I'm wondering what alternatives there are to a grid convergence study to judge solution accuracy for a given grid resolution when doing steady-state RANS simulations on an automatically generated complex unstructured co-located grid (using OpenFOAM). Ideally, such a criterion would lend itself to automatic grid refinement.
One of the conventional methods to assess the spatial convergence of RANS simulations is to perform grid refinement studies based on Richardson extrapolation. Its formulation and practice can be found in the link.
In addition, one should not select the grid refinement ratio arbitrarily. As a rule of thumb:
The grid refinement ratio (i.e. r) should be a minimum of r >= 1.1 to allow the discretization error to be differentiated from other error sources (iterative convergence errors, computer round-off, etc...).
The terms such as accuracy of a CFD simulation should always be formalised for a serious study. This link might be a good start to those who may concern.