This paper is concerned with the problem of how to take account of the variability of the fatigue crack propagation rate da/dN in the structural reliability analysis. Specifically, various methods for randomization of the parameters in the crack propagation law da/dN=C(ΔK)m or da/dN=C0(ΔK/K0)m were examined. These methods were divided into two classes. In the first class, randomization was made based on the inter-specimen variability of the parameters, whereas in the second class randomization was made based on the intra-specimen variability of the parameters. With respect to the first class, the following three methods were examined for da/dN=C0(ΔK/K0)m: (a) m and log C0 are treated as normal random variables independent of each other, (b) only m is treated as a normal random variable, and (c) only log C is treated as a normal random variable. By using each of the three methods, the variability of the crack propagation life N was analysed, where N was defined as the number of repeated cycles for crack growth from a=a1 to a=a2. In this calculation, a1 was fixed and a2 was varied. The results of the calculation as a function of a2 were compared with the authors' experimental result. It was shown that the method (a) was the most reasonable among the three methods. With respect to randomization based on the intra-specimen variability of the parameters, the method was examined in which only C0 was treated as a spatial random variable in da/dN=C0(ΔK/K0)m. It was shown that this method could also predict the experimental tendency of the variability of the crack propagation life as a function of a2.
