Artificial intelligence (AI)-based friction factor models for large piping networks

In large piping networks, evaluation of friction factor is a time-consuming process and poses computational complexity. This is because the friction factor has to be evaluated for every pipe segment and that too by using implicit correlations. In the present study, this issue has been addressed by developing artificial intelligence (AI)-based friction factor models namely, support vector regression (SVR), artificial neural networks (ANN) and gene expression programing (GEP) to predict the friction factor for the turbulent flow regime. The developed models have been compared with the existing correlations based on the statistical parameters and have shown excellent prediction accuracy with the lowest average absolute relative error (AARE), root mean square error (RMSE) and highest correlation coefficient (R) as 1.43%, 0.0003, 0.9993 for SVR while for ANN they are 2.11%, 0.00095, 0.9978 and for GEP they are 7.14%, 0.0024, 0.9864, respectively. Leave-one-out cross-validation on the test set for SVR, ANN, and GEP are obtained as 0.9976, 0.9957, and 0.9726, respectively. Furthermore, the performance of these AI-based models, i.e. SVR, ANN, and GEP models and the various well-known correlations have been studied for estimating pipe friction factor in both smooth and rough pipes with different values of relative roughness. The SVR-based model significantly outperforms the existing correlations and the GEP-based model and marginally the ANN-based model. AI approach reduces the computational complexity and the time-consuming iterative solution of implicit correlations for large pipe networks without compromising the accuracy.