Efficient deep learning-based models for physics simulation and computational design

Gladstone, Rini Jasmine

Permalink

https://hdl.handle.net/2142/129368 Copy

Description

Title

Efficient deep learning-based models for physics simulation and computational design

Author(s)

Gladstone, Rini Jasmine

Issue Date

2025-02-20

Director of Research (if dissertation) or Advisor (if thesis)

Meidani, Hadi

Doctoral Committee Chair(s)

Meidani, Hadi

Committee Member(s)

• Ravaioli, Umberto
• Sychterz, Ann
• Zhang, Shelly

Department of Study

Civil & Environmental Eng

Discipline

Civil Engineering

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Scientific computing
• Deep learning
• Machine learning
• Graph neural networks, Physics informed neural networks
• Topology optimization
• Solid mechanics

Abstract

Scientific computing problems are widely used in an array of problems with applications ranging from micro- to macro-world in the field of physics, biology, material science, and civil and mechanical engineering. Many scientific computing problems such as design optimization, and design space exploration require resource-intensive repetitive simulation runs of a model with different input values. For systems characterized by numerous input parameters, the response calculation is particularly challenging as one also has to deal with the curse of dimensionality, which is the exponential increase in the volume of the input space, as the number of parameters increases linearly. Many real-world computational problems, require running a large number of high fidelity simulations of physical systems, governed by partial differential equations, using numerical methods such as finite element, which are often limited by the available computational resources. While data driven and physics-informed deep learning-based surrogate models have demonstrated success in tackling many of these problems, they still grapple with limitations in generalizability across problems, ability to handle complex domains and dependence on high data quality (for supervised models). The overarching objective of this dissertation is to take a step toward addressing these computational challenges and contribute to the promotion of efficient computational approaches based on deep learning for scientific computing problems such as physics simulations and computational design. In particular, and in moving toward this objective, I introduce various deep learning approaches, both data-driven and physics-informed, for fast and efficient modeling of physical systems. In the first part of the dissertation, I tackle supervised deep learning algorithms, primarily Variational Autoencoders (VAEs) and Graph Neural Networks (GNNs) and their applications for various array of problems such as Robust Topology Optimization (RTO), time-independent physics simulations and multi-fidelity methods. In particular, I introduce multi-fidelity architectures for VAE and GNN for improving the computational efficiency of the models. I also propose novel GNN architectures for accurate evaluation of time-independent physical systems. In the second part of the dissertation, I introduce two variants of physics-informed neural networks (PINNs), namely FO-PINN and PINN-FEM, to tackle some of the challenges of PINNs, including strong imposition of boundary conditions, solving higher-order PDEs and parameterized systems. A variety of engineering applications are considered to verify the accuracy and efficiency of the proposed methods.

Graduation Semester

2025-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129368

Copyright and License Information

Copyright 2025 Rini Jasmine Gladstone

Owning Collections