Use Of Genetic Expression Programming For Inferring Roughness Correlations From a DNS Database

Understanding and predicting the impact of surface roughness on fluid flow plays a crucial role in many fluid-related fields. Direct Numerical Simulations (DNS) offer detailed insights, but their high computational cost still makes them impractical for realworld applications with roughness. In this study, an extensive collection of DNS data over rough surfaces is utilized to establish a correlation between equivalent sand-grain roughness ks and roughness parameters. A
genetic approach, called Gene Expression Programming (GEP), is employed to determine these correlations by posing the problem as a symbolic regression
problem. The resulting fully-rough model formula is compared with some empirical formulas and a neural network approach. GEP proved highly effective
at predicting the equivalent sand-grain roughness, outperforming the other methods. Additionally, a correlation for the full roughness range is sought. Finally, the developed ks correlation is tested on a linear turbine cascade to determine whether the RANS predictions improve with use of the new correlation, or not