Modeling Variability in Service Loading Spectra

This paper describes a methodology for statistically extrapolating a single measured service loading history to the expected long-term service usage spectra. The measured time history first is processed into a rainflow counted histogram. Nonparametric kernel smoothing techniques are employed to convert the rainflow histogram of cycles into a probability density histogram. Once the probability density histogram is obtained, Monte Carlo methods are used to produce a rainflow histogram of any desired number of cycles. A new loading history then is reconstructed from the expected rainflow histogram, which can be combined with a probabilistic fatigue analysis to obtain an estimate of the durability of a structure. Obtaining an estimate of the loading spectra for a ground vehicle is difficult because there are many users, each with different service usage. The extrapolating methodology is extended to combine data from several users to obtain loading spectra that represent more severe users in the population.