Electrical and Computer Engineering ETDs

Publication Date

9-12-2014

Abstract

Computer simulation techniques are widely adopted nowadays in many areas like manufacturing, engineering, graphics, animation, virtual reality and so on. However, the standard finite element based simulation is notorious for its expensive computation. To address this challenge, I present a GPU-based parallel implementation for simulating large elastic deformation. Classic modal analysis provides a set of orthonormal bases vectors, which span a spectral space encoding the dynamics of the elastic body. As each basis vector is orthogonal to each other, the computation is completely decoupled and can be well-fit into the modern GPGPU platform. We further explore the latest feature of NVIDIA CUDA so that the result of GPU computation can be directly used for upcoming rendering/visualization and a significant amount of overheads for transmitting data from client GPU and host CPU via the PCI-Express bus are avoided. Real-time simulation is made possible with this technique for many cases that otherwise is not possible.

Keywords

gpgpu, gpu, parallel, computing, deformed, simulation, object, model, cuda, parallelization

Sponsors

Consejo Nacional de Ciencia y Tecnologia (Mexico)

Document Type

Thesis

Language

English

Degree Name

Computer Engineering

Level of Degree

Masters

Department Name

Electrical and Computer Engineering

First Advisor

Yang, Yin

First Committee Member (Chair)

Shu, Wei Wennie

Second Committee Member

Pattichis, Marios S.

Share

COinS