Track 2 Session 10
9:10 to 10:10 a.m. Thursday June 11, 2009

Making Conservative Estimates of Demonstrable Reliability When Model Parameters Are Unknown

When using a parametric model for the calculation of demonstrable reliability, the estimates on model parameters can come from historical or published data, but more often they are simply assumed, e.g. β = 1. When taking into consideration the calculations for the lower confidence bound on the demonstrable reliability, the assumed parameters can have a significant effect on the final estimate. Under the assumption of using a Weibull model to predict reliability against a specific failure mode, an estimate on the shape parameter can be made that truly minimizes the demonstrable reliability, i.e. the lower bound estimate. This estimate accounts for the target confidence level and a given testing strategy. This estimate, however, is strongly influenced by the number of tests and the duration of each individual test. Therefore, a sensitivity analysis will be made to highlight the dependencies between the test plan, model parameters and the demonstrable reliability.

Key Words: Weibull Analysis, Statistical Model Parameter Estimation, Statistical Model Sensitivity Analysis, Reliability Estimation