IsaacGentle
commented
4 years ago Hi there, I am helping out yannmorisseau (see above) and have found some interesting behaviour which might give further insight into the cause of the problem.
Issue - when using the Specular_Standard_Material model for a parabolic dish model, inconsistent results are observed for various high resolution meshes
Observations - 1) The issue is can be replicated on different meshes types (triangle, curvilinear grid).
2) The number of mesh elements does not impact the size of the overestimation of the peak concentration, but the mesh 'aspect ratio' does (ratio of n_\theta to n_r in Fig. 1)
3) This behaviour does not appear when using the rough material model (with sigmaSlope = 0)
Figure 1 - Ideal model is the inbuilt parabolic dish, mesh is a curvilinear grid which is divided into n_\theta radially and n_r facets (see inset plot). spec is Specular_Standard_Material model (sigmaSlope = 5), rough is Specular_Rough_Standard_Material model (with sigmaSlope = 0 and sigmaSpecularity = 10).
We can provide a minimal working example but the files are large due to the mesh size. If this is useful, we could use google drive etc.
Thanks for the open software and thanks in advance for looking into our issue.
We are comparing different mesh approximations of the ideal parabolic dish and have noticed some weird unexplainable behaviour when we apply the Specular_Standar_Material model. Specifically, for no slope error the results look the same, whereas when a slope error is applied we get an over-estimate of the peak concentration:
Assuming the in-built Tonatiuh parabolic dish is accurate, this apparent improvement (higher peak flux) for large size meshes should be impossible as we expected that any representation of the parabolic shape as a mesh should spread the flux distribution more than the analytical model. Furthermore, the size of the over-estimate actually increases as the mesh is refined (number of mesh facets increase) contrary our expectations:
Have you ever experienced this issue before? Are there any known bugs for Specular_Standar_Material applied to very high resolution meshes?
NB: Flux Analysis were run with N=10e8 rays, the reflectivity was set to R=0.55, Slope error was set to S=0 mrad and then S=4.5 mrad to observe the discrepancy, the sun environment is the Pillbox_sunshape provided by Tonatiuh.