Author:
Knott Christine E.,Kabban Christine Schubert
Abstract
Probability of detection (POD) studies for evaluating the capabilities of an inspection system for Air Force aircraft structural components commonly use a Logistic Regression model with a Wald 95% confidence interval. However, hit/miss POD data is distributed as a Binomial, and the sample sizes are commonly too small for Wald’s identically and independently normality distributed assumption to be true. This paper uses a large set of simulated representative hit/miss data to compare and contrast the performance of the four confidence intervals methods: Standard Wald, Modified Wald, Profile Likelihood Ratio, and Profile Modified Likelihood Ratio. Performance is measured in terms of bias and existence of a90/95 with respect to data distribution, sample size, overlap, and evenness. This paper provides guidance and methodology on new POD methods that more reliably and accurately estimate a90/95.
Publisher
The American Society for Nondestructive Testing, Inc.
Subject
Mechanical Engineering,Mechanics of Materials,General Materials Science
Reference18 articles.
1. Agresti, A. 2002. Categorical Data Analysis. 2nd ed. Hoboken, New Jersey: Wiley. https://doi:10.1002/0471249688.
2. Annis, Charles P.E. 2014. “Influence of sample characteristics on probability of detection curves.” AIP Conference Proceedings 1581, 2039-2046. https://doi.org/10.1063/1.4865074
3. Annis, C., L. Gandossi. 2012. ENIQ TGR Technical Document - Influence of Sample Size and Other Factors on Hit/Miss Probability of Detection Curves (ENIQ report N. 47). EUR 25200 EN. Luxembourg (Luxembourg): Publications Office of the European Union. JRC68677
4. Annis, C., J. C. Aldrin, and H. A. Sabbagh. 2015. “Profile likelihood: What to do when maximum probability of detection never gets to one.” Materials Evaluation, 73 (1): 96–99.
5. Annis, C., L. Gandossi, and O. Martin. 2013. “Optimal sample size for probability of detection curves.” Nuclear Engineering and Design, 262 (September 2013): 98-105. JRC74951
Cited by
2 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献