Bayesian optimization and topographic exploration of drag-reducing dimples for aerodynamic surfaces

Abstract

Dimples offer a promising route to reducing drag on aerodynamic surfaces. However, whether such shallow concavities yield a net benefit depends sensitively on their topography, which demands systematic mapping and exploration of their comprehensive design space. In this study, dimple design is examined as a mixed-variable optimization over four design variables: dimple type, depth, in-plane scale, and streamwise stretch. The design space is explored using MixMOBO, a Bayesian optimizer, coupled with immersed-boundary large eddy simulations of channel flows at a constant flow rate corresponding to a flat channel at a friction Reynolds number of 180. The optimal solution, a relatively deep, fully packed, streamwise-elongated diamond dimple, attains a 13.2% drag reduction, notably above previously reported values. A Gaussian process metamodel sensitivity analysis identifies dimple topology as the dominant factor, with coverage and elongation acting mainly through interactions, and depth itself carrying no universal sign. A near-wall flow analysis links the leading designs to fully attached, groove-like flow, whereas poorer designs tend to produce local flow separation that incurs adverse form drag. From these findings, key design insights for drag-reducing dimples are provided.