Properties of parameter estimation techniques for a beta-binomial failure model. Final technical report

Technical Report · Tue Dec 01 04:00:00 EST 1981

OSTI ID:5536466

Shultis, J K; Buranapan, W; Eckhoff, N D

Of considerable importance in the safety analysis of nuclear power plants are methods to estimate the probability of failure-on-demand, p, of a plant component that normally is inactive and that may fail when activated or stressed. Properties of five methods for estimating from failure-on-demand data the parameters of the beta prior distribution in a compound beta-binomial probability model are examined. Simulated failure data generated from a known beta-binomial marginal distribution are used to estimate values of the beta parameters by (1) matching moments of the prior distribution to those of the data, (2) the maximum likelihood method based on the prior distribution, (3) a weighted marginal matching moments method, (4) an unweighted marginal matching moments method, and (5) the maximum likelihood method based on the marginal distribution. For small sample sizes (N = or < 10) with data typical of low failure probability components, it was found that the simple prior matching moments method is often superior (e.g. smallest bias and mean squared error) while for larger sample sizes the marginal maximum likelihood estimators appear to be best.

• Research Organization: Kansas State Univ., Manhattan (USA). Dept. of Nuclear Engineering
• OSTI ID: 5536466
• Report Number(s): NUREG/CR-2372
• Country of Publication: United States
• Language: English

Related Subjects

22 GENERAL STUDIES OF NUCLEAR REACTORS
220900* -- Nuclear Reactor Technology-- Reactor Safety
COMPUTERIZED SIMULATION
FAILURES
MATHEMATICAL MODELS
MAXIMUM-LIKELIHOOD FIT
NUMERICAL SOLUTION
PROBABILITY
REACTOR COMPONENTS
RELIABILITY
SIMULATION