Determining mesh sizes

[mcintya]

Is there a method for pre-determining an appropriate mesh density? I know one method is to keep refining the mesh size until you get a converged solution. This is fine for linear analysis but I don't want do use the same methodology for non-linear analysis as there is considerable wait time between solves. What I have been doing is running a linear analysis, refining the mesh in areas where there are high stress gradiants and then running the non-linear analysis. Is this reasonable?

Any comments / words of wisdom would be appreciated.

Thanks