Downloaded from https://www.everyspec.com I~Ii AMCP 70-.19t tested against the hypothesis that "the rell- s-Confidence interval estimates are mote use- abilities are not equal". These tests provide ful than point estimates because they give a useful guidance to equipment designers who much better idea of the uncertainty involved can then make decisions based on test results in the estimation process. and rigorous statistical mnalyses. Distribution estimates are used when it is The tests must be well planned and test desired to estimate the probability distrlbu- data properhy evaluated in order to avoid tion governing a particular reliability measure. costly enron and delays.
This is especially This is usually a 2-step process: (1) the form true for system reliability testing in which of the distribution must be hypothesized or components frequently are destroyed and in determined from the failure data, and (2) the which expensive equipment must be built to parameters that describe the distribution simulate the operational environment. Test must be estimated. planning ih very important in complex pro- grams that operatu under time and budget Nonparametric methods can be used to limitations.
Critical trade-offs must be made estimate reliability measures without making among test time, number of units tested, and any assumptions concerning the time-to- achieved s-confidence level. failure distribution. Generally, nonparametric estimates are not as efficient as parametric Reliability measurement tests are used to estimates. Nonparametric reliability estimates make estimates of the reliability of a popula- apply only to the specific test interval and "tion of items. Both pmrametric and nonpara- cannot be extrapolated. Both point estimates metric estimates can be used.
Parametric and interval estimates can be made using non- estimates are baed on a known or assumed parametric techniques. distribution of the characteristic of interest. The constants in the equation that describe Before one decides to choose a type of the probability distribution are called para- distribution and estimate Its parameters from meters. Nonparametric estimates are made the data, one should have a very good idea without assuming any particular form for of why it is being done. For example, one the probability distribution.
may wish to use it to interpolate among the data, one may wish to extrapolate, or one The three type, of parametric estimates may wish to r itimate certain charaAteristics most frequently used are (Ref. 3): of the data such as a mean, median, or tenth percentile. If one is going to use the distribu- , I. Point estimate--a singli-valued estimate tion to interpolate among the data, goodnes- of a reliability paanmeter of-fit tests are quite appropriate to help de- termine a good type of distribution to use. 2.
IntervYJ estimate-an estimate of an if one is going to extrapolate, then goodness- interval that is believed to contain the true of-fit tests (which operate only in the region value of the parameter of the data) are not appropriate because they do not tell how well the distribution will fit 3. Distribution estimate-an estimate of in the region where there are no data; in the parametets of a reliability distribution, fact, goodness-of-fit tests for this purpose can " ~be extremely misleading.
A s-confidence interval estimate is one for b which there is a known probability that the If the purpose of knowing the distribution true value of the unknown parameter or Is to estimate soneu characteristics of the characteristic lies within a computed interval, population, one should give serious considers- j~ . .......... j