Chihiro Fujio, Hideaki Ogawa
DOI Number XXX-YYY-ZZZ
Conference Number HiSST-2022-437
Design of scramjet intakes requires sophisticated methodologies to achieve desirable compression reducing total pressure loss and drag simultaneously, and multi-objective design optimization (MDO) using
evolutionary algorithms (EA) is one of the most promising approaches. In addition, the dataset obtained
from computational fluid dynamics (CFD) simulations in the MDO process can serve expected as a rich
mine of physics-based information. However, the substantial computational cost of a large number of
CFD evaluations represents an obstacle to high-fidelity design search. The present study proposes an
evolutionary algorithm which employs a multi-dimensional flow prediction model via deep learning for
objective/constraint function evaluations to replace CFD simulation by predicting flowfields inside scramjet intakes. The proposed EA-based MDO framework is applied to an optimization study of axisymmetric
3-ramp intakes and the performance and utility are discussed. This EA approach using multi-dimensional
predictive modeling is applicable for various optimization problems which require reduction of the computational cost of function evaluations.