RSM and SPC
Tuesday 3 June, 2:40--4:30
- Moderator: Tammy COFFEY
- Wyeth-Ayerst Research
- Dennis COX, Mark GEBERT
- Rice University, Houston, TX, USA
A Bayesian approach to nonparametric
response surface methods consists of a flexible
stochastic process prior for the response function
and a model for the noise. The proposed procedure
is a sequential design methodology which seeks to
balance between exploring regions of the design
space which are relatively unknown (as determined
by the posterior variance of the response) and
local searches in promising regions (as determined
by the posterior mean of the response). An application
is described which involves correlated errors in
addition to a highly complex response surface.
[Dennis D. Cox, Department of Statistics, Rice University, P.O. Box 1892, Houston, Texas 77251, USA, dcox@stat.rice.edu]
- Radhika RAMJEE
- Stevens Institute of Technology, Hoboken, NJ, USA
- Nuno CRATO, Bonnie K. RAY
- New Jersey Institute of Technology, Newark, NJ, USA
Statistical process control charts have been widely used in
industry. Often the least appropriate assumption has been that the process
data to which these charts are applied are uncorrelated. In reality this
assumption is often violated. The presence of autocorrelation has adverse
effects on the performance of control charts, causing an increase in out of
control signals.
The purpose of this paper is to monitor the behavior of control charts on
data sets having persistent properties. The performance of Shewhart, EWMA
and CUSUM charts in presence of long-memory processes is studied by
analyzing the out of control signals obtained both in the mean and range
charts.
[Radhika Ramjee, Department of Mathematics, Stevens Institute of
Technology, Hoboken, NJ 07030, rramjee@stevens-tech.edu]
- Chun K. CHAN, Mohcene MEZHOUDI, Michael TORTORELLA
- Bell Laboratories, Lucent Technologies, Holmdel, NJ, USA
Robust design (RD) and reliability engineering
are technologies promoting the realization of reliable products.
We describe a conceptual framework for unifying
them to increase the effectiveness of failure avoidance.
In RD, the distribution of a product performance factor is
compared to its acceptable window of operation;
units in the population that fall outside the
window are failures (alternatively, a ``reliability loss function'' could be
defined, with ``failure'' defined in terms of this loss function).
RD aims to center the distribution within the window and decrease
its dispersion so that the number of units that fall outside the
window is suitably small.
The added feature that reliability engineering provides is
quantitative treatment of the way these distributions drift over time,
thereby permitting estimation or prediction of the number of failures
(units that will drift outside the acceptable window as time passes).
We include a simple example of resistance drift that also shows the
role of physical information in the theory.
[Michael Tortorella, Bell Laboratories, Room 2L-536, 101 Crawford's Corner Rd., Holmdel, NJ 07733, mtortorella@lucent.com]
Back to JRC program
Back to JRC Home Page