A Bayesian method for establishing fatigue design curves

Abstract From a reliability viewpoint, the simple conventional procedure for establishing design S − N curves from laboratory fatigue test data suffers from two important limitations. Firstly, the calculated fatigue reliability on the design curve only reflects the observed “physical” uncertainty associated with the fatigue process itself. “Statistical” uncertainties, connected with estimating the parameters of the fatigue model, are not considered. Secondly, it is not possible to account for fatigue run-outs (non-failures) in a rational manner. A new procedure, based on Bayesian statistical inference, is presented which is capable of handling both the above problems. Its use is illustrated via the analysis of some typical fatigue data sets and a number of general points arising from the results are discussed.