I dont think that is a valid thing to do. First, you need to know what sampling converges you are looking for (i.e., how do you define convergence); and second, I think it is more important to investigate why a model does not converge (or output divergence samples), an automatic process doesnt seems to be the answer here.