MULTIVARIATE EQUIVALENCE TESTING FOR COMPOSITEMATERIALS

摘要

At the National Center for Advanced Materials Performance (NCAMP), we are developing alibrary of information about composite materials. One goal is to examine the data from a set ofqualification tests, which are used to develop basis values for the material, and an equivalencydata set showing performance of the same material either produced by a different manufactureror with slight changes to the recipe and conclude if they are sufficiently similar to be consideredequivalent. Tests are performed for many different properties under various environmentalconditions.The current approach examines each set of test results individually and determines if thatproperty and condition combination is statistically significantly different at the 95% level ofconfidence. Since there are many sets of tests, the probability of at least one failure is high andvirtually all equivalency’s fail on one or more test conditions. The final assessment of thevarious test results is a subjective decision based on engineering judgment.In order to reduce the subjectivity of this assessment, a multivariate approach is proposed. Thispaper will discuss a proposed procedure to compare multivariate normal data from two samplegroups including their covariance relationships and will set an objective decision criteria todetermine equivalence based on the combined results of multiple tests.