Many full-scale component fatigue limits are developed on the basis of less than five tests. Use of statistical concepts on such a limited number of specimens would result in extremely low and unrealistic allowable stresses for any reasonable probability and confidence level.
Fatigue data from element tests and full-scale component tests, where many specimens have been tested, can be used to evaluate a coefficient of variation that is representative of the material and type of component being considered. By use of this information, and with the application of basic statistical techniques, a design S-N-P (probability) curve can be determined for components where only a limited number of tests are available, for any probability and confidence level desired. The methods employed for the evaluation of available data are discussed.
Typical values of coefficient of variation are tabulated for components of typical steel and aluminum alloys such as 4130 and 4340 alloy steel, 150,000 to 160,000 psi ultimate tensile strength, and 2024-T4, 2014-T6, and 7075-T6 aluminum alloys, which contain stress concentrations or areas of fretting corrosion or both.
Suitable probability and confidence levels are selected for components where a single noninspectable failure would result in a catastrophic loss.
Determination of a design fatigue limit can be obtained directly by using the method described in this paper or by applying a factor of safety to the average test fatigue limit that simplifies the determination of an appropriate design S-N-P curve for the designer and structures engineer. A table is presented showing the factor of safety required to ensure a minimum desired probability and confidence level when only a limited number of test specimens is available.