Large-eddy Simulations of the Flow over Airfoils with Leading-Edge Roughness

Abstract

We performed large-eddy simulations of the flow over a NACA4412 airfoil to understand the effects of leading-edge roughness designed to mimic ice accretion. The roughness elements protrude outside the boundary layer, which, near the leading edge, is very thin; thus, the configuration does not represent a classical rough-wall boundary layer, but rather the flow over macroscopic obstacles. A grid convergence study is conducted and results are validated by comparison to numerical and experimental studies in the literature. The main effect of the obstacles is to accelerate transition to turbulence. Significant variations in structure generation are observed for different roughness shapes. The three-dimensionality of the irregularities has a strong impact on the flow: it creates alternating regions of high-speed (“peaks”) and low-speed (“valleys”) regions, a phenomenon termed “channelling”. The valley regions resemble a decelerating boundary layer: they exhibit considerable wake and higher levels of Reynolds stresses. The peak regions, on the other hand, are more similar to an accelerating one. Implications of the channelling phenomenon on turbulence modelling are discussed. In the second part of the thesis, we investigate further the channelling phenomenon by performing simulations at three angles of attack. Downstream of the roughness zone, the fast regions slow down under the action of hairpin vortices generated by the roughness elements, and merge with neighbouring slow regions. The flow channels remain coherent over the boundary layer developing on the suction side of the airfoil and affect the trailing-edge separation. With increasing angle of attack in the linear- lift regime the intensity of flow-channelling can increase or decrease depending on the topology of the leading-edge roughness. Its effect on the trailing-edge separation remains, however, significant. The mean-separation line is highly distorted, and the separation length can vary by up to 30% of the chord length along the span.