Downloaded from https://www.everyspec.com AMCP 706.198 If all items are failed, the probability state- FHI r/n (2-1a) ments about the true Cdf (relative to the sample points) are simple and straightforward. FL.o (r 1)/n (2-1b) This method is recommended whenever it can be used. Connect the points with horizontal and verti- cal lines; this is the sample Cdf 2. ?he distribtiton of the order statistics is used to provide 3 plotting positions for 2. 2-sided s-confidence bounds on the 4 each point. This spread provides a feel for actual Cdf:
choose the x-confidence level, the uncertainties Involved, Simple, straight- near it (l/n)l is reasonable; then find KS, forward probability statements can be made from the body of Table 2-1(A) (e,g., n a 10, only for an individual point, not for the Cdf s-confidence w 95%, KSM M 0.41). The upper as a whole. Extensive tables are necessary, bound is plotted at As mentioned in the first paragraph, these rLo 4 KSl and FHj + KSR (2-2ra) methoda can be used provided no failure time is longer than a censoring time, A con- the lower bound is plotted at soring (censored item) occurs when an item is removed from test before it fails.
ThL PHIt - KS,, and FLo - KS.1 (2-2b) cause for removal cannot be related to the apparent condition of the item if an analysis For each bound, connect the points with in Chapter 2 is to be valid. For example, suppose as failure becomes quite likely, an item begins to vibrate slightly. Then If items TABLE 2.1(A) that vibrate are removed from test before they actually fail, the removal cause is related TABLE OF K.8 BOUNDS to the condition of the item, and a legitimate analysis of tebt results is virtually impossible n Kil, eo.oon codewee) unless the whole test and population desLrip- (90%) (9%) (9%) (90%) tion are redefined. 5 .51 .56 .03 .07 2-2,1.1 Proatkcal Plotting (K-S Bounds) .
/ ,52 le8 .62 8 .41 .45 .51 .54 Specific Instructions for this kind of 10 .37 .41 .46 .4g plotting are on the back of each sheet of 12 ,34 ,38 .42 .45 Practical Probability Paper in par. 2-2. They 14 .31 .35 .39 .42 are repeated here for a general case. 16 .30 .33 .37 .39 18 .28 .31 .35 .37 Notation: 20 .26 .29 .33 .38 30 .22 .24 .27 .29 F a Cumulative distribution function (Cdl) 40 119 .21 .24 .25 na - ample size r - failure number; r - 1,2... . n Plotting instructions follow: (formula Is o~k. for n;;e) q1 i. Plotting data: plot failure r at the For censored samples, use an n which is between the two points originaI-nLumber-on-test and the number-of-fallures. (V Y 2,64