Degree Type

Dissertation

Date of Award

1997

Degree Name

Doctor of Philosophy

Department

Electrical and Computer Engineering

First Advisor

John P. Basart

Abstract

This dissertation presents a comprehensive study on the forward modeling methods, signal processing techniques, and image restoration techniques for two-dimensional eddy current nondestructive evaluation. The basic physical forward method adopted in this study is the volume integral method. We have applied this model to the eddy current modeling problem for half space geometry and thin plate geometry. To reduce the computational complexity of the volume integral method, we have developed a wavelet expansion method which utilizes the multiresolution compression capability of the wavelet basis to greatly reduce the amount of computation with small loss in accuracy. To further improve the speed of forward modeling, we have developed a fast eddy current model based on a radial basis function neural network. This dissertation also contains investigations on signal processing techniques to enhance flaw signals in two-dimensional eddy current inspection data. The processing procedures developed in this study include a set of preprocessing techniques, a background removal technique based on principal component analysis, and grayscale morphological operations to detect flaw signals. Another important part of the dissertation concerns image restoration techniques which can remove the blurring in impedance change images due to the diffusive nature of the eddy current testing. We have developed two approximate linear image restoration methods--the Wiener filtering method and the maximum entropy method. Both linear restoration methods are based on an approximate linear forward model formulated by using the Born approximation. To improve the quality of restoration, we have also developed nonlinear image restoration methods based on simulated annealing and a genetic algorithm. Those nonlinear methods are based on the neural network forward model which is more accurate than the approximate linear forward model.

DOI

https://doi.org/10.31274/rtd-180813-13537

Publisher

Digital Repository @ Iowa State University, http://lib.dr.iastate.edu/

Copyright Owner

Bing Wang

Language

en

Proquest ID

AAI9737777

File Format

application/pdf

File Size

271 pages

Share

COinS