ABSTRACT

Historically, sampling plans have been used to assess the present quality of the material examined. They are employed to determine the acceptability of the product against specifications at a given time. This has usually been the time of sale. Of course, the implication is that items presently acceptable will retain their utilitarian properties upon reaching the consumer. An important quality characteristic of some products, however, is degradation in use, that is, the useful life of the product with regard to some property. The advent of considerations of reliability imposed by high technology programs, such as space

and atomic power, consumerism, and conformance testing to government mandatory standards, have placed a new dimension on the sampling problem, that of time. Reliability sampling plans are used to determine the acceptability of the product at some future point in its effective life. This usually involves some form of life testing. An Advisory Group on Reliability of Electronic Equipment (AGREE) was formed in 1952 under

the assistant secretary of defense to ‘‘monitor and stimulate interest in reliability matters and recommend measures which would result in more reliable electronic equipment.’’ AGREE (1957) defined reliability as follows: ‘‘Reliability is the probability of performing without failure a specified function under given conditions for a specified period of time.’’ Reliability testing is to provide assurance of reliability. In this sense, it is not testing what the product is, but rather how it will operate, over time, in the hands of the consumer. The standard plans discussed so far determine whether the product is made to specifications. Reliability plans assess how it will perform. The time dimension implicit in reliability testing is superimposed on the sampling problem as an

additional criterion. Samples must be tested for a specified length of time. When all units are tested to failure, the standard plans can be utilized to assess the results against specified requirements. If lifetimes are measured, these results can be used in a variables sampling plan, such as MIL-STD414 or its derivatives, provided the distributional assumption of the plan is satisfied. Also, the number failing before a required time can be used with standard attributes plans in determining the disposition of the material, e.g., MIL-STD-105E. In reliability and safety testing, extremely low levels of probability of acceptance are often used.

When a test based on a two-point plan (p1, p2, a, b) has been passed, it is often said that a reliability of at least p¼ 1 p2 has been demonstrated with g¼ 1b confidence. Specifications are often written in this way. Clearly, a variety of plans could satisfy such a requirement on what amounts to limiting quality (LQ). For example, it follows from the Schilling-Johnson Appendix Table T5.2 that to demonstrate .9995 confidence of .99 reliability, the plan n¼ 1000, c¼ 1 could be used since, for c¼ 1, when Pa¼ .0005, np¼ 10.000, and