|
Activity Number:
|
226
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, August 3, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
| Abstract - #303831 |
|
Title:
|
Exploratory Data Analysis at the Boundary of Statistics and Engineering Design Optimization
|
|
Author(s):
|
Theodore Allen*+ and Mikhail Bernshteyn and Ravishankar Rajagopalan
|
|
Companies:
|
The Ohio State University and Sagata Ltd. and The Ohio State University
|
|
Address:
|
1971 Neil Avenue, Columbus, OH, 43221,
|
|
Keywords:
|
Regression ; Robust Design ; Design of Experiments ; Taguchi Methods ; Bayesian Modeling ; Plotting
|
|
Abstract:
|
Two types of plots are described to clarify how modeling decisions interact with system design decision-making. The first type is called "consequence-likelihood diagrams" with Box-Meyer posterior probabilities as the likelihood measure and box-and-whisker plots of the objective function under various hypotheses representing the consequences. The usefulness is illustrated with real world applications in welding engineering. The second type of plot is called "tolerance-box yield plots" which help the user explore robust design decision-making. These plots are based on regression modeling with interactions in the model and the possible inclusion of noise factors. Tolerance-box yield plots are compared Taguchi marginal plots. The benefits include a natural ability to integrate multiple responses, incorporate separate models for mean and variance, and display the yield which is intuitive.
|
