Many people face problems in getting a stable and converged solution in CFD. Here are some elementary guidelines that I am sharing can help you get a better convergence.
1) Always start with the first order upwind discretization scheme, as this is the most stable scheme.
2) Once the first order upwind scheme solution is converged or the residuals (specially the continuity) are significantly lower <1 x10-3, then change the discretization scheme to second order upwind scheme and use the solution of first order scheme as the initial guessed values.
3) If large fluctuations in the residuals are observed, then change the under relaxation factors. Lowering the under relaxation factors help in converging the solution.
4) Sometimes residuals do not change with the iteration or fluctuate with the iteration about a mean value, in that case, it is better to plot cross-sectional averaged properties of any important variable like velocity, pressure etc to see how that is changing with iterations. If it is not changing appreciably, you can take that as your converged solution.
5) Sometimes the nature of the flow in the geometry is so complex that convergence using steady state approximation is not possible. In that case, transient simulation should be used.