A Reduced Order Modeling Methodology for the Multidisciplinary Design Analysis of Hypersonic Aerial Systems
Author(s)
Decker, Kenneth H.
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Recent years have seen a significant increase in the demand for an advance and diverse fleet of hypersonic aerial systems. As computational power has increased, high-fidelity physics-based numerical analyses have emerged as feasible alternatives to physical experimentation, especially during early design phases. Due to the complexity of the underlying physics that govern hypersonic aerodynamics, these numerical tools can be very costly and not practical for systems engineering tasks that require many queries. To overcome these challenges, Reduced Order Models (ROMs) have been implemented that are capable of replacing expensive numerical analyses with inexpensive field surrogate models that can accurately predict aerodynamic flow features. This dissertation puts forth a methodology for the development of accurate, efficient, data-driven ROMs capable of predicting complex off-body hypersonic flow features. This methodology uses both linear and nonlinear Dimensionality Reduction (DR) to reduce high-dimensional aerodynamic field data into low-dimensional representations that faithfully represent the original data set. After this reduction, state-of-the-art surrogate modeling techniques are used to map parametric design inputs into this low-dimensional space to enable predictions. Manifold Alignment (MA), has also been implemented to enable models to leverage data from multiple fidelity sources. The performance of this method is evaluated experimentally using a number of test problems that exhibit a range of size and feature complexity. It is observed in many of these experiments that nonlinear ROMs outperform equivalent linear ROMs when the underlying fields exhibit complex discontinuous behavior. Furthermore, nonlinear ROMs consistently reduce data to lower dimensional forms than equivalent linear models, which results in nonlinear ROMs having lower evaluation costs and being more resilient to the “curse of dimensionality” then their linear counterparts. Similar trends are observed with multi-fidelity ROMs. When implemented into a coupled analysis, ROMs trained using the proposed methodology are able to achieve superior performance to state-of-the-at scalar models when predicting integrated force coefficients. Moreover, the proposed ROMs offer the novel capability of providing parametric flow-field data within a coupled analysis, which enables more sophisticated assessments of system-level performance, objectives, and constraints.
Sponsor
Date
2021-07-26
Extent
Resource Type
Text
Resource Subtype
Dissertation