A new fatigue reliability technique has been developed using a strain-
based analysis. A probabilistic strain-life dime, where the variabilit
y in cycles to failure at constant strain range has been modeled with
a three-parameter Weibull distribution, has been incorporated into the
strain-based fatigue analysis. This formulation which includes a notc
h strain analysis, rainflow cycle counting and damage accumulation acc
ording to Miner's rule, is used to estimate fatigue life to crack init
iation for notched components using smooth specimen laboratory data. U
nlike other probabilistic fatigue models, the technique developed here
does not include a distribution model for stress peaks such as the co
mmonly-used stationary narrow band Gaussian random process assumption
but rather uses strain histories directly. Using this model, technique
s have been developed to estimate the number of cycles to failure at a
specified reliability and to predict the reliability and failure rate
at a specified time in the analysis.