摘要
The purpose of this note is to point out a fact not commonly known, namely, that Johns and Lieberman in their interesting and useful paper [1] used the first extreme value (Gumbel) distribution in finding the bounds for Weibull reliability. If the underlying distribution is Gumbel, one can use their method (omitting one step-taking logarithms) and tables of [1] to find bounds for reliability. Johns and Lieberman consider X the two parameter Weibull variable with c.d.f., Fx(t; a, 13) = 1 exp {-(t/a)}, t > O, a > 0, (1) where a, iS are scale and shape parameters. They use the well known relationship [2], that the distribution of W = log X has the Gumbel c.d.f., Fw(u; v, a) = 1 exp {exp ([u v]/ao)}, 0 (2) where v = log a, a = p-1, are location and scale parameters, and u = log t, so that Y = W o = W log to has the translated Gumbel distribution, Fy(u; pt, v) = 1 exp {exp ([u ,t])/o-)}, (3) with parameters , = v Uo and a. The crux of the Johns-Lieberman procedure is to realize that Equations (1), (2) are different ways of writing P(X to) = 1 F(to ;a, ft) = 1 Fw(uo ;v, o-) = 1 Fy(0; /z, a) = exp { -exp (--/a}. A lower 100y% confidence limit, B(to) for R(to) satisfying P[R(to) > R(to)] = y,