Centrifugal pump convergence issue
 

Dear All,
I am doing a centrifugal pump simulation in CFX general mode ( not using the turbo module ).
My B.C's are Total Pressure at inlet ( Atmospheric ) and mass flow outlet and the flow in incompressible.
I am facing this in my simulation. Initially the residuals drop rapidly without any issues and my monitor of total pressure rise across the pump gradually converges to a value with almost a flat curve. But after some 300 iterations or so the residual and pressure rise monitor start fluctuating with the amplitude increasing as the simulation progresses. Shown below are the pics showing various residuals and the total pressure rise monitor. My pressure rise fluctions are of the order of 4-5% of the mean value. Also what I observe is that the frequency of fluctuations is exactly the same in all the monitors and residuals and this value is close to half the frquency of my rotor revolution, are these related ?
Can someone tell me what might the possible reasons be ? and HOW to GET Rid of these fluctuations

Thanks in Advance,
Dinesh

Hi,
I tried playing around with the time scale to get the convergence. In the simulation shown above I had given a time scale of 1/2w = 0.0015, then after the post I changed it to 0.003(1/w) and got better convergence, then I reduced it further to 0.005 and got huge drop in the residuals.
The CFX documentation says 1/w is a good estimate for the physical time scale. My doubt is what are the criteria based on which we should decide the physical time scale and most of all WHAT ARE THE LIMITS OF PHYSICAL TIME SCALE THAT WE SHOULD NOT CROSS ???

Thanks in Advance,
Dinesh

There are no general limits on timescale. For a steady state flow the time scale you use should be irrelevant - the flow is the same in 1 nanosecond and 1 year.

But practically, larger timesteps mean that the turbulence modelling works better. If the timesteps are in the range of the turbulence timescale then the turbulence modelling does not work well and you get convergence problems. Bigger time steps fix this problem - as you found in your case.

Thanks for the clarification.

But sir, what is the relevance of physical timescale in steady analysis?

The physical time scale is only used as a way of advancing towards convergence. If you take big time steps you progress faster but risk divergence, if you take small time steps you progress slower but are more numerically stable. Once you have a converged solution the time step makes no difference as all the time derivatives are zero.