Hydraulic Model Calibration of a Laboratory Water Distribution Network Using Hydraulic and Water Quality Measurements

In the realm of hydraulic modeling for water distribution networks, the calibration process plays a pivotal role. Calibration involves precise adjustments to align a model with observed data. However, when the measured data is scarce, the calibration process becomes challenging. In such cases, laboratory models prove valuable for simulating real-world conditions. Variability in parameters like pipe dimensions, length, roughness coefficients, and nodal demand as well as nodal elevations often leads to disparities between computer-based model outcomes and reality. Despite extensive research on computer-based models, laboratory water distribution network models and their calibration have received relatively limited attention due to the challenges and costs associated with them. In this study, a laboratory model of a water distribution network was constructed and subjected to hydraulic calibration. Roughness coefficients and minor head losses within the network were determined using a meta-heuristic method. Then, pipe roughness coefficients for polyethylene pipes were assessed and compared with values from scientific references. In the following, hydraulic validation of the network was conducted using the water quality simulation of a conservative substance. This approach illustrates the level of concurrence of flow ratios in the network pipes between the model and reality.
The laboratory network was made of PE40 and consists of a square looped system with 1-meter pipe lengths, employing a 1000-liter tank to maintain a constant water head. This research was conducted in three stages. In the first stage, network calibration was performed using piezometric pressure and output flow data. Roughness coefficient and pipe minor head loss coefficients were selected as decision variables. The objective function was defined to minimize the total weighted difference in piezometric head and discharge between the model and reality. In the second stage, validation was performed based on pressure and output flow data. In the third stage, the network's hydraulic validation with respect to pipe flow rates was performed through the modeling of a conservative substance. This is because the dissolution of a conservative substance occurs solely due to mixing at nodes and flow division, allowing it to represent the correspondence of flow ratios in network pipes between the model and reality. In this research, pressure data was recorded using piezometers, and salt concentration was calculated using TDS (Total Dissolved Solids) sensor.
After performing the optimization, a value of 0.008 was obtained for the Darcy-Weisbach friction coefficient (ɛ), This value aligns well with the assumption of new pipes in the network, in agreement with previous research (e.g., 0.050 by The Plastics Pipe Institute, 2008, and 0.070 by Padilla et al., 2013). Also, values of 1.20 and 0.89 were obtained for the minor loss coefficients of 0.5 and 1-meter pipes, respectively. Furthermore, the optimized minor loss values effectively reflect differences attributed to the number of connections in the 0.5 and 1-meter pipes, falling within recommended scientific ranges. Notably, unlike previous field studies, this research uniquely focuses on a laboratory model.
After hydraulic calibration and validation using pressure and output flow data, further validation was conducted using the water quality model. The saltwater solution was injected at a specific point in the network, and the TDS quality parameter was measured at the two points in the network. Subsequently, utilizing the TDS values and the established relationship between TDS and salt concentration calculated in the laboratory, the salt concentration at the location of two sensors was determined. It's worth mentioning that the network's water supply contained dissolved solids. Subsequently, initial and injection concentrations were applied to the model and, the simulation was performed. A comparison of salt concentration results at two sensor locations revealed an 8.5% error in the first experiment and 2.5% in the second, confirming excellent agreement between the laboratory and computer-based network hydraulic model.