3.5.2 presented methods of determining which tasks to be sampled under the
fault inducement approach, which are applicable to each of the test methods
presented herein. When the demonstration is a requirement of the development
program, the procuring activity historically has had the option of
surveillance over and/or participation in the random selection of tasks
comprising the demonstration population (see Column 11 of Table B-V) down to
and including the specific faults to be simulated or induced. It is
recommended that this practice continue. Further details on this and other
management aspects of maintainability demonstration are contained in Section
of this appendix. In all cases, whenever a chosen task can result in events detrimental to safety of personnel or property, appropriate redesign action must take place. In the event that secondary failures result from an induced fault, they should be documented and their impact on item maintainability assessed. A report of such findings is typically made to the procuring activity or demonstration authority.
Two basic types of tests may be used for statistical
maintainability demonstration: sequential and non-sequential. In sequential
testing, testing continues until a decision to accept or reject the hypothesis
under consideration (see Section
B.3.1 of this Appendix for discussion of hypothesis testing) can be made. One drawback of sequential testing is that the length of the test cannot be determined in advance. However, sequential testing will accept very low MTTRs or reject very high MTTRs very quickly. A non-sequential, or fixed sample size is best when the maintainability must be demonstrated with a given confidence level.
Whenever sequential test plans are employed (see Test Method 1, for example), care must be exercised in selecting and sampling tasks to insure that a true simple random sample is obtained. Departures from simple random sampling, such as proportionate stratified sampling, can affect the validity of the test procedures presented herein. However, this effect is considered minimal for the sample sizes required by the test procedures that are not sequential tests. In short, simple random sampling must be used for sequential test methods.