SED navigation bar go to NIST home page SED Home Page SED Contacts SED Projects SED Products and Publications Search SED Pages
Contributed Session: Statistical Methods in Quality Improvement

Contributed Session: Statistical Methods in Quality Improvement

Session Chair: Carroll Croarkin, NIST
 

A Designed Experiment for 747 Aircraft "Hump Section" Frame Mismatch

Bob Haukap
Process Engineering, Boeing Commercial Airplane Group

The Boeing 747 fuselage is assembled from large body panels at a factory in Everett, Washington. The frames of the panels are designed to fit one another, but at times they don't match up. These frame mismatches, while small, must be corrected with excessive rework during the build process. Many possible variables could contribute to this problem including variability in the panels coming from suppliers, variability in the build/assembly process and variability in the use of the tools, such as jacks, clamps, and spreader bars, during the build process.

Based on the large number of factors potentially affecting assembly, cost considerations, and other issues, a screening experiment was performed using a 2^13-8 fractional factorial design. With this design we were only able to discern "main effects" yet this proved to be useful as it was possible to eliminate at least seven factors as not being significant.

A follow-up experiment is in the planning process to further study the remaining significant factors and their possible interactions

[Bob Haukap, Boeing Commercial Airplane Group, P.O. Box 3707, MS 0X-EL, Seattle, WA 98124-2207 USA; evie179@kgv1.bems.boeing.com ]

 
What a Difference a Day Makes: Before and After a Deming Lecture in Ex-Yugoslavia

Charles J. Stiffler
Vesna Luzar-Stiffler
Croation Applied Industrial Research Center

This is basically a multivariate change score analysis on a small sample of individuals who were exposed to their very first Deming lecture. The purpose of the study was to measure the short term, "single shot" impact of Deming's ideas on the attitudes and beliefs of Slovenian managers and engineers. The method was to gather attitudes, beliefs and demographic information immediately prior to and after a 2 and one half hour lecture on the basics of TQM (including a brief red bead experiment). Compatibility of the samples (before vs. after) was determined in two ways: 1) as crosscorrelations among before and after factor scores, and 2) as congruencies of the corresponding factor patterns. Analysis indicated a significant change on majority of 17 items, but also a differential impact based on subpopulation (engineer vs. manager). The findings confirm the delivered theory that one must study the audience's initial position and variation so as to maximize message impact.

[Vesna Luzar-Stiffler, Croation Applied Industrial Research Center, Vlaska 26, 10000 Zagreb, CROATIA; Vesna.Luzar.Stiffler@srce.hr]

 
Structural Models & Operational Measures in Analyzing Customer Satisfaction

Jarrett K. Rosenberg
Sun Microsystems

The most common method of assessing customer satisfaction is the survey. Analyses of such surveys, however, are typically very superficial and yield little insight into the relationships among the obtained responses. Furthermore, there is rarely an explicit linkage between the company's actions (as measured internally by operational process metrics) and the effect they have on customer attitudes (as measured by satisfaction survey responses). This paper describes the design and analysis of a suite of customer service satisfaction surveys which use structural (latent variable) models to elucidate the structure of satisfaction, both in terms of customer attitudes and their connection to internal operational measures such as response time.

[Jarrett Rosenberg, Sun Microsystems, 2550 Garcia Avenue, MS MPK17-307, Mountain View, CA 94043 USA; Rosenberg@Eng.Sun.COM ]

 

Using Basic Statistical Thinking in Industrial Situations

Bob Haukap
Process Engineering, Boeing Commercial Airplane Group

Many problems arise in industry that do not fit traditional statistical assumptions. This paper uses basic data exploration skills to recognize when these assumptions do not hold and provides examples with solutions that have arisen in an industrial setting. Issues of normality, linearity and censoring arise when trying to apply statistical techniques from basic statistics textbooks. Simple graphs of the data can readily reveal non-normality and non-linearity and often suggest a suitable transformation to normality or linearity.

The three examples selected address complications that arise when constructing tolerance and prediction intervals due to non- normal, non-linear and censored data. The first example illustrates sample size selection for constructing a tolerance interval. The second example involves constructing tolerance intervals for an engineering commitment with skewed and censored data. The last example involves a non-linear modelling situation with a corresponding prediction band.

All of these examples result in conclusions that require the user to understand concepts from basic statistics textbooks. The underlying calculations are somewhat more complicated, yet the explanation to the end user (customer) remains straight forward.

[Bob Haukap, Boeing Commercial Airplane Group, P.O. Box 3707, MS 0X-EL, Seattle, WA 98124-2207 USA; evie179@kgv1.bems.boeing.com ]

Date created: 6/5/2001
Last updated: 6/21/2001
Please email comments on this WWW page to sedwww@cam.nist.gov.