Estimating extreme wind pressure for long-span roofs: Sample independence considerations

Abstract

Observing wind loads directly in model scale is an effective way to predict extreme loads from the tail of the wind pressure distribution, but to ensure the reliability of the predictions, long-term data may be necessary. Such data are expensive in a wind tunnel. In this study, a criterion is established for dividing a short-term wind pressure sample into independent sub-samples. The observation duration of the sub-samples needed is derived based on mutual information theory. A modified observed extreme method for estimating extreme wind pressure from a short-term sample is proposed which combines an observed extreme method with an extreme value conversion relationship based on the Gumbel distribution. The proposed method is demonstrated by estimating the extreme wind pressures on a flat long-span roof from model-scale test results. The analysis shows that the test data can be segmented independently using the mutual information method. The delay time distribution is revealed to be highly correlated with the turbulence acting on the roof. The compensation value in the extreme value conversion relationship is shown to strongly influence the estimation result. Compared with the commonly-used methods for estimating extreme values, the proposed method has the advantage of more stable and effective statistical results.