Differential Reynolds-stress models (DRSM) provide an advantage over two-equation models when studying flows with strong acceleration and curvature. Unlike two-equation turbulence models, DRSM features a description of the redistribution of energy between components of the Reynolds-stress tensor, and this is known as the pressure-strain correlation ϕij . Developing a description of ϕij in a near-wall region without a wall-echo term, however, has been an ongoing challenge. The elliptic blending model of Manceau and Hanjali´c (2002) exploits an elliptic blending function α, to blended the ‘homogeneous’ ϕh ij of Speziale et al. (1991) in the freestream (ϕSSG ij ), with a boundary condition ϕw ij . The derivation of ϕh ij does not account for near-wall effects, and the functions scaling the finite tensor polynomial of ϕh ij significantly change in the near-wall region. This behaviour is evident in the logarithmic region of the boundary layer where the elliptic blending effects are most prominent. The present work develops a machine-learning framework for reassessing the coefficients of the pressure-strain correlation ϕSSG ij , which considers near-wall blending effects of the elliptic blending DRSM. This methodology uses gene-expression programming to re-evaluate ϕij , while respecting the boundary conditions of ϕw ij at the wall, and the generalised polynomial tensor decomposition of the ϕh ij model in the free stream. This framework allows for ϕij development in problems that have incomplete Reynolds-stress budget datasets, and this is advantageous for the study of complex flow configurations that do not obey the attached boundary layer theory of Cole’s Law of the Wall. The present methodology has yielded good improvements in the elliptic blending model on the periodic hills benchmark.     «

Differential Reynolds-stress models (DRSM) provide an advantage over two-equation models when studying flows with strong acceleration and curvature. Unlike two-equation turbulence models, DRSM features a description of the redistribution of energy between components of the Reynolds-stress tensor, and this is known as the pressure-strain correlation ϕij . Developing a description of ϕij in a near-wall region without a wall-echo term, however, has been an ongoing challenge. The elliptic b...     »