서브메뉴
검색
Uncertainty Quantification in High-Dimensional Supersonic Flows Using Nonlinear Reduced Order Modeling
Uncertainty Quantification in High-Dimensional Supersonic Flows Using Nonlinear Reduced Order Modeling
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105514
- ISBN
- 9798263336813
- DDC
- 330
- 저자명
- Iyengar, Nikhil.
- 서명/저자
- Uncertainty Quantification in High-Dimensional Supersonic Flows Using Nonlinear Reduced Order Modeling
- 발행사항
- [Sl] : Georgia Institute of Technology, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 285 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Mavris, Dimitri N.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2024.
- 초록/해제
- 요약Across the world, there is a growing interest in re-introducing commercial supersonic transport aircraft (SST). However, the development of SSTs is dependent on the mitigation of sonic boom loudness to acceptable levels. Moreover, uncertainties in the atmosphere and flight conditions can drastically impact the loudness of an aircraft and must be accounted for during the design process to avoid certification delays. Given this need, there has been significant research on modeling the aerodynamic field around SSTs, which is characterized by the presence of strong shocks and nonlinearities, to accurately shape the aircraft sonic boom signature and guide downstream sub-system analyses.Computational fluid dynamics (CFD) simulations closely match experimental aerodynamic data, but each simulation can take hours or days to output a solution. For multi-query problems, such as uncertainty quantification (UQ), which require repeated evaluation of the expensive simulation, this cost becomes computationally intractable. Indeed, this requirement for physics-based models conflicts with the high computational cost and leads to a gap that must be tackled to make uncertainty propagation feasible. This thesis addresses the challenge of performing UQ in such high-dimensional, uncertain fields.While there exist several methods to reduce this cost, surrogate models hold promise because they are non-intrusive, data-driven, and cheap to evaluate. Traditionally, UQ has relied on data-fit surrogates to predict scalar random variables. However, when applied to high-dimensional fields, where there are thousands or millions of coupled random variables, it is both impractical and computationally expensive to train individual data-fit models. Alternatively, Reduced Order Models (ROM) have served as promising candidates for providing parametric predictions of high-dimensional fields at a reduced computational cost. The thesis first explores a recently developed ROM-based tool for high-dimensional UQ called POD-PCE. This method relies on Proper Orthogonal Decomposition (POD) for dimensionality reduction and a generalized Polynomial Chaos Expansion (PCE) for uncertainty propagation. While its application had been previously restricted to canonical test cases or linear problems, this study thoroughly investigates the robustness of the methodology in empirical data sets characterized by sharp discontinuities, large nonlinearities, and high dimensionality. Specifically, this thesis explores the performance of the PODPCE method in predicting aerodynamic fields with shocks. From these studies, it is seen that this tool is unable to predict high-speed flows with nonlinearities as POD and PCE both rely on approximating a nonlinear space with a linear model.Thus, to address this deficiency, the research is decomposed into two technical gaps: The first is to identify a suitable alternative to linear dimensionality reduction and the second is to extend PCE models to handle nonlinear probability spaces. To address the first gap, manifold learning algorithms are identified as a way to potentially capture discontinuous flow features more effectively than their linear counterparts. Using a series of experiments from one-dimensional uncertain internal flow through ducts to two-dimensional flow over an airfoil with several uncertainties, the performance of the manifold learning methods is assessed. It is observed that manifold learning approaches can better predict shocks and maintain accuracy throughout the field as well. It is recommended that global approaches to manifold learning, which attempt to preserve the overall geometry, be used as they consistently capture both local and global features.
- 일반주제명
- Aircraft
- 일반주제명
- Mean square errors
- 일반주제명
- Sample size
- 일반주제명
- Monte Carlo simulation
- 일반주제명
- Aerodynamics
- 일반주제명
- Neural networks
- 일반주제명
- Decomposition
- 일반주제명
- Pressure distribution
- 일반주제명
- Multidimensional scaling
- 일반주제명
- Visualization
- 일반주제명
- Geometry
- 일반주제명
- Aerospace engineering
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2024 us c eng d■001000017360370
■00520260202105514
■006m o d
■007cr#unu||||||||
■020 ▼a9798263336813
■035 ▼a(MiAaPQ)AAI32309268
■035 ▼a(MiAaPQ)GeorgiaTech75195
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a330
■1001 ▼aIyengar, Nikhil.
