Estimation of Pr{X>Y} or Pr{X/spl les/Y}

The paper presents a procedure for the estimation (point and interval) for a reliability model of the form Pr{X>Y} or Pr{X/spl les/Y} and Gaussian (normal) distributed variables. The procedures are derived from a quantitative approach that allows the user to get sufficient evidence, concerning the product reliability, by testing a small sample but at severe conditions, rather than testing a large sample at the nominal working conditions. Y has known parameters while X is a Gaussian r.v. with an unknown mean and a known coefficient-of-variation. The procedure is practical and cost effective when facing situations such as high test-cost, or severe contractual reliability requirements.<
>