Estimating Colebrook-White Friction Factor Using Tree-Based Machine Learning Models

Colebrook-White friction factor is the common hydraulic roughness coefficient used in water supply systems. Analysis, design, and management of water distribution networks basically rely on an adequate estimation of friction factor. In turbulent flows, it depends on the Reynolds number and the relative roughness of pipes. This study assesses the performances of four tree-based Machine Learning (ML) models, including Decision Tree (DT) Regression, Adaboost, Gradient Boosting Regressor (GBR), and XGBoost. Their performances were compared with the equation of Swamee-Jain, which is utilized in EPANET hydraulic solver. For the comparative analysis, a reliable database comprising more than one million data points in the turbulent flow zone. Based on the results, GBR and DT performed better than the equation of Swamee-Jain. The improvement made in friction factor estimations using ML-based suggests further studies on the topic. Thus, future studies can be conducted by implementing ML-based estimator in hydraulic software for simulating pipe networks.