Adaptive computing for optimizing high-fidelity simulation runtimes

Physics-based and high-fidelity simulations are often leveraged to predict real-world trends and optimize resource consumption. However, these simulations are often computationally expensive and time consuming. Alternatively, small scale simulations can be performed at a fraction of the cost, but generate a host of scalability issues when translated to large-scale applications. To address model scalability issues, it is necessary to identify cost efficient methods for running physics-based models.Here we demonstrate how adaptive computing can be leveraged to create surrogate models that accurately approximate high-fidelity simulation behavior at a reduced runtime. We introduce a pipeline for training surrogate models that reaffirms the effectiveness of low-fidelity simulations. We elaborate on this pipeline for multi-scale simulations which demonstrate how adaptive computing can be used to stagger high-fidelity queries during surrogate training.These implementations demonstrate how adaptive computing can be used to manipulate model run-time and increase accuracy with reduced computational budgets.These implementations can be leveraged in any existing modeling pipeline regardless of discipline.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Domantay, Janelle

Contributors dc:contributor

• Driggs-Campbell, Katherine

• Energy Modeling
• Adaptive Computing
• Simulation
• Sustainability
• Optimization

Statement dc:rights

• Copyright 2024 Janelle Domantay

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/127235

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F127235