Load models for fatigue reliability from limited data

Probability distributions of wind turbine loads are estimated from limited data. The impact of different models on fatigue damage is shown. Common one-parameter probability models, such as the Rayleigh and exponential, are found to give significantly different estimates of load distributions and damage. Greatest differences occur in materials with relatively high values of the S-N exponent b, such as composites. In such cases more accurate damage estimates are found by matching at least two moments of the load data (Weibull model), or still higher moments as well. For this purpose, a new, four-moment ``generalized Weibull`` model is introduced. For edge-wise loads it appears a notable improvement over the basic Weibull model, while it supports the Weibull model in the flapwise case. Uncertainty in damage estimates is also quantified, along with the implied data needs.