Numerical calculation of wind loads over solar collectors

For non-standard structures, such as parabolic trough collectors, the application of codes of practice to compute load coefficients usually results in oversized wind loads. Therefore, it is common practice to support the analysis experimentally with wind tunnel test results. This approach, however, is time-consuming, expensive, flow intrusive and requires a new set of tests for any modification in geometry, configuration, topography or load conditions, entailing additional time and expense. The results obtained are restricted to a finite set of points and variables, and most importantly, the size and wind velocities in available boundary layer wind tunnels impose Reynolds numbers well below those occurring in open air full-scale structures of this type. This similitude limitation prevents a direct and fully accurate extension of the results to the characterization of full scale collector structures. These wind tunnel shortcomings have positioned CFD as an appealing alternative for determining wind load distributions over solar collectors, solving most of these problems. Although extensively used in other applications, however, the application of CFD methods requires additional development, further testing and verification, proper modeling conditions and additional validation to gain acceptance as an alternative to physical wind tunnel tests. In this paper, CFD has been used as a virtual wind tunnel to compute the three-dimensional flow around a single model-scale module for a range of yaw and pitch angles, and the resultant load coefficients have been compared with those obtained experimentally in a physical wind tunnel. After validation against experimental data, the computational methodology has been applied to compute the wind loads on a full-scale module (including the complete supporting structure) and an array configuration of such modules. It will be shown that the relative mean errors of the numerical results with respect to the reference experimental data are within 10%, thus of the same order as experimental uncertainty. (C) 2013 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).