Frequently Asked Questions About Quality

Please contact Media Sciences if your questions are not answered on these pages.

How do AQL, LTPD, and SPC sampling plans differ?

Sampling is often used for quality evaluation of large lots. Statistical sampling methods that minimize subjective elements are best. There are two types of statistical sampling methods; attribute and variable.

Attribute plans test the sample, rank each part as GOOD or BAD, and decide whether to accept or reject the lot based upon the number of BAD units. Attribute plans are simple and easy to execute, but do not detect marginal results and are not predictive, ignoring the history of past lots.

Attribute sampling is usually based on Poisson statistics, where sample size and acceptance number (maximum number of BAD units in the sample) are specified. This generates an Operating Characteristic (OC) that is a plot of the probability of lot acceptance (or confidence level) vs. percent defective in the total lot from which the sample was randomly selected.

Acceptable Quality Level (AQL) sampling fixes the probability of lot acceptance at 95%, automatically giving the AQL percent defective from the OC plot. For example, AQL=2% means that, on average, 95 of 100 lots containing 2% defective parts will be accepted, while only 5 of the 100 will be rejected. AQL sampling was implemented shortly after World War I.

AQL sampling results in a high probability that a bad lot will be accepted, being favorable to manufacturers but not to end users. Consequently, Lot Tolerance Percent Defective (LTPD) sampling was introduced later for high reliability military and aerospace applications, and is found in MIL-S-19500 (discrete semiconductors) and MIL-M-38510 (integrated circuits).

LTPD sampling uses the same OC, but fixes the probability of lot acceptance at 10%. For example, LTPD=5% means that, on average, 10 of 100 lots containing 5% defective will be accepted while 90 of the 100 will be rejected. Since this is more favorable to end users, it is recommended for all attribute programs.

Variables methods select and evaluate small samples, perhaps only 3 or 5 parts, on a regular predetermined schedule. Using Statistical Process Control (SPC) methods, their quality characteristics are evaluated, and the values of each sample are graphed using both the average value and range (variation) for each sample. After a given number of samples have been tested and the results plotted as a control chart, upper and control limits are calculated and used to determine whether the process is "in control", and predictable, and also the statistical probability that a part will be out of spec.

Attribute plans can be likened to ranking weather only as hot or cold, while variables methods examine the exact temperature and trends. The application of SPC methods decades ago allowed our semiconductor and automotive industries, who had used obsolete attribute methods, to survive the challenges of Japanese manufacturers who used SPC to deliver superior products.

SPC is best implemented by the manufacturer of mass produced parts, since it requires regular sampling and tracking of each individual production line. This is uneconomical for most end users who would have to station a knowledgeable representative at each plant. Media Sciences recommends hybrid methods whereby test results are ranked as acceptable, minor defects, major defects, and critical defects, each subject to a different LTPD. This approach is contained in specifications and test plans that Media Sciences can prepare for its clients.

When conducting tests, Media Sciences also provides a summary, or overview of test results, containing a list of major and minor positive variances as well as critical, major, and minor defects. Such product evaluations, when properly used, provide our customers with highly useful information that allows them to clearly differentiate between potential suppliers and to avoid unnecessary field failures when quality is regularly monitored.