SED navigation bar go to SED home page go to SED projects page go to NIST home page SED Home Page SED Contacts SED Projects SED Products and Publications Search SED Pages

Education and Training: Experiment Design for Engineers and Scientists
Time and Location Experiment Design for Scientists and Engineers
James Filliben and Dennis Leber
Statistical Engineering Division, NIST
Thursday-Friday September 6-7, 2007 8:30am - 4:30pm
Monday-Teusday September 10-11, 2007 8:30am - 4:30pm
Adminstration Building, Lecture Room D
NIST
Gaithersburg, MD
Abstract Experiment design is a systematic, rigorous, data-based approach to scientific/engineering problem-solving. The goal of experiment design is to generate valid, crisp, unambiguous, and reproducible conclusions about the scientific/engineering process of interest--and to do so in a time- and cost-efficient fashion. Statistically designed experiments--especially "orthogonal" designed experiments--markedly enhance scientific insight, rigor, and robustness, while saving both time and money. Such designs have already benefited a variety of NIST projects of both data types:
  1. the usual laboratory data type and
  2. the ever-increasing/important virtual data (= computational experiments) type.
Such NIST projects include, for example, bullet casing forensics, high rise building safety factors, highway concrete durability, optical microscopy, carbon nanotube purification, bio-cartilage regeneration, nanocantilever compliance coefficients, soil leeching, dental polysac adhesion, Apache-Linux net processing time, SMS: smart machining systems, peak deconvolution algorithm accuracy, HRTEM error in gate dielectrics, Abilene network performance, trace plutonium detection, automated machine-tool performance, RAVE visualization facility calibration, security and radiation detection, World Trade Center FEA impact analysis, etc.) Such NIST examples will be drawn on as needed during class.

The class itself covers the fundamental principles and techniques for the

  1. organization,
  2. construction, and
  3. analysis
of designed experiments. This class illustrates the powerful problem-solving role of designed experiments by application to a wide variety of scientific/engineering problems.
Objective The objective of this course is to provide to the scientist/engineer both the "why" and the "how" of the experiment design construction and analysis tools necessary to allow the scientist/engineer to efficiently evaluate, characterize, optimize, and model their instrument/process. Upon completion, the scientist/engineer will be able to
  • classify his/her problem into the proper design category;
  • translate his/her problem into design essentials;
  • construct the proper time/cost-efficient design;
  • run the design in a conclusion-protective fashion;
  • analyze the data ensuing from the design with quantitative
  • and state-of-the-art graphical/visual techniques;
  • gain insight into the underlying physical mechanisms of a system; and
  • extract from the analysis the appropriate conclusions regarding

    1. evaluation: whether or not a parameter had an effect on the overall system response(s);
    2. characterization: what are the important parameters and interactions in a system;
    3. optimization: what are the optimal settings for system parameters;
    4. modeling: what is a parsimonious prediction equation for the system.
Participants will organize and construct an optimal design for their own scientific/engineering project. The classroom learning experience will be reinforced by in-class exercises.
Audience All members of the NIST scientific/engineering/technical staff; non-NIST personnel on a "space-available" basis.
Class Outline
  1. Design of Experiment (DEX) Fundamentals
  2. Problem Classification
  3. DEX Principles & Techniques
  4. Comparative Designs
  5. Full Factorial Screening Designs
  6. Fractional Factorial Screening/Sensitivity Designs
  7. Taguchi Screening Designs
  8. Regression Designs
  9. Optimization Designs
  10. Conclusion
Comments on Course CLASS SIZE IS LIMITED TO 40.
REGISTRATION FEE IS $150.

An instructor notebook will be provided for the class, and a textbook (Box, Hunter, Hunter: Statistics for Experimenters, Wiley) will be included as part of the class fee.

Further Information For further information, contact or register online (as of August 29, 2007, the class is full).

Date created: 8/29/2007
Last updated: 8/29/2007
Please email comments on this WWW page to sedwww@nist.gov.

[ SED Home |  Education Home |  Previous ] Next ]