Let us go back to my first reply. Is there a reason why you are running this in transient? Second your initial condition is not great, that is why the first timesteps are not converged. You should definately not consider these in your final analysis. One thing to do would be to increase the maximum number of iterations per timestep. But a better approach would be to use a steady-state solution as the initial condition.

On your question whether the residual curve is acceptable, that is just one data point to determine convergence. Simply looking at the residuals is not enough to determine if the solution is good enough. You should have other monitors that will help you determine if the solution is "good enough". Without knowing your problem it is difficult for any of us to recommend what monitor points you should include. That is why I asked above why are you running it as a transient solution. Usually the first step is to solve the steady solution.