This preview shows page 1. Sign up to view the full content.
Unformatted text preview: actors -~ Figure 5. Different
sponse Variables. 1 measurable 1 Categories of Factors Affecting Re- 4.2 Current Ability Use (col. 2) to Measure and Set (col. 3) With control variables, there is an additional consideration rarely mentioned in the literature. The
experimentation team not only needs to know how
measurements will be obtained and the precision of
measurement, a,,,, but also how the control variable
settings will be obtained and “setting error,” E,. These
different types of deviation from the ideal have different effects on experimentation. Large 0, will mean
that either errors-in-variables methods will have to
be used (e.g., methods that will allow estimation of
bias in effects estimates) or, alternatively, many samples will have to be collected for measurement during
experimentation to get an acceptably small a,,,lfi,
especially if IE,Jis also large. If IE,~ large, traditional,
class-variable-based analysis of variance will have to
TECHNOMETRICS, FEBRUARY 1993, VOL. 35, NO. 1 DAVID E. COLEMAN AND DOUGLAS Table 3. Control Control variable
(units) Normal level
and range x-axis shift*
Tool vendor O--.020 inches O-.020 inches O--.020 inches Internal, external a-axis shift*
nominal) O--.030 O-.025 Variables Measurement
setting errorhow known?
(experience) - Difference /1 0, .015 inches Difference 7 ? Difference 7 degrees 0, .030 degrees External is more
Unknown 90%. 110% None? ,002 inches
panel) 0, .015 inches Unknown 90%, 110% None? .OOl degrees go-110% inches be replaced by regression analysis. The result of large
setting variation may be unwanted aliasing, greater
prediction error, violation of experiment constraints,
and difficulty conducting split-plot analyses.
Often, one finds that a, = [es/, such as when the
measurement system is part of a controller, and equilibrium conditions can be achieved. Measurement
precision and set...
View Full Document
- Fall '14
- The Land