| ||Format||Pages||Price|| |
|PDF (236K)||13||$25||  ADD TO CART|
|Complete Source PDF (12M)||415||$142||  ADD TO CART|
Regression analysis is useful when limited data on a material is obtained under various experimental conditions, and design allowables are required for certain (perhaps different) conditions. Since composite material properties typically exhibit considerable variability, and since composite material testing is expensive, the use of regression methodology for the efficient analysis of experimental data is even more important than for metals. However, test data for composites often come from several batches, with nonnegligible between-batch variability. The usual approach when confronted with such data is to combine all batches at each test condition, and then to use conventional regression methodology to determine allowables. This approach does not take the variability between batches into account, and, consequently, the allowables can be nonconservative. New methodology has been developed for regression allowables in the presence of between-batch variability that overcomes this limitation for a wide class of practical situations. This article discusses the interpretation of statistically-based design allowables, particularly for realistic situations where the data come from several batches over a test matrix of fixed conditions. A computer program (available from the author), which is an implementation of the new methodology, is used to analyze a set of graphite/epoxy compressive strength data.
design allowables, regression analysis, tolerance limits, testing, design, batch-to-batch variability, composite materials
Mathematical statistician, National Institute of Standards and Technology, Gaithersburg, MD