Statistical fatigue data of various materials at different conditions are increasingly required by engineers for reliabillity design purposes. The present paper describes a new way of analysing fatigue data, which is promising to take out maximum statistical informations from ordinary small sample test results. The method is essentially based on examining the distribution in relative strength deviation of individual test data, evaluated against the mean strength at the considered life level. The Probit analysis using weighting coefficients, under the condition of a known variation coefficient, is quite suitable for the mean strength estimation. Starting with a provisional variation coefficient, the test data could be transformed into a set of relative deviation values in strength; a real variation coefficient is then evaluated according to the theory of multiply censored order statistics.
The proposed method could also be used most effectively in analysing a pile of small sample data sets plotted on a common strength scale, e.g., the ratio against tensile strength of the material.
