Statistical Engineering Division Seminar

Peter A. Parker
National Aeronautics and Space Administration
Langley Research Center
Hampton, Virginia 23681
Statistical Engineering Division Seminar
Friday April 28, 2006, 10:30-11:30 PM
Administration Building, Lecture Room C

Abstract

Classical response surface methodology offers a collection of statistical techniques that are useful for process and product characterization and optimization. However, in practice, we often encounter apparatus-based constraints that make some factors difficult, costly, or time consuming to manipulate, referred to as hard-to-change factors. Consequently, complete randomization of the experimental run order may be infeasible, resulting in a restriction on randomization. Typical examples of hard-to-change factors include environmental conditions, such as temperature and pressure, and configuration changes that require mechanical disassembly. Alternatively, after a hard-to-change factor is set, there is an opportunity to obtain relatively inexpensive information about other factors that are easier to change. In this presentation, classical response surface designs are extended to accommodate restrictions on randomization with a second-order split-plot structure, thereby minimizing the manipulation of the hard-to-change factors. Examples from aeronautical research are used to illustrate experimental design planning, construction methods, and evaluation criteria. In particular, a class of second-order split-plot designs is presented in which ordinary least squares is equivalent to generalized least squares for model estimation; providing best linear unbiased estimates of the model coefficients that are independent of the variance components. Throughout this presentation, compromises among multiple design criteria are emphasized to develop effective experimental strategies that accommodate hard-to-change factors.

Biographical Information

Peter Parker is a member of the Aeronautics Systems Engineering Branch at the National Aeronautics and Space Administration's Langley Research Center in Hampton, Virginia. He has over 16 years of experience in the development of measurement systems used in experimental aeronautics. He holds a B.S. in Engineering from Old Dominion University (1989), a M.S. in Applied Physics and Computer Science from Christopher Newport University (2000), and a M.S. (2003) and Ph.D. (2005) in Statistics from Virginia Tech. His current research interests include experimental design and analysis, response surface methodology, and statistical quality control and improvement.