Stochastic analysis of peak pressures on building façade from Wind Tunnel measurements

Abstract

The design of high-rise buildings is mainly dictated by wind-induced loads. An important part of the design is the façade. Nowadays, they are usually made up of a metallic structure where glass panels are attached to with cladding elements. They are submitted to pressures either positive or negative (suction). The latter can be a design challenge. To assess the values of those pressures, wind tunnel tests on scaled models are performed. The output data is the time history of the pressure. It is scaled into a non-dimensional form, the pressure coefficient. To get to a design value of load, the common practice is to first proceed to time-filtering (typical windows of 10 min), then assign the resulting value to a relying area. Recently, due to the increase of performance of the instrumentation, which now have higher data acquisition frequency, some extremely negative pressures have been observed. Those peak events are characterised by a very short duration and happen locally. Therefore, they are simply neglected when filtered. One important question is, apart from understanding what flow conditions create them, if and how they should be accounted for. To answer this problematic, a large campaign of tests was conducted at the wind tunnel of Politecnico di Milano. The approach was to have specific pre-defined zones with a large number of closely spaced taps. This leads to a very dense spatial description of the acting pressure.

This work proposes to study those pressure data through their probability density functions. The high negative pressures make the processes highly skewed and the common model (Hermit model) to assess the peak factor cannot be employed, as the condition of applicability (monotonic criterion) is not verified. To get around this, the pdf are decomposed using a bimodal model of skew gaussian distributions. Once decomposed, the statistics of both component, each linked to a different physical phenomenon, can be accessed. A new evaluation of the peak factor is made. Along with the pdf, the autocorrelation function is also modelled and decomposed to estimate the time-scales associated to both component respectively.