Sponsoring Section/Society: ASA Quality and Productivity
Session Slot: 8:30-10:20 Monday
Estimated Audience Size: xx-xxx
AudioVisual Request: xxx
Session Title: Utilizing Large Industrial Data Sets for Product and Process
Quality Improvement
Theme Session: No
Applied Session: Yes
Session Organizer: Grimshaw, Scott D. Brigham Young University
Address: Brigham Young University Department of Statistics Provo, UT 84602
Phone: (801) 378-6251
Fax: (801) 378-5722
Email: grimshaw@byu.edu
Session Timing: 110 minutes total (Sorry about format):
Opening Remarks by Chair - 0 minutes First Speaker - 30 minutes Second Speaker - 30 minutes Third Speaker - 30 minutes Discussant - 10 minutes Floor Discusion - 10 minutes
Session Chair: Grimshaw, Scott D. Brigham Young University
Address: Brigham Young University Department of Statistics Provo, UT 84602
Phone: (801) 378-6251
Fax: (801) 378-5722
Email: grimshaw@byu.edu
1. Interrogating Large Industrial Databases
MacGregor, John, McMaster University
Address: McMaster University Department of Chemical Engineering 1280 Main Street West Hamilton, Ontario, Canada L8S4L7
Phone: (905) 525-9140 x24951
Fax: (905) 521-1350
Email: macgreg@mcmaster.ca
Abstract: It is quite common in the process industries for on-line computers to routinely collect measurements every few seconds or minutes on hundreds of process variables. As a result most industries have very large databases containing non-causal data. Fisher's famous statement that ``all one can do with such happenstance data is a postmortem to see what they died of'' is a warning to all those that might try to do too much with such data.In this presentation we show that with the use of latent variable models and estimation methods based on them (PLS/PCR) one can still extract very useful information from such large multivariate databases for problems such as: (i) multivariate SPC; (ii) building soft (inferential) sensors, (iii) finding process conditions to yield a product with specific properties, or to make the same product in different plants; (iv) establishing multivariate specification limits for product properties; (v) and in certain cases finding directions of steepest ascent for RSM. A key point in the success of these methods is understanding the nature of these databases and how they were generated.
The approach to using latent variable models to interrogate large databases for these problems will be presented and illustrated with some industrial applications.
2. A SAS Procedure for Partial Least Squares and Related Methods
Tobias, Randy, SAS Institute Inc.
Address: SAS Institute Inc. SAS Campus Drive Cary, NC 27513-2414
Phone: (919) 677-8000 x7933
Fax: (919) 677-8123
Email: sasrdt@unx.sas.com
Abstract: Chemometrics, which is the field of statistical analysis for chemical processing, often faces a problem that researchers in other fields would envy--too much data! Typically, instruments monitor and control dozens to hundreds of aspects of a complicated manufacturing process. Analysis must reveal major modes of variation in product quality characteristics, and must identify important factors for predicting and ultimately controlling these characteristics. The method of partial least squares (PLS) is a popular technique for achieving these objectives.This talk will focus on a new SAS/Stat procedure implementing PLS and related methods. I will discuss the commonalities of the different procedures I've chosen to implement, give an overview of the syntax and structure of the procedure, and present several examples of its use, in conjunction with other features of SAS software for data manipulation and graphics.
3. Partitions and Hierarchies for SPC with Massive Datasets
Runger, George C., Arizona State University
Address: Arizona State University Industrial and Management Systems Engineering Tempe, AZ 85287-5906
Phone: (602) 965-3193
Fax: (602) 965-8692
Email: runger@asu.edu
Abstract: Increases in information technology continue to increase the size of the datasets and the number of variables available for SPC. An overview of the impact of massive datasets on traditional multivariate SPC will be presented. Specific work on the use of hierarchies and contrasts to decompose the process control problem will be presented. Applications to semiconductor manufacturing and coated products will be made.
Discussant: Grimshaw, Scott D. Brigham Young University
Address: Brigham Young University Department of Statistics Provo, UT 84602
Phone: (801) 378-6251
Fax: (801) 378-5722
Email: grimshaw@byu.edu
List of speakers who are nonmembers: None