Browse Prior Art Database

Automated Attribute Pre-control for Infrequent Defects

IP.com Disclosure Number: IPCOM000110498D
Original Publication Date: 1992-Dec-01
Included in the Prior Art Database: 2005-Mar-25
Document File: 5 page(s) / 238K

Publishing Venue

IBM

Related People

Baumann, GW: AUTHOR [+3]

Abstract

This invention is an automated statistical process for quickly identifying negative shifts in manufacturing quality. It uses attribute properties (e.g., good/bad) obtained from either 100 percent inspection or limited sampling. In an ideal environment, it would be implemented in an executable program run in a process control computer. It would automatically receive all quality data (in real time or frequent batches), and it would automatically shut down the process line when the defects exceeded the maximum acceptable criteria. It is designed to address the process quality, rather than the quality of a particular inspection sample. In the software manufacturing environment (such as ISMD), the relevant attribute would be hard read errors of media or pick and pack errors.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 30% of the total text.

Automated Attribute Pre-control for Infrequent Defects

       This invention is an automated statistical process for
quickly identifying negative shifts in manufacturing quality.  It
uses attribute properties (e.g., good/bad) obtained from either 100
percent inspection or limited sampling.  In an ideal environment, it
would be implemented in an executable program run in a process
control computer.  It would automatically receive all quality data
(in real time or frequent batches), and it would automatically shut
down the process line when the defects exceeded the maximum
acceptable criteria.  It is designed to address the process quality,
rather than the quality of a particular inspection sample.  In the
software manufacturing environment (such as ISMD), the relevant
attribute would be hard read errors of media or pick and pack errors.

      Manufacturing processes are typically controlled by statistical
methods such as Shewhart's Statistical Process Control and Military
Standard sampling processes.  Where variable data are unavailable or
inappropriate, attribute data are used (for example, with a Shewhart
P-Type charts and the Mil Spec. sampling, plans are based on batch
inspection, and neither responds quickly to defects nor makes
efficient use of sparse data.  In practice, neither one works when
failures occur less often than several per day (or shorter sampling
period).  Moreover, neither is optimal when the production rates vary
several fold over the relevant production window (e.g., one day or
one shift).

      The new Pre-Control system makes up for these shortcomings.  It
has the following benefits:
1) It automatically tracks production volumes - enabling the fastest
response to quality problems.  (The likelihood of any given number of
defects being shipped prior to a line down condition is independent
of production levels, given the quality level.)
2) It is a continuous process, and hence does not lose, discard, or
disregard information (or signal) from prior batches when evaluating
new data.
3) It is easily automated in a computerized data collection (and
control) system.
4) It enables the unambiguous identification of acceptable quality
levels versus an unacceptable level (in terms of defect frequency).
5) It applies equally well to deterministic failures and
probabilistic ones.  In magnetic recording, most hard and soft read
errors are unrepeatable, but the frequency is repeatable.  Even in
this environment, the claim in 1) above is still true.
6) It is readily adapted to a multi-level process control
environment.  For example, if defects rise above some rate p1, an
emergency task force may be automatically brought into action.  At
some higher rate p2, the manufacturing line would be shut down.

      Process quality control is derived from probability and
statistics, built around the null hypothesis.  In the attribute
environment, the relevant statistic is the binomial process, with
probability...