A Novel Computationally Efficient Asset Management Framework Based on Monitoring Data from Water Distribution Networks

Drinking water infrastructure in the U.S. is in a deteriorated state needing immediate intervention that is sustainable. Although many technologies are being developed to inspect buried pipeline assets, they are still expensive and human-dependent to use for comprehensive condition assessment and prioritization of the most critical assets for immediate rehabilitation and replacement planning. This paper presents a novel system-level condition assessment framework where monitoring data from distribution infrastructure is leveraged to predict the condition of assets using evolutionary optimization and machine learning algorithms. Pipeline roughness values and effective hydraulic diameters (given the possibility of graphitization/corrosion) are two parameters that would reveal their overall condition, and therefore these two parameters will be used to demonstrate the framework presented in this paper. In this respect, a modified benchmark water distribution network is used to represent an ageing, deteriorated network by randomly reducing effective pipe diameters and roughness coefficient values. Subsequently, a novel reverse engineering optimization method is leveraged to minimize the mean square errors of operational parameters (e.g., pressure and flow) via both predicted (through optimization) and modeled data obtained from a given set of monitoring stations. Roughness values and effective hydraulic diameters are the decision variables in this optimization framework that are to be predicted. EPANET 2.0 software is used for modeling the water distribution network performance in this study. Faster convergence is achieved through fine-tuning of genetic algorithm properties. Specifically, the computational efficiency and prediction accuracy benefits derived from appropriately narrowing down on the upper and lower bounds of the decision variables through multiple runs of the optimization process will be demonstrated in this paper. The framework proposed in this study offers great analytical capability to predict the condition of various assets in a water distribution network without having to undertake expensive inspection procedures.