Advanced fatigue reliability analysis

Abstract Factors in fatigue life prediction models can be treated as random variables reflecting inherent uncertainties in material behaviour and loading. Given the probability distribution of these factors, U , the distribution of life, N , is derived when a reliability analysis is performed. Numerical reliability analysis methods require a large number of function evaluations of N(U) . But algorithms to estimate N , such as local strain analysis to predict crack initiation or fracture mechanics to predict crack propagation, are complicated and may require significant computer time for only one function evaluation. An advanced mean value (AMV) method has been proposed as an efficient scheme for deriving a probability distribution of an implicit function using a minimum of function evaluations. The AMV method is reviewed in the context of the fatigue problem. Examples of reliability analysis for local strain and fracture mechanics fatigue models illustrate that the AMV method produces accurate probability estimates with a minimum of function evaluations.