Publication: Forecasting pressure coefficients on a circular cylinder at Re=106 by cognitive classifiers and linear transformations
All || By Area || By Year| Title | Forecasting pressure coefficients on a circular cylinder at Re=106 by cognitive classifiers and linear transformations | Authors/Editors* | X. Gavalda1a, J. Ferrer-Gener1b, G.A. Kopp2, Francesc Giralt1a and J. Galsworthy2 |
|---|---|
| Where published* | To submit |
| How published* | Other |
| Year* | 2008 |
| Volume | |
| Number | |
| Pages | |
| Publisher | |
| Keywords | |
| Link | |
| Abstract |
The prediction of surface pressure time series in buildings is important to avoid structural failure under extreme wind conditions. Numerical calculations such as LES are not readily applicable to estimate loads on buildings at the high Reynolds numbers that occur under atmospheric conditions and the use of databank information is necessary. Nevertheless, not all building geometries are readily available in databanks and interpolation might be necessary. In the current study forecasting and interpolation tools were assessed for surface pressure time series measured at 32 locations over a circular cylinder in cross flow at Re = 106. A system formed by several fuzzy ARTMAT neural algorithms operating in parallel was first applied to capture the dynamics of the measured pressure field and to forecast the instantaneous pressure field over the cylinder surface. Results showed that pressure data measured at all locations and at previous time instants were simultaneously needed to yield accurate time-predictions for pressure, lift and drag. The generated pressure time-series showed, however, deviations mainly in the predicted kurtosis, indicating that three-dimensional and/or blocking effects could be present. As a consequence, the simpler alternative of classifying pressure sensors by means of Self-Organizing Maps (SOM) and obtaining pressure signal estimates within each class of surface sensors was considered. Representative locations for each sensor class were identified and used as proxies to generate pressure signals by linear transformation for the remaining class members, yielding in most cases good approximations up to the 4th moment. Average lift and drag coefficients were satisfactorily predicted. |
Back to page 52 of list