A procedure for using multivariate normal tolerance regions for the certification of related characteristics
The purpose of this investigation was to develop a complete procedure for the use of multivariate normal tolerance regions for the simultaneous certification of up to 30 related characteristics. A Monte-Carlo integration technique was developed to investigate the accuracy of an approximation to the multivariate normal tolerance region. The results indicated that the approximation is sufficiently accurate for relatively small sample sizes.
The first step in the development of the procedure involved the determination of a method for setting specification regions. The method proposed involves using the univariate specification intervals to define a rectangular solid and using this rectangular solid in conjunction with an estimate of the variance-covariance matrix to inscribe a maximum volume ellipsoid within the rectangular solid. This inscribed ellipsoid is the specification ellipsoid.
The second step in the development of the procedure involved generating a complete set of tables for determining the equation for the calculated tolerance ellipsoid that contains with a certain degree of confidence a proportion P of the population of items. In generating these tables, the most complete set of non-central Chi-Square tables in existence were generated.
The final step involved the development of a mathod for testing if the calculated tolerance ellipsoid is within the specification ellipsoid. If this test is positive then it can be said with 100y% degree of confidence that at least 100p% of the items meet the specification.
Thesis74b.H644.pdf
4.13 MB
Adobe PDF
661902482b40d62686bf7f372c97357c