Files in this item

FilesDescriptionFormat

application/pdf

application/pdfROSS-THESIS-2015.pdf (6MB)
(no description provided)PDF

Description

Title:Accelerating induction machine finite-element simulation with parallel processing
Author(s):Ross, Christine Anne Haines
Advisor(s):Krein, Philip T.
Department / Program:Electrical & Computer Engineering
Discipline:Electrical & Computer Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:M.S.
Genre:Thesis
Subject(s):finite element
simulation
finite
element
MATLAB
Graphics Processing Unit (GPU)
parallel
parallel
processing
linear
nonlinear
transient
eddy current
eddy
induction
Machine
induction machine
electrical machine
speedup
electromagnetic
Compute Unified Device Architecture (CUDA)
sparse matrix-vector multiplication
Sparse Matrix-vector Multiply (SpMV)
Krylov
iterative solver
Finite Element Method (FEM)
Finite Element Analysis (FEA)
Galerkin
Abstract:Finite element analysis used for detailed electromagnetic analysis and design of electric machines is computationally intensive. A means of accelerating two-dimensional transient finite element analysis, required for induction machine modeling, is explored using graphical processing units (GPUs) for parallel processing. The graphical processing units, widely used for image processing, can provide faster computation times than CPUs alone due to the thousands of small processors that comprise the GPUs. Computations that are suitable for parallel processing using GPUs are calculations that can be decomposed into subsections that are independent and can be computed in parallel and reassembled. The steps and components of the transient finite element simulation are analyzed to determine if using GPUs for calculations can speed up the simulation. The dominant steps of the finite element simulation are preconditioner formation, computation of the sparse iterative solution, and matrix-vector multiplication for magnetic flux density calculation. Due to the sparsity of the finite element problem, GPU-implementation of the sparse iterative solution did not result in faster computation times. The dominant speed-up achieved using the GPUs resulted from matrix-vector multiplication. Simulation results for a benchmark nonlinear magnetic material transient eddy current problem and linear magnetic material transient linear induction machine problem are presented. The finite element analysis program is implemented with MATLAB R2014a to compare sparse matrix format computations to readily available GPU matrix and vector formats and Compute Unified Device Architecture (CUDA) functions linked to MATLAB. Overall speed-up achieved for the simulations resulted in 1.2-3.5 times faster computation of the finite element solution using a hybrid CPU/GPU implementation over the CPU-only implementation. The variation in speed-up is dependent on the sparsity and number of unknowns of the problem.
Issue Date:2015-07-17
Type:Thesis
URI:http://hdl.handle.net/2142/88070
Rights Information:Copyright 2015 Christine Ross
Date Available in IDEALS:2015-09-29
Date Deposited:August 201


This item appears in the following Collection(s)

Item Statistics