CHARACTERISTICS OF A RELIABILITY DEMONSTRATION TEST

Eight characteristics are important in reliability demonstration testing. These are:

1. Specified reliability, R _s : This value is sometimes known as the "customer reliability." Traditionally, this value is represented as the probability of success (i.e., 0.98); however, other measures may be used, such as a specified MTBF.

2. Confidence level of the demonstration test: While customers desire a certain reliability, they want the demonstration test to prove the reliability at a given confidence level. A demonstration test with a 90% confidence level is said to " demonstrate with 90% confidence that the specified reliability requirement is achieved."

3. Consumer's risk, ² : Any demonstration test runs the risk of accepting bad product or rejecting good product. From the consumer's point of view, the risk is greatest if bad product is accepted. Therefore, the consumer wants to minimize that risk. The consumer's risk is the risk that a test can accept a product that actually fails to meet the reliability requirement. Consumer's risk can be expressed as: ² = 1 - confidence level

4. Probability distribution: This is the distribution that is used for the number of failures or for time to failure. These are generally expressed as normal, exponential, or Weibull.

5. Sampling scheme

6. Number of test failures to allow

7. Producer's risk, ± : From the producer's standpoint, the risk is greatest if the test rejects good product. Producer's risk is the risk that the test will reject a product that actually meets the reliability requirement.

8. Design reliability, R _a : This is the reliability that is required in order to meet the producer's risk, ± , requirement at the particular sample size chosen for the test. Small test sample sizes will require a high design reliability in order to meet the producer's risk objective. As the sample size increases , the design reliability requirement will become smaller in order to meet the producer's risk objective.

THE OPERATING CHARACTERISTIC CURVE

The relationship between the probability of acceptance and the population reliability can be shown with an operating characteristic (OC) curve. An operating characteristic curve can also be used to show the relationship between the probability of acceptance and MTBF or failure rate. Given then an OC curve, one may calculate the:

1. Producer's risk, ±

2. Consumer's risk, ²

3. Probability of acceptance at any other population reliability or MTBF or failure rate

Obviously, a specific OC curve will apply for each test situation and will depend on the number of pieces tested and the number of failures allowed.

ATTRIBUTES TESTS

If the components being tested are merely being classified as acceptable or unacceptable, the demonstration test is an attributes test. Attributes tests:

• May be performed even if a probability distribution of the time to failure is not known

• May be performed if a probability distribution such as normal, exponential, or Weibull is assumed by dichotomizing the life distribution into acceptable and unacceptable time to failure

• Are usually simpler and cheaper to perform than variables tests

• Usually require larger sample sizes to achieve the same confidence or risks as variables tests

VARIABLES TESTS

Variables tests are used when more information is required than whether the unit passed or failed, for example, "What was the time to failure?" The test is a variables test if the life of the items under test is:

• Recorded in time units

• Assumed to have a specific probability distribution such as normal, exponential, or Weibull

FIXED-SAMPLE TESTS

When the required reliability and the test confidence/risk are known, statistical theory will dictate the precise number of items that must be tested if a fixed sample size is desired.

SEQUENTIAL TESTS

A sequential test may be used when the units are tested one at a time and the conclusion to accept or reject is reached after an indeterminate number of observations. In a sequential test:

1. The "average" number of samples required to reach a conclusion will usually be lower than in a fixed-sample test. This is especially true if the population reliability is very good or very poor.

2. The required sample size is unknown at the beginning of the test and can be substantially larger than that in the fixed-sample test in certain cases.

3. The test time required is much longer because samples are tested one at a time (in series) rather than all at the same time (in parallel), as in fixed-sample tests.

Now that you are familiar with the four test types, let us look at the test methods . Note that the four test types are not mutually exclusive. We can have fixed-sample or sequential-attributes tests as well as fixed-sample or sequential-variables tests.