 60 pipe segments that explicitly incorporates gradual, incipient mechanical degradation prior to burst onset. Benchmarked against a conventional static pressure threshold (F1 = 0.830) and standard supervised classifiers, the proposed hybrid model achieves an F1-score of 0.992 and provides a median warning lead time of 1.9 hours (mean 1.8 hours) ahead of physical rupture, with zero false alarms and zero missed events across held-out test segments. Finally, we analyze model sensitivity across noise regimes and establish key deployment requirements for district metered area field implementations.
