Water supply networks, a critical infrastructure of modern cities, were intensively installed during the 1970s and 1980s. Their rapid deterioration has led to severe social and economic costs caused by pipeline leakage and corrosion. As of 2020, the a...
Water supply networks, a critical infrastructure of modern cities, were intensively installed during the 1970s and 1980s. Their rapid deterioration has led to severe social and economic costs caused by pipeline leakage and corrosion. As of 2020, the average leakage rate in South Korea reached 10.4%, resulting in an economic loss of approximately 658.6 billion KRW. While conventional leak detection methods, such as acoustic and correlation techniques, are widely used, they possess distinct limitations as reactive and discontinuous management approaches that respond only after a leakage incident has occurred. Consequently, the ultimate objective of this study was to develop the AIoT-based Buried Pipelines Integrated Management System (ATIMS). By converging Internet of Things (IoT) and machine learning technologies, this system aims to detect and predict hazardous precursors—specifically leakage, corrosion, and breakage—in underground water supply networks prior to the occurrence of actual incidents.
To achieve this objective, the entire research process was systematized into three core components. These components were not developed in isolation but were designed to establish a feedback loop system where they continuously interact with one another. First, for fundamental data acquisition, four key soil factors determining the corrosion environment of water networks were selected: soil resistivity, water content, pH, and Oxidation Reduction Potential (ORP). Subsequently, a multi-purpose IoT sensor capable of measuring these soil factors in real time was developed. This sensor was designed with durability to withstand underground conditions and equipped with an LTE communication module to enable the remote transmission of collected data.
Second, to construct a high-quality training dataset essential for algorithm development, a hybrid dataset construction strategy was adopted. Raw data collected from the field underwent preprocessing, including Box–Cox transformation, to eliminate sensor-specific bias and skewness and to ensure statistical significance for future predictions. Based on this refined hybrid dataset, a Deep Multi-Layer Perceptron (DMLP) model was applied to learn the complex non-linear relationships between the four soil factors and pipeline risks. Finally, the ATIMS software (S/W) was constructed to implement and apply the developed sensors and algorithms in the field.
As a result of verification, the leak status prediction model achieved optimal accuracy with a coefficient of determination (R²) of 0.99. Furthermore, the system achieved excellent performance indicators: 82% (R² = 0.82) for leak location prediction, 92% (R² = 0.92) for corrosion thickness prediction, and 93% (R² = 0.93) for corrosion rate prediction. These results confirm that the proposed algorithms possess sufficient effectiveness as tools for diagnosing and managing anomalies in water supply networks.
In the final phase of the study, an integrated platform system (S/W) encompassing the entire process—from sensor data collection to the visualization of machine learning analysis results—was established, and its completeness was demonstrated through copyright registration. This study presents an end-to-end integrated solution where hardware and software are seamlessly connected in water supply network management. As one of the pioneering studies verified with actual field data, it proposes a practical approach to shift the existing paradigm of qualitative and reactive management toward a quantitative and predictive management system.