Abstract:
This paper introduces ABC2, an advanced framework for rotor blade design optimization that can effectively consider the airfoil shape variations during optimization process. A major component of this framework is an reduced-order model (ROM) that leverages deep-neural-network techniques both for airfoil parameterization and performance prediction. Utilizing the UIUC airfoil database and a two-dimensional unsteady Reynolds-averaged Navier-Stokes (URANS) solver, the ROM can effectively control the airfoil shapes and predict the resulting aerodynamic performance across the wide range of flow conditions. A comprehensive aerodynamic solver is incorporated for blade design optimization. Enhancement of the fidelity of the comprehensive solver is achieved through the integration of a three-dimensional URANS solver, which also plays a crucial role in analyzing the aerodynamics of the optimized blade and uncovering its underlying physics. The competence of the present framework is demonstrated by a hovering optimization problem, yielding satisfactorily optimized blades and identifying the underlying physics.