Downloaded from https://www.everyspec.com AMCP 70819 especially If the system is very large. For reliability level considered to be example, it would be almost impossible to unacceptable. perform a reliability demonstration test on a complete ballistic missile system, including Incorrect inferences are of two types: missile, launcher, command and control, etc. In this case, techniques which permit system 1. Type I Error: Ho may be rejected reliability (and the uncertainty therein) to when it is true. (Producer risk) be estimated from subsystem reliability data can be used. - Probpblity of Type I error, producer risk. 3.2 CONCEPTS 2. Type II Error:
Ho may be accepted This paragraph treats the statistical con- when H, is true. (Consumer risk) cepts for tests used to accept or reject an equipment or system, based on its degree of 0 o Probability of Type II error, compliance with iecifled reliability goals. consumer risk. 3.2.1 TERMINOLOGY The probabilities of making these errors de- pend on the sample size and decision criteria. The statistical basis of a decision test Table 3-1 summarizes the relationships. (Accept/Reject) Is the theory of testing hy- potheses. In rellallity testing, the null It is easy to get boxed in by statistical hypothesis under test is:
"the submitted terminology and assertions, For example, a lot or system conforms to the reliability valid conclusion of some statistical tests (not requirement". An alternate hypothesis is mentioned so far) is that there wore not also specified (or at least implied): "the enough data to detect a difference-If it submitted lot or system does not conform exists. An engineer can always call for more to the reliability requirement in some way". tests. It may mess up some previously cal- Rejecting the null hypothesis is often con- culated s-confidence oi s-significance leveis, sidered statistically equivalent to accepting nrd it may make new ones virtually Impossi- the alternate hypothesis (Refs.
I and 5), but ble to calculate, but those levels are not the other engineering considerations often apply. be-all and end-all of testing. The prime put- pose of testing is to be sure only good equip- When components are qualification tested, samples-rather than whole lots-are used. Therefore, the possibility of incorrect infer- ences due to sampling fluctuations has to be TABLE 3.1 considered. RELATIONSHIPS BETWEEN TEST DECIIION The following notation is used: AND TRUE 8ITUATION STmTrue Situaton He .
the null hypothesis that the product conforms to the reliability requirements (e.g., the mean life Accept Correct Decision Type II Error is equal to a specified value) NO Probability -w - a Probability H, U the alternate hypothesis that the Accept Type I Error Correct DecIsion reliability of the lot is at some HN Probability- a Probability- 1 -0