Introduction

A run-out in fatigue testing occurs when the specimen endures a predetermined large number of cycles and the test is then terminated to conserve time. The data recorded for a run-out specimen are the alternating stress amplitude imposed (⁠$Sa$⁠) (or equivalent information) and the number of cycles (⁠$Nf*$⁠) defining a run-out for the given test program. The corresponding data recorded for a failed specimen are the observed fatigue life (⁠$Nf$⁠) and the alternating stress amplitude imposed (⁠$Sa$⁠), but the observed fatigue life itself is not subsequently used in statistical analysis to estimate the fatigue limit (that is, the fatigue strength at $Nf*$ cycles). Rather, the fatigue data are analyzed as quantal response data where the relevant issue is only whether each given specimen did or did not fail prior to $Nf*$ cycles. Fatigue limits are generally estimated using only a few specimens. Thus it is vital that each specimen be allocated efficiently to the alternating stress amplitude imposed in testing. But even with optimal allocation the only goal remotely attainable in a test program employing only a few specimens is a reasonably precise estimate of the median fatigue limit based on the assumption that the standard deviation of the fatigue strength distribution is known from prior experience and data. Such median fatigue limit estimates may be sensitive to certain assumptions made in analysis. Thus engineering judgment is required in interpreting these fatigue limit estimates. This publication presents tables generated by alternative analyses, thereby providing the fatigue analyst quantitative means to examine the effect of various assumptions on the resulting fatigue limit estimate. It is expected that this information will enable the analyst to estimate the median fatigue limit using a combination of statistical procedure and engineering judgment.