Improved Submerged Inlet Conceptual Design Process Using Data Mining and Surrogate Modeling

Submerged inlets have the potential to increase aircraft performance, but are difficult to design and integrate within an aircraft fuselage. This difficulty comes from the inherent complexity of submerged inlet flowfields. The strong vortices, thick boundary layers, and turbulent behavior of the submerged inlet requires that Navier-Stokes simulation or wind tunnel testing be used to provide design analysis. For the designer who wishes to optimize submerged inlet geometries at the low to medium fidelity “conceptual design” level, few viable optimization strategies are available. RANS CFD-based shape optimization is useful for producing localized optima for detailed design but remains too computationally expensive for large scale conceptual design studies. The execution of a coarse set of wind tunnel experiments at the conceptual level is prohibitively expensive and time consuming. Published wind tunnel studies provide limited sensitivity information, tend to be univariate in nature, and provide data within a limited range of applicability. The ability to interpolate a large mixed database of separate wind tunnel or computational experiments would provide an enhanced multivariate simulation capability for the inlet designer. This paper presents a methodology for the efficient data mining of a large non-homogenous, mixed database of submerged inlet experiments for the purposes of conceptual design. The methodology presented uses the techniques of design of experiments, surrogate modeling, and genetic algorithm optimization to efficiently determine an optimized inlet design. Steps of the proposed methodology are described in detail. Finally, a practical example problem is shown using NACA submerged inlet wind tunnel experiments as a database and ANSYS Fluent® with Sculptor® by Optimal Solutions for final solution verification.