This study capitalizes on a dataset, originally including 280 sensory measurements from a laboratory-scale water distribution system, to advance the concept of leakage diagnosis and localization. The water distribution test rig are formulated in two configurations, namely looped and branched layouts. The paper processed time-domain data from accelerometers and dynamic pressure sensors into advanced statistical features of: Autocorrelation Coefficient (Au-C), and Signal Energy (Sig-E), to detect and localize the water leakage. By the Employment of these two features, the research developed an expert system of an Artificial Neural Network (ANN) model designed with optimal parameters, neurons, and hidden layers to classify the presence and pinpoint the location of leaks within the water test rig. The effectiveness of the current approach is quantitatively evaluated using F1-scores and accuracy metrics. A robust capability for both detecting and localizing leaks under varying conditions was established with a highest accuracy and F1-score of 86.5 % and 86.2 %, respectively. The findings underscore the potential of integrating advanced features with Artificial Intelligence (AI) in enhancing the reliability and dependability of water management expert systems. This approach contributes to the broader application of AI in managing water resources and infrastructure resilience with its support to improve leakage whereabouts.
Read full abstract