Physics-Infused Reduced-Order Modeling of Aerothermal Loads for Fluid-Thermal-Structural Interactions
Open Access
- Author:
- Vargas Venegas, Carlos
- Graduate Program:
- Aerospace Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- March 27, 2022
- Committee Members:
- Daning Huang, Thesis Advisor/Co-Advisor
Puneet Singla, Committee Member
Amy Pritchett, Program Head/Chair
Sven Schmitz, Committee Member - Keywords:
- hypersonic
fluid-thermal-structural
aerothermoelasticity
data-driven
physics-driven
physics-based
reduced-order modeling
machine learning
field inversion
aerothermal
turbulent viscous-inviscid interactions
neural ordinary differential equations
direct collocation
optimization
kriging
proper-orthogonal decomposition
latin hypercube sampling
thermoelastic
hypersonic vehicle
RANS
skin-panel
two-dimensional
gaussian process regression
computational fluid dynamics
supported panel
augmentation
augmented
model augmentation
surrogate
POD-kriging
POD
algebraic
differential
algebraic augmentation
differential augmentation
NODE
PIROM
analysis - Abstract:
- Although considerable advancements have been made during the past several decades in hypersonic vehicle technologies, there are still numerous unresolved experimental and computational challenges that limit practical and affordable hypersonic flight. One of the main challenges is associated to the accurate and robust computational modeling of the flow surrounding the hypersonic vehicle. The prediction of aerothermal loads over a hypersonic structure during sustained atmospheric flight is critical for the design, analysis and optimization of structures for this class of vehicles. Due to the high-energy physics encountered at hypersonic speeds, the computational modeling and coupling of structural dynamics, heat transfer, and hypersonic aerothermodynamics, viz. aerothermoelasticity (ATE), becomes a challenging phenomenon to resolve computationally. Reduced-order models (ROMs) and surrogates are typical approaches to reducing the computational cost of hypersonic aerothermoelasticity simulations to a tractable level, while maintaining a desirable level of accuracy. While ROMs tend to accelerate predictions by several orders of magnitude, they still suffer from limitations that prohibit their practical use in real-world applications. In this thesis, a new physics-infused reduced-order modeling (PIROM) methodology is presented which leverages from field inversion and machine learning (FIML) techniques to improve first-principle-based low-fidelity physical models. In a nutshell, the PIROM performs an inference step between high-fidelity data and a low-fidelity model output to extract spatio-temporal discrepancies that can be learnt by a machine learning model in functional form. These functionals are later embedded into the low-fidelity model to inform it of the missing physics. Eventually, this process creates an augmented model, capable of high-fidelity predictions comparable to full-order models while maintaining a low computational cost. Specifically, the PIROM methodology is adopted to generate an improved ROM capable of accelerated and accurate hypersonic aerothermal load prediction for arbitrary thermal and elastic responses from a skin panel structure. On the practical side, the PIROM methodology is used to predict pressure and heat flux loads in a fully-coupled hypersonic aerothermoelastic simulation for a two-dimensional skin panel configuration; a typical structure in hypersonic vehicles. Numerous structural boundary conditions are considered to assess the generalization capabilities of the PIROM. These boundary conditions include clamped, spring, simply-supported, and combinations of these. Specifically, the results of this thesis demonstrate how the PIROM approach can considerably enhance the predictive capabilities of the low-fidelity turbulent-viscous inviscid (TVI) interaction model for aerothermal load prediction, and obtain an efficient and robust augmented TVI (ATVI) model, which is practical in high-fidelity analysis and design of hypersonic structures.