Mixture model approach for peak factor estimation and its application in cladding design

To design reliable structures, extreme pressures and peak factors are required. In many applications of Wind Engineering, their statistical analysis has to be performed by consid- ering their non-Gaussian nature. In particular, in the shear layers and where local vortices are formed, it is commonly observed that measured pressures might exhibit two or more significant components, which can be modeled with mixture models. Mixing two components with unequal weight may result in much larger non-Gaussianities (even if the two components are Gaussian) than a single (non-Gaussian) component. For this reason and some others, it is important to split a mixed flow into its underlying components. In this paper, we propose a novel de-mixing method. Unlike existing approaches which are based on a nonlinear least-square fitting of the probability density function (PDF), the proposed method is supported by the different timescales of the mixed phenomena. The method is briefly described, and two examples of application are given.