Prediction of Water Pipeline Condition Parameters Using Artificial Neural Networks

Pipeline condition assessment needs routinely challenge water utilities in the US. Our buried water pipeline infrastructure is old and deteriorated, resulting in significant leakage and an increasing number of main breaks. The consequences of these failures are serious, thereby warranting proactive approaches to prevent them. While many technologies currently exist to support water pipeline condition assessment, they are either expensive to cover larger sections of the distribution network or intrusive, limiting their suitability. Many existing technologies also only offer a snapshot assessment at the time of inspection. Continuous monitoring networks enabled by sensors that serve multiple purposes in distribution network monitoring can support primitive pipeline condition assessment. This paper proposes and demonstrates a model-based approach for primitive prediction of pipeline condition parameters such as leakage location and severity as well as pipeline roughness. The model-based approach uses the approximate hydraulic model of the distribution network to predict pipeline condition parameters that would result in pipe flows and pressure values matching the actual pipe flows and pressure values obtained through monitoring equipment. Machine learning approaches are leveraged to predict the pipeline condition parameters in this study. The whole approach is demonstrated on a benchmark water distribution network with promising results. This approach requires synchronized monitoring data comprising pressures, pipe flows, and water consumption from distribution networks. The paper offers value to water utilities that are embracing monitoring technologies such as the AMI to transform into smart water utilities.