By using the fatigue crack growth test system developed in the previous study, statistical data of fatigue crack growth rate da/dN in 2024-T3 aluminum alloy were obtained. Eighteen specimens were tested. In the test system used, the crack length was measured using the DC electrical potential method and precise load control was made using a microcomputer. Thus, scatter in testing conditions and labor needed were decreased as compared with the previous test by the present authors in which the crack length was measured using a travelling microscope. From analyses of the experimental data, the coefficient of variation, ηΔN, of the number of stress cycles for crack growth from a given value of a to a+Δa (Δa=0.1mm) was found to be nearly constant independent of a. This implies that m in Paris' law da/dN=C(ΔK)m can be regarded to be deterministic in the present data. Then, the composite variability model as a stochastic model of da/dN proposed previously by the present authors was examined using the present data. In this model, C is treated as a random variable involving the inter-specimen variability and the intra-specimen variability, whereas m is treated as deterministic. From comparison of the model with the data, the inter-specimen variability was found to be considerably smaller in the present data than in the previous data. This seems to be due to the decrease of scatter in testing conditions. Furthermore, the correlation distance δ of the intra-specimen variability of C was found to be much smaller in the present data than in the previous data. The value of δ estimated from the present data was of the order of a grain size. This suggests that the intra-specimen variability of C occurs from grain to grain.
