Dear Fadzli,

For any problem in which a residual (or relative change) approaches zero, you would have to define such an "acceptable" value of relative change between different grid resolutions. As long as you report the evolution of your chosen residual with increasing grid refinement, the reader can draw own conclusions about convergence and the trustworthiness of your results. As a practical guide, a value of less than 1% change in the residual for a doubling of grid resolution appears to me as very close to a stationary (and by inference converged) solution.

Philip

Hi,

The important thing is to see how your solution depends as a function of grid size. So if you select a certain result ( such as VM stress /strain) and check it as a function of grid (mesh) size. Once you have such a plot, you can make decision about how much you can compromise in computation cost (finer grid) for a certain increase in accuracy.

Aritra.

Because the solutions of numerical simulation are only approximations, the analysis of the quality of the mesh and its influence on the results deserve particular attention. Theoretically, the errors related to the mesh must disappear for meshes more and more fine, until asymptotically reaching values independent of the size of the meshes.

Thanks for all the answers...yes I realized that my result is greatly depending on the mesh size since it shows some good trend with the changing of the mesh size. In my case, the finest mesh size (#3mm) gives a result with about 10% difference with previous mesh size (#4mm). So I'm not sure whether this difference is an acceptable range since my computer cannot run further simulation due to very high computation cost.

Please help, thanks.