Case Studies in Data-driven Methods Applied to Engineering Problems in Solid Mechanics and Heat Transfer
Year: 2023
Type: dissertation
Abstract: This thesis explores the efficacy of data-driven methods for solid mechanics and heat transfer problems, challenging traditional computational approaches. Three case studies are presented: firstly, a Data-driven Reduced-order Model for transient temperature fields is created using Dynamic Mode Decomposition. The reduced-order model is constructed solely from temperature data, without knowledge of boundary conditions or material properties. Secondly, a Visual Transformer Neural Network is employed for representative volume element homogenization. The model predicts effective elastic properties based on elastic properties and images of two-phase microstructures. Lastly, One-Dimensional Partial Differential Equations for heat diffusion and beam deflection problems are solved using physics-informed neural networks and deep operator networks. Despite data dependence for training, these models provide a computationally efficient alternative to existing numerical methods. This thesis advocates for the integration of these emerging data-driven methods into the engineer’s toolbox.
Author: Dankhara, Dhruvin
Institution: University of Guelph
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7212335998