Data-Driven Techniques for Materials Characterization and Intelligent Experimental Design at Advanced Scattering Facilities
Data-Driven Techniques for Materials Characterization and Intelligent Experimental Design at Advanced Scattering Facilities
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211153054
- ISBN
- 9798346385042
- DDC
- 620.11
- 서명/저자
- Data-Driven Techniques for Materials Characterization and Intelligent Experimental Design at Advanced Scattering Facilities
- 발행사항
- [Sl] : Stanford University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 211 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-05, Section: A.
- 주기사항
- Advisor: Dunne, Mike;Dionne, Jennifer.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2024.
- 초록/해제
- 요약This dissertation addresses the need to accelerate materials discovery, particularly within the context of advanced scattering facilities. The complexity of materials discovery arises from vast design space sizes, the need to achieve multiple simultaneous property requirements, and the limitations of ab initiotheory leading to the necessity of performing time-consuming and expensive synthesis and characterization experiments. Advanced X-ray and neutron scattering facilities often play a key role in this materials design process, but now face challenges such as exploding data rates due to improvements in sources, detectors, and increased experimental complexity. This work focuses on developing methodology to enable closed-loop materials discovery at advanced scattering facilities. Specifically, data-driven techniques are introduced for the task of real-time analysis of scattering experiments and to guide materials experiments towards specific objectives.Real-time analysis, often a challenging bottleneck for closed-loop systems, is addressed through the design of methods for solving inverse scattering problems using supervised statistical learning. Emphasis is placed on using simulated data derived from physics-based forward models to aid in scattering analysis, as well as on the rational architectural design of supervised learning models to incorporate inductive biases that align with experimental data. Key frameworks introduced in this dissertation include automatic unit-cell estimation from powder X-ray di↵raction (PXRD) patterns, real-time photon assignment and contrast extraction for X-ray Photon Fluctuation Spectroscopy (XPFS), and robust Hamiltonian parameter inference from inelastic neutron scattering (INS) data.Furthermore, the dissertation presents novel methods for experimental materials design, adapting the recently proposed Bayesian algorithm execution (BAX) approach to frame materials discovery as the task of locating relevant regions of a design space aligned with specific experimental goals. This methodology o↵ers an automated way to convert complex experimental objectives into intelligent, goalaligned data acquisition strategies, allowing for addressing tasks that are impossible or impractical under current methods. The SwitchBAX data acquisition strategy, formulated in this dissertation, achieves state-of-the-art performance on datasets for TiO2nanoparticle synthesis and magnetic materials characterization, and should generally be applicable for any compound discovery task involving complex synthesis protocols, long data acquisition times, or limited sample quantities.Ultimately, this research helps enable closed-loop discovery at national facilities, o↵ering ecient solutions for accelerating materials research and discovery.
- 일반주제명
- Materials science
- 일반주제명
- Spectrum analysis
- 일반주제명
- Neutrons
- 일반주제명
- Fourier transforms
- 일반주제명
- Neural networks
- 일반주제명
- Pandemics
- 일반주제명
- Mathematics
- 일반주제명
- Point defects
- 일반주제명
- Large language models
- 일반주제명
- X-rays
- 일반주제명
- Parameter estimation
- 일반주제명
- Analytical chemistry
- 일반주제명
- Atomic physics
- 일반주제명
- Design
- 일반주제명
- Epidemiology
- 일반주제명
- Optics
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 86-05A.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017164844
■00520250211153054
■006m o d
■007cr#unu||||||||
■020 ▼a9798346385042
■035 ▼a(MiAaPQ)AAI31643373
■035 ▼a(MiAaPQ)Stanfordsg415nc8104
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620.11
■1001 ▼aChitturi, Sathya Ranjan.
■24510▼aData-Driven Techniques for Materials Characterization and Intelligent Experimental Design at Advanced Scattering Facilities
■260 ▼a[Sl]▼bStanford University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a211 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-05, Section: A.
■500 ▼aAdvisor: Dunne, Mike;Dionne, Jennifer.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2024.
■520 ▼aThis dissertation addresses the need to accelerate materials discovery, particularly within the context of advanced scattering facilities. The complexity of materials discovery arises from vast design space sizes, the need to achieve multiple simultaneous property requirements, and the limitations of ab initiotheory leading to the necessity of performing time-consuming and expensive synthesis and characterization experiments. Advanced X-ray and neutron scattering facilities often play a key role in this materials design process, but now face challenges such as exploding data rates due to improvements in sources, detectors, and increased experimental complexity. This work focuses on developing methodology to enable closed-loop materials discovery at advanced scattering facilities. Specifically, data-driven techniques are introduced for the task of real-time analysis of scattering experiments and to guide materials experiments towards specific objectives.Real-time analysis, often a challenging bottleneck for closed-loop systems, is addressed through the design of methods for solving inverse scattering problems using supervised statistical learning. Emphasis is placed on using simulated data derived from physics-based forward models to aid in scattering analysis, as well as on the rational architectural design of supervised learning models to incorporate inductive biases that align with experimental data. Key frameworks introduced in this dissertation include automatic unit-cell estimation from powder X-ray di↵raction (PXRD) patterns, real-time photon assignment and contrast extraction for X-ray Photon Fluctuation Spectroscopy (XPFS), and robust Hamiltonian parameter inference from inelastic neutron scattering (INS) data.Furthermore, the dissertation presents novel methods for experimental materials design, adapting the recently proposed Bayesian algorithm execution (BAX) approach to frame materials discovery as the task of locating relevant regions of a design space aligned with specific experimental goals. This methodology o↵ers an automated way to convert complex experimental objectives into intelligent, goalaligned data acquisition strategies, allowing for addressing tasks that are impossible or impractical under current methods. The SwitchBAX data acquisition strategy, formulated in this dissertation, achieves state-of-the-art performance on datasets for TiO2nanoparticle synthesis and magnetic materials characterization, and should generally be applicable for any compound discovery task involving complex synthesis protocols, long data acquisition times, or limited sample quantities.Ultimately, this research helps enable closed-loop discovery at national facilities, o↵ering ecient solutions for accelerating materials research and discovery.
■590 ▼aSchool code: 0212.
■650 4▼aMaterials science
■650 4▼aSpectrum analysis
■650 4▼aNeutrons
■650 4▼aFourier transforms
■650 4▼aNeural networks
■650 4▼aPandemics
■650 4▼aMathematics
■650 4▼aPoint defects
■650 4▼aLarge language models
■650 4▼aX-rays
■650 4▼aParameter estimation
■650 4▼aAnalytical chemistry
■650 4▼aAtomic physics
■650 4▼aDesign
■650 4▼aEpidemiology
■650 4▼aOptics
■690 ▼a0794
■690 ▼a0405
■690 ▼a0486
■690 ▼a0800
■690 ▼a0748
■690 ▼a0389
■690 ▼a0766
■690 ▼a0752
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g86-05A.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17164844▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


