Robust design is a multi-objective optimization framework for obtaining designs that perform favorably under uncertainty. In this paper robust design is used to redesign a highly loaded, transonic rotor blade with a desensitized tip clearance. The tip gap is initially assumed to be uncertain from 0.5 to 0.85% span, and characterized by a beta distribution. This uncertainty is then fed to a multi-objective optimizer and iterated upon. For each iteration of the optimizer, 3D-RANS computations for two different tip gaps are carried out. Once the simulations are complete, stochastic collocation is used to generate mean and variance in efficiency values, which form the two optimization objectives.
Two such robust design studies are carried out: one using 3D blade engineering design parameters (axial sweep, tangential lean, re-cambering and skew) and the other utilizing suction and pressure side surface perturbations (with bumps). A design is selected from each Pareto front. These designs are robust: they exhibit a greater mean efficiency and lower variance in efficiency compared to the datum blade. Both robust designs were also observed to have significantly higher aft and reduced fore tip loading. This resulted in a weaker clearance vortex, wall jet and double leakage flow, all of which lead to reduced mixed-out losses. Interestingly, the robust designs did not show an increase in total pressure at the tip. It is believed that this is due to a trade-off between fore-loading the tip and obtaining a favorable total pressure rise and higher mixed-out losses, or aft-loading the tip, obtaining a lower pressure rise and lower mixed-out losses.
On the BIC for determining the number of control points in B-spline surface approximation in case of correlated observations
