Probably there's a stearic clash or some other overlap between a protein and the membrane that the minimizer cannot resolve or untangle.
Try Desmond to run your your membrane-bound protein MD. They have an automatic membrane-specific system builder facility, so getting a simulation going, including minimization and lipid equilibration, is really easy.
Their ca be many reasons. those mentioned above are first to be looked at. adjusting the lipids is always tricky. you should have a look at your starting structure. if nothing is obvious consider also that membrane systems are usualy large and may simply need more time to get minimized.