Quantitative Evaluation of Neural Networks for NDE Applications Using the ROC Curve
Date
Authors
Major Professor
Advisor
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Authors
Research Projects
Organizational Units
Journal Issue
Is Version Of
Versions
Series
Begun in 1973, the Review of Progress in Quantitative Nondestructive Evaluation (QNDE) is the premier international NDE meeting designed to provide an interface between research and early engineering through the presentation of current ideas and results focused on facilitating a rapid transfer to engineering development.
This site provides free, public access to papers presented at the annual QNDE conference between 1983 and 1999, and abstracts for papers presented at the conference since 2001.
Department
Abstract
This paper arose from the need to quantitatively evaluate performance of a neural network example used to distinguish corrosion from noise. Although this is a simple GO/NO-GO situation, many NDE systems can be reduced to this level; for example, corrosion/no corrosion, crack/no crack, dangerous crack/ineffective crack. Even when the size of a crack is to be determined, the instrument or operator must first be able to detect it. Many NDE managers would probably prefer this GO/NO-GO decision since it can easily be verified, is likely to be more accurate, predictable and fast