摘要
The detrimental effects of surface roughness on the growth of turbulent boundary layers in turbines are well known. However, the robust prediction of these effects can be problematic, especially for surfaces which are not similar to sand grains (which form the basis of the standard predictive approach). Several publications have proposed additional parameters, such as effective slope and skew, to augment wall-normal measures such as Sa for correlating surface roughness with aerodynamic penalty. However, many of these parameters possess limitations that make them unsuitable for correlating topography to aerodynamic penalties for aero-engine designers. To address the above limitations, a geometric framework is developed in this thesis whereby geometrically grounded parameters which are robust to scan resolution can be reliably extracted from high-resolution (typically < 0.1 µm) 2-D areal rough surface scans. Two parameters are introduced as explicit measures of the wall-parallel length scales: feature (i.e. a roughness pit or peak) size, d70, and mean feature (pit-to-peak) separation, mean(Dl). In this thesis, Sa is adopted as a wall-normal measure due to its popularity within the gas turbine industry. The decomposition of Sa into its contributions above and below the mean surface roughness height (the above-below decomposition) is also demonstrated. This provides a geometrically grounded alternative to skew for quantifying the wall-normal distribution of roughness heights. Scans of turbine blade surfaces show that their mean feature separations can be up to six times larger than those of sand grain roughness with the same Sa. For the turbine surface (BladeA) that is examined in this thesis, both the feature size and mean feature separation were approximately 2.4 times greater than those of the sand grain surface with the same Sa. In addition, the above-below decomposition for BladeA is different from a sand grain surface, such that it exhibits a strongly negative skew whereas a sand grain surface has a strongly positive skew. An experimental framework, Scan-Scale-Print-Measure, where a surface scan is scaled up, 3D-printed, and tested in a flat-plate facility where the resulting turbulent boundary layer loss is measured, has been developed. This framework was validated geometrically and aerodynamically, showing that for a topography to be accurately 3D-printed (i.e. to generate 100% of its expected aerodynamic penalty), the 3D-printed surface needs to have a cross-correlation coefficient of 96% or better (i.e. nearer to 100%) and a Sa match of 102% or better of its intended surface. The resulting scaling criteria, as a function of printer resolution, for the 3D-printer used is established. This experimental framework enables systematic investigations of the introduced geometric roughness parameters and provides evidence of their aerodynamic relevance through correlation with measured aerodynamic penalty. Studies in which sand grains were either stretched or spaced apart demonstrate that the feature size, d70, and the mean feature separation, mean(Dl), are distinct topographical measures and that both are needed within robust correlations. This finding allows “effective slope” and “density parameter” approaches to be incorporated into the proposed single framework. Further systematic investigations show that the above-below decomposition can be used to augment Sa, providing a more reliable description of the distribution of heights relative to a sand grain surface. It is also shown that a single equivalent sand grain diameter cannot be assigned to a non-sand grain rough surface in the transitionally rough regime (where the total drag has non-negligible contributions from both the viscous and form drags). Finally, the experimental framework was applied to an engine-representative turbine surface to determine its measured aerodynamic penalty under a zero pressure gradient, and the geometric framework was used to predict it. The results show that the surface generates 48% of the aerodynamic penalty of a sand grain surface with the same Sa. By evaluating the topographical differences relative to a nominal sand grain equivalent, relevant parametric studies were identified and applied. For this engine-representative surface, the wall-parallel stretch and wall-normal above-below decomposition studies were combined to produce a prediction within the range 42% to 53% of the sand grain aerodynamic penalty. The geometric framework approach outperformed enhanced correlations from the literature by incorporating aerodynamically relevant parameters and trends with Reynolds number. This framework provides a foundation for extending such predictions to more realistic turbine flow low conditions, including non-zero pressure gradients.