Multi-Fidelity Design Optimization Studies for Supersonic Jets Using Surrogate Management Frame Method

Abstract

Design of the realistic supersonic jets has been challenging problem. The design space for a ground boom is known to be non-smooth and discontinuous. Due to those particular characteristics as such, gradient-based optimizer can not be employed to locate the global minimum, and gradient-free search method is more suitable for an extensive exploration in design sites. In this study we employ one of the direct search method, pattern search method which has been used for a general constrained optimization on filter methods. A general pattern search (GPS) method is the search algorithm which composes of the search step and the local poll step. Random global search is performed in the search step. In the poll step, the search directions are selected in a way that they can accelerate the convergence of search. A filter algorithm is combined with the PS to handle the mission constraints in a aggregate way. Moreover, to overcome the restriction to a finite set of directions in the GPS for local exploration, mesh adaptive direct search (MADS) method is introduced providing more search directions. A virtual mesh on the design space is constructed and as the algorithm iterates, the mesh can be refined or coarsened corresponding to the success of the search. This method is applied to the shape optimization of the low boom supersonic jet and the pattern search proceeds through the several iterations, until all the mission constrains are satisfied and aerodynamic/boom performance is improved. The objective function is the minimization of MTOW and the low ground boom is treated as one of the constraints.