Probabilistic fatigue life prediction using AFGROW and accounting for material variability

The fatigue performance of components contains significant amounts of scatter, and variability has been characterized in initial crack sizes, crack growth rates, and material properties. Probabilistic methods have recently been gaining acceptance as an approach to account for uncertainty in various sources to predict component fatigue life. However, computation time associated with the accepted standard Monte Carlo method can be prohibitive during design phase evaluations. Accordingly, the objectives of the current study were to develop a probabilistic interface for the AFGROW life prediction software and to demonstrate the use of efficient probabilistic methods, as an alternative to Monte Carlo analysis, to accurately predict fatigue lives for three verification cases. The verification cases were based on experimental data for compact tension, single edge notched tension, and single lap joint specimens from the literature. Based on experimentally determined distributions of crack growth rate, material properties, and initial crack size, predicted distributions of fatigue life agreed closely with replicate experimental test results. Computation time with the Advanced Mean Value (AMV) and FORM methods were reduced by 100-fold compared to Monte Carlo, promoting the notion of utilizing probabilistic assessments within the design process.