Estimation of Lower Bound Properties from Material Test Data
Abstract
Monte Carlo simulations were performed to determine how the accuracy of lower bound values estimated from experimental data is influenced by sampled size required confidence level, and assumed statistical mode. Population distributions having different degrees of skewness, selected to bracket those expected in actual experimental data, were studied. For nearly every case considered, lower bound estimates calculated using Log-Normal statistics were more accurate than estimates calculated using either Normal or Weibull statistics. It was demonstrated that testing more than three samples per condition can greatly reduce the error associated with the lower bound estimate. However, after the thirteenth sample no additional sample will reduce the lower bound estimation error by more than 2.5% for all statistical distribution/ confidence level combinations considered. When applied to material properties for which the population distribution has been established by previous testing, it was demonstrated that a Monte Carlo simulation can be used to assess the maximum expected lower bound estimation error as a function of sample size and confidence level. Keywords: Lower bond; Statistical evaluation; Monte Carlo; Material characterization. (MJM)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1989
- Accession Number
- ADA207378
Entities
People
- M. T. Kirk