Downloaded from https://www.everyspec.com AMO 70W.IN ability of results, The larger the experimental tests, "Identical" systems must be tested uncertainty, the less precise the results. The (Ruf. I). also of the experimental uncertainty can be decreased through replication of the experi- The principal requirement of any sampling msnt. Replication is the term used to lndi- procedure is that It yield representative cate the number of parts with which a samples, A representative sample is a min- speciflc test condition is associated.
The lature of the population, To make inferences total sample size is the sum of all the replica- about the population from the sample re- tions, For example, if a test is conducted at suits, the sample selection must be random. A levels or temperature, I levels of vibration, The simplest kind of random sample results and C levels of acceleration, there are a total when every item int thq population has an of AB'C combinations, If each or these equal chanue of being chosen for the sample. combinations has R replications, tho total A simple random sample does not necessarily number of parts to be sampled is (AB'C)R. result in a representative sample.
Stratifled Increasing the number of' replications will in- sampling can be used to obtain a sample that mrea the precision of the test, I.e,, reduce Is representative of the universe of Items to the experimental uncertainty, be tested, However, within each stratum, the l•nal selection procedure is random. A valid test must provide a good estimate S of the experimental uncertainty. Randomize- In system demonstration testing, the uni. tion provides the test engineer with a tech- pie selection Is limited to a few systems. nique that ensures good estimates of un. Often, a random sampling procedure cannot certainties.
To support the results of an be followed because the testing is limited to experiment with probability statements, the first few preproduction models, Then it randomization is necessary in the sample is ir•portant to distinguish between the spe- selection and in the experimental design, cial attontion preproduction models are given (from purchased parts to final Inspection) vs Standard deviations and э'confldence state- the kind of actual production system in the mente are two popular, good ways of meas- future, For example, how will the quality uring uncertainty, of Incoming steels bh checked to be sure they meet the specifications, month after 34.4 SAMPLE SELECTION month?
What will long term variations in all kinds of' properties do to estimates of The elimination of possible tources ol bias standard deviations which were based on a is an important consideration in the choice few samples? of sample items, Beioru a sample selection scheme is choar.c, the population from which 3.1 TYPES OF TESTS the sample is to be drawn must be precisely and correctly deflated. If resistors are th. 'rests often are classified as "attributes" or padts under consideration, the test engineer "variables".
must ask If the population i1 to contain all resistors, or Just carbon composition, all An attributes test is one in which each , values of r.eshtance; 5%, 10%, or 20% tuler- Item Is tested, and judgud to be a success or ance limits; and so on. The population must a failure. Attributes tests uэually are used be precisely and correctly detined, since con. for testing a [,ehot Item In which time or clusions of the expertwoeutt ats limited to cycles are not involved, It is possible, how- I that populatiun from which the actual sample ever, to include time by testing each item ,, , . is twuly a random on,', fi demonstration for a specified period und counting the num- ( j ', '!. ,• ' •• , ,