On the Existence of Logarithmic Terms in the Drag Coefficient and Nusselt Number of a Single Sphere at High Reynolds Numbers

Abstract

At the beginning of the second half of the twentieth century, Proudman and Pearson (J. Fluid. Mech.,2(3), 1956, pp.237-262) suggested that the functional form of the drag coefficient ($C_D$) of a single sphere subjected to uniform fluid flow consists of a series of logarithmic and power terms of the Reynolds number ($Re$).\ In this paper, we will explore the validity of the above statement for Reynolds numbers up to $10^6$ by using a symbolic regression machine learning method.\ The algorithm is trained by available experimental data and data from well-known correlations from the literature for $Re$ ranging from $0.1$ to $2\times 10^5$.\ Our results show that the functional form of the $C_D$ contains powers of $\log (Re)$, plus the Stokes term, fulfilling partially the statement made above. The logarithmic $C_D$ expressions can generalize (extrapolate) beyond the training data and are the first in the literature to predict with acceptable accuracy the rapid decrease (drag crisis) of the $C_D$ at high $Re$.\ We also find a connection between the root of the $Re$-dependent terms in the $C_D$ expression and the first point of laminar separation.\ We did the same analysis for the problem of heat transfer under forced convection around a sphere and found that the logarithmic terms of $Re$ and Peclect number $Pe$ play an essential role in the variation of the Nusselt number $Nu$.\ The machine learning algorithm independently found the asymptotic solution of Acrivos and Goddard (J. Fluid. Mech., 23(2),1965, pp.273-291).

arXiv comment51 pages, 8 Figures

References in corpus (2)

• TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
• Understanding deep learning requires rethinking generalization