ABSTRACT
ABSTRACT Bridge loading constitutes a complex random process, characterized by a mixture of different loading event types contributing to extreme load effects on bridges. In contrast, classic extreme value analysis applying the block maxima method generally assumes asymptotic models such as Gumbel or generalized extreme value distribution. To account for the complexity of the bridge loading process in statistical extrapolation of extreme traffic load effects on bridges, two different strategies are investigated. Tail fitting aims for data reduction to a subset significant for modeling the extreme value behavior of the random variable. For tail identification, empirical criteria from literature are reviewed and evaluated. Based on selected benchmark results, own empirical criteria for an optimal tail are determined and compared to the ones from literature. Alternatively, fitting of composite distribution models attempts to directly consider the inhomogeneities arising from the underlying bridge loading process. Results from long term traffic simulations evaluated for several structures and response parameters serve for benchmark testing and evaluating the effectivity of the proposed methods.