■24510▼aUncertainty Quantification in High-Dimensional Supersonic Flows Using Nonlinear Reduced Order Modeling
■260 ▼a[Sl]▼bGeorgia Institute of Technology▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a285 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Mavris, Dimitri N.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2024.
■520 ▼aAcross the world, there is a growing interest in re-introducing commercial supersonic transport aircraft (SST). However, the development of SSTs is dependent on the mitigation of sonic boom loudness to acceptable levels. Moreover, uncertainties in the atmosphere and flight conditions can drastically impact the loudness of an aircraft and must be accounted for during the design process to avoid certification delays. Given this need, there has been significant research on modeling the aerodynamic field around SSTs, which is characterized by the presence of strong shocks and nonlinearities, to accurately shape the aircraft sonic boom signature and guide downstream sub-system analyses.Computational fluid dynamics (CFD) simulations closely match experimental aerodynamic data, but each simulation can take hours or days to output a solution. For multi-query problems, such as uncertainty quantification (UQ), which require repeated evaluation of the expensive simulation, this cost becomes computationally intractable. Indeed, this requirement for physics-based models conflicts with the high computational cost and leads to a gap that must be tackled to make uncertainty propagation feasible. This thesis addresses the challenge of performing UQ in such high-dimensional, uncertain fields.While there exist several methods to reduce this cost, surrogate models hold promise because they are non-intrusive, data-driven, and cheap to evaluate. Traditionally, UQ has relied on data-fit surrogates to predict scalar random variables. However, when applied to high-dimensional fields, where there are thousands or millions of coupled random variables, it is both impractical and computationally expensive to train individual data-fit models. Alternatively, Reduced Order Models (ROM) have served as promising candidates for providing parametric predictions of high-dimensional fields at a reduced computational cost. The thesis first explores a recently developed ROM-based tool for high-dimensional UQ called POD-PCE. This method relies on Proper Orthogonal Decomposition (POD) for dimensionality reduction and a generalized Polynomial Chaos Expansion (PCE) for uncertainty propagation. While its application had been previously restricted to canonical test cases or linear problems, this study thoroughly investigates the robustness of the methodology in empirical data sets characterized by sharp discontinuities, large nonlinearities, and high dimensionality. Specifically, this thesis explores the performance of the PODPCE method in predicting aerodynamic fields with shocks. From these studies, it is seen that this tool is unable to predict high-speed flows with nonlinearities as POD and PCE both rely on approximating a nonlinear space with a linear model.Thus, to address this deficiency, the research is decomposed into two technical gaps: The first is to identify a suitable alternative to linear dimensionality reduction and the second is to extend PCE models to handle nonlinear probability spaces. To address the first gap, manifold learning algorithms are identified as a way to potentially capture discontinuous flow features more effectively than their linear counterparts. Using a series of experiments from one-dimensional uncertain internal flow through ducts to two-dimensional flow over an airfoil with several uncertainties, the performance of the manifold learning methods is assessed. It is observed that manifold learning approaches can better predict shocks and maintain accuracy throughout the field as well. It is recommended that global approaches to manifold learning, which attempt to preserve the overall geometry, be used as they consistently capture both local and global features.
■590 ▼aSchool code: 0078.
■650 4▼aAircraft
■650 4▼aMean square errors
■650 4▼aSample size
■650 4▼aMonte Carlo simulation
■650 4▼aPartial differential equations
■650 4▼aAerodynamics
■650 4▼aNeural networks
■650 4▼aDecomposition
■650 4▼aPressure distribution
■650 4▼aMultidimensional scaling
■650 4▼aVisualization
■650 4▼aGeometry
■650 4▼aAerospace engineering
■690 ▼a0538
■690 ▼a0800
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360370▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


