I am trying to define a benchmark for a Karamba brep model but whenever I change the mesh resolution my results of stress change in the final chart, so I’m wondering if there is anything wrong in my script or how do I define a benchmark that won’t make the results of the stress change that drastically?
The proper way to do this is to do a mesh refinement study and analyse if the results converge to a value for finer and finer meshes. Google for Uncertainty quantification.
As @menno has mentioned, A course mesh (large mesh sizes) can give you inaccurate results. A general rule of thumb is to reduce the mesh resolution so that the difference in results is very minimal. However this can have a significant affect on your calculation time.
Thanks, so is there a way I can know what would be the ideal mesh scale for my model size? Or how to define a benchmark to confirm the model inadequacy, model bias, or model discrepancy? So far the smallest mesh I’ve tried is 0.1m but it’s result is a very heavy model to analyze, is there any way to do this without a mesh resolution and just a surface? Would that correct the model discrepancy?
No, you can’t know the ideal mesh scale without such a convergence study, at least once. Then you know for comparable models what the minimum requirements for your mesh are to get results that are not fully untrustworthy.