|
Statistical Tolerancing
Statistical tolerancing encompasses two main
activities. When data can be obtained directly for some
“parameter” of interest, statistical tolerance limits
can be computed such that a “high” proportion of the
population will be within the limits with some “high”
level of confidence. Statistical tolerance limits are
useful in setting specifications since they provide an
objective view based on measured process capability.
Statistical tolerance limits are not, however, the only
consideration in setting product specifications which
would be warranted to the customer.
A
second activity within the area of statistical
tolerancing is to infer “system” level tolerance limits
from the tolerance limits of the “components” going into
the system – the classic tolerance stack problem. One
would consider doing this activity when the system
variation cannot be measured directly or it is
inconvenient to do so. The end result of working
through the tolerance stack would again be to place
limits on the likely performance of some parameter of
interest at the system level. A topic closely related
to statistical tolerancing is, given the product
specifications, where should the production test limits
be set? This brings in the subject of measurement error
in the test process and the notion of setting test
limits such that some cost function of the test errors
is minimized.
A
paper describing a statistical tolerancing application
for an in-circuit board tester can be obtained by
request.
Larsen, G.A., "Estimating
Contact Reliability Using Tolerances in a Complex PC
Board Test Fixturing System", 12th Annual Rocky Mountain
Quality Conference Proceedings, June 1988, pp. 361-375.
|
