Abstract

This study explores aerodynamic shape optimization of Low Pressure Turbine (LPT) blades using Implicit Large Eddy Simulation (ILES) coupled with the Mesh Adaptive Direct Search (MADS) optimization algorithm. Aerodynamic shape optimization in the aerospace industry relies heavily on the adjoint method combined with steady-state Computational Fluid Dynamics (CFD) solvers, namely the Reynolds-Averaged Navier-Stokes (RANS) approach. While this methodology has proven quite useful and efficient for common optimization problem, its effectiveness is reduced when faced with separated turbulent flow. As LPT designs move increasingly towards high lift configurations to reduce engine weight, accurately predicting transition and turbulent separation for each optimization cycles becomes a necessity. The T106D blade with an exit Mach number of 0.4 and Reynolds number 80,000 is used as the initial conditions before optimization, because of the presence of turbulent transition and separation on the Suction Surface (SS). First, the ILES framework employing the Flux Reconstruction (FR) approach is presented along with the benefits of its implementation on modern many-core hardware, specifically Graphical Processing Units (GPUs). The formulation of the MADS algorithm is then presented and compared against gradient-based methods and other gradient-free methods. A Bézier curve is used to modify the original blade camber line, with the use of 4 control points, permitting a wide range of shapes to be obtained. Then, the results of the baseline case are presented with a comparison with experimental results to validate the methodology selected. The results of optimization cycles using two different objective functions are then presented and compared with the original blade. Results demonstrate that a total pressure loss coefficient reduction of about 16% was achieved in the first optimization, while a tangential force increase of more than 29% was achieved with the second optimization. Finally, recommendations are made as to how this methodology could be successfully applied in industrial applications, and what future research should focus on.