Session: 34-06 Axial compressor design methods

Paper Number: 125897

125897 - Aerodynamic Shape Optimization of a Stator 

Abstract

In this paper, a numerical study is conducted on a compressor stator. The results are compared with existing literature and experimental data. For this purpose, a grid refinement study for the baseline geometry at the design condition with an unstructured mesh family is performed. The total pressure loss coefficient and the exit flow angle are compared against the experimental data. Moreover, the corner separation on the suction side is compared against the experimental data. Subsequently, a multi-objective optimization is performed on the baseline geometry with the aim of improving two objectives: flow deviation angle and total pressure loss coefficient at the stator exit. For the CAD-based optimization, a parametric model with 54 design variables is built. This parametric model is fitted to the baseline geometry to start the optimization process. The Flow360 compressible flow solver is employed for the steady RANS analysis, utilizing the Spalart-Allmaras turbulence model. Multi-objective genetic algorithms (MOGA) are employed for the optimization. Non-uniform profile boundary conditions, obtained from experimental tests, are applied at the inlet. The results for the baseline simulation at the design condition demonstrate strong agreement with both experimental and numerical findings. In the optimal Pareto front solutions, both objectives exhibit improvement. The optimal solution is compared with previously published results in the literature.

Introduction:
The aviation industry wants to reduce emissions, noise, and fuel consumption. For turbomachinery applications and gas turbine engineering, maintaining efficiency while meeting the new environmental standards requires innovative solutions. The unsteady flow at the outlet of the compressor, which is due to the pulse detonation engine (PDE) combustion tubes, impacts the last compressor stator. The last compressor stator is exposed to a variation of the incidence angle at the leading edge and unsteady flow. As a result, the overall aerodynamic performance is influenced. The corner vortices and corner separation between a blade’s suction side and the hub are known as the culprits for the highest losses. Different studies are performed to control the flow on the suction side. The presented work focuses on 3D global optimization of the stator's blade in order to reduce the losses due to corner separation based on CAD-based design parameters. This work presents a 3D optimization of a stator cascade with 3D separated flows using 54 design variables to cover a rich design space.


Gradient-free Optimization

The multi-objective genetic algorithm (MOGA) is applied to two objective functions: mass-averaged exit flow angle and mass-averaged total pressure loss coefficient at the outlet. Initially, 350 samples are generated with respect to 54 design variables using a quasi-random sequence generator. Next, in the first round of optimization, a response surface is constructed from the initial 350 sample solutions. Three design iterations, each producing 5 solutions based on Dakota’s multi-objective genetic algorithm (MOGA), are conducted.

In the second round of optimization, the results obtained in the first round are combined with the sample solutions to create a response surface. Then, three design iterations, each yielding 5 solutions per design iteration, are carried out. Finally, in the third round of optimization, the results obtained in the previous rounds are added to the samples to update the response surface. Python’s global multi-objective optimization with one design iteration is then employed to determine if the solutions in the Pareto front show no further improvement. The last round is only conducted to verify the optimal Pareto solutions found in the second round.

Presenting Author: Payam Dehpanah Flexcompute Inc

Presenting Author Biography: Aerospace Engineer at Flexcompute
Interests in aerodynamic shape optimization, design & automation

Authors:

Payam Dehpanah Flexcompute Inc
Cj Doolittle Flexcompute Inc
Carsten Fuetterer FRIENDSHIP SYSTEMS AG
Mattia Brenner FRIENDSHIP SYSTEMS AG