Exploring dissimilarities in momentum and heat transfer over homogenous rough surfaces

Two neural networks are successfully deployed to predict the influence of homogenous and isotropic rough surfaces on the velocity and temperature profile. The networks, in combination with data from direct numerical simulations (DNS), are then used to investigate the roughness influence. Surfaces with similar values in the temperature roughness function but different values in the velocity counterpart and vice versa are identified and further analyzed to investigate these dissimilarities. Especially, the blanketing layer reveals significant differences for sparse surfaces, which propagates into the roughness functions $\Delta U^+$ and $\Delta \Theta^+$. To further understand which topological features quantify these functions, multiple statistical parameters are analyzed with focus on the ratio of sheltered regions to the total plane area as well as the total surface area.