Dear Allstat,
Please find below details of the next event organised by the Business
and Industrial Section (BIS) of the Royal Statistical Society. All are
welcome, but please let RSS know by email so that we can arrange
refreshments.
Kind regards,
David
---
Screening Designs for Industrial Experiments
Wednesday, 4th June 2014, 16:00 – 18:15
Location: Royal Statistical Society, 12 Errol Street, London, EC1Y 8LX
"Design and Analysis of Screening Experiments when Factor Interactions
are Present"
Angela Dean, The Ohio State University USA and University of
Southampton UK.
It is increasingly common for screening experiments to be run during
the development of high quality products and processes. Screening
involves sifting through a large number of potentially important
factors to search, as economically and effectively as possible, for the
few "active" factors. These are factors whose influence on the measured
response is sufficiently large to be of value in improving the system.
The active factors are followed up in later studies for building
detailed models for prediction and optimization. In complex systems
where there are several aspects which must function efficiently
together, studies are needed to discover how factors interact. It is
then vital that a screening strategy can identify active interactions
as well as main effects.
This talk describes two types of screening design (supersaturated
designs and two-stage group screening) illustrated by experiments run
in the automotive engine and lubricant oil industries. A variety of
data analysis methods is investigated and recommendations on the choice
of screening strategy are provided through simulation studies.
Joint work with Danel Draguljic, David Woods, Susan Lewis, Anna Vine
"Designed experiments for semi-parametric models and functional data
with a case-study in Tribology"
Dave Woods, University of Southampton, UK
Experiments with functional data are becoming ubiquitous in science and
engineering with the increasing use of online monitoring and
measurement. Each run of the experiment results in the observation of
data points that are realised from a smooth curve. Although large
quantities of data may be collected from each run, it may still only be
possible to perform small experiments with a limited number of runs.
We describe statistical methodology for an example from Tribology,
concerning the wear-testing of automotive lubricants. Here, we
investigated how lubricant properties and process variables affected
the shape of a functional response measuring wear. Novel techniques
were developed for the initial design of a screening study where the
levels of some of the factors could not be set directly. A two-stage
semi-parametric modelling approach was applied, using a varying
coefficient model and principal components. New methods for the design
of follow-up experiments for such models were also developed and
applied. In addition to the new methodology, we present conclusions
from the case study about which factors had substantial effects, and
how they influence the shape of the wear curves.
Joint work with Chris Marley and Sue Lewis (University of
Southampton).
Open to both RSS and non-RSS members
Please register via email to [log in to unmask] so we can plan
catering.
Tea and coffee will be served after the first talk
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|