Abstract
This paper presents a generalized approach for predicting (i.e., interpolating) the magnitude and distribution of roof pressures near separated flow regions on a low-rise structure based on freestream turbulent flow conditions. A feed-forward multilayer artificial neural network (ANN) using a backpropagation (BP) training algorithm is employed to predict the mean, root-mean-square (RMS), and peak pressure coefficients on three geometrically scaled (1:50, 1:30, and 1:20) low-rise building models for a family of upwind approach flow conditions. A comprehensive dataset of recently published boundary layer wind tunnel (BLWT) pressure measurements was utilized for training, validation, and evaluation of the ANN model. On average, predicted ANN peak pressure coefficients for a group of pressure taps located near the roof corner were within 5.1, 6.9, and 7.7% of BLWT observations for the 1:50, 1:30, and 1:20 models, respectively. Further, very good agreement was found between predicted ANN mean and RMS pressure coefficients and BLWT data.
Highlights
Boundary layer wind tunnel (BLWT) testing is still considered the primary experimental instrument to accurately reproduce and assess wind-induced loads on building structures
To address gaps in experimental databases, this study makes use of existing experimental boundary layer wind tunnel (BLWT) datasets and artificial neural networks (ANN) to assist in the development of robust and reliable mathematical models for accurately quantifying peak wind loading on low-rise structures and their inherent dependence on freestream turbulent flows
The performance function was chosen as the combined mean squared error (MSE) of the predicted (i.e., ANN) and observed (i.e., BLWT) mean, RMS, and peak Cp values
Summary
Boundary layer wind tunnel (BLWT) testing is still considered the primary experimental instrument to accurately reproduce and assess wind-induced loads on building structures. A robust feed-forward multilayer artificial neural network (ANN) using a backpropagation (BP) training algorithm is employed to analytically predict the mean, RMS, and peak pressure coefficients on the roof of a low-rise structure given the freestream turbulence intensity (at eave height) and the normalized plan roof coordinates. An ANN using a backpropagation training algorithm was employed to predict the distribution of mean, RMS, and peak pressure coefficients on the roof of a low-rise structure from the turbulence characteristics of the freestream. The number of validation checks represents the number of consecutive iterations that the validation performance fails to decrease Upwind terrains for both narrow and wide edge roughness element orientations were used for training the network; including the smoothest (h = 0 mm; i.e., flush floor) and roughest (h = 160 mm, wide edge) Terraformer configurations.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call