A statistical model for lifespan prediction of large steel structures

Abstract

The fatigue life of steel structures under operating conditions inevitably depends on various random factors. Among the most influential factors are the characteristics of load cycles, such as stress means and amplitudes. A knowledge of their probability distribution is thus crucial for fatigue life analysis and prediction. Finite probabilistic mixture models have previously been used for this purpose. This paper presents a study of the possible benefits of mixture models with log-normal components, using a large experimental data set from the slew bearing substructure of a stacker. The study shows that for this particular situation, the log-normal mixture model performs significantly better than Gaussian mixtures, and thus can be used as a suitable model in similar areas of application.