CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Reverse flow at boundary (

andyross33 August 22, 2013 03:20

Reverse flow at boundary

I am seeing (heavy) reverse flow at my outlet boundary when there shouldn't be. I am modelling one side of a plate heat exchanger, with subcooled water at the inlet which flows vertically past a hot isothermal surface (which will eventually be simulated explicitly as the other side of the heat exchanger). I am including homogeneous phase change with IAPWS water/vapor definitions as I want to see how much vapor is formed.

The inlet condition is defined as 10C at a prescribed mass flow rate. The outlet is defined as an Opening at an opening pressure of -2 psig and flow direction set to Normal to Boundary. I have hand calculated the expected velocity through the domain and set that as the initial condition.

I monitor the mass flow rate at the outlet during the solve and it steadily climbs but in the positive direction (flow into the domain). I have tried various combinations of boundary conditions including Outlet and reversing the mass flow definition at the outlet vs the inlet but no luck. When I used the Outlet BC it placed a wall over 100% of the outlet.

This is not reverse flow caused by turbulence, as far as I can tell, because my flow is low speed and laminar with no geometry that would impart turbulence (flow past a vertical plate).

The weird thing to me is that I have modelled the other side of the heat exchanger separately, where I have superheated steam passing its heat to the wall and the solver has no issues. I do have to babysit it to ensure convergence but at least the flow seems correct.

I know phase change simulation can be tricky but this seems so elementary to me I'm kicking myself for not finding the culprit yet and I have been at it for several weeks now. Any thoughts anyone?


ghorrocks August 22, 2013 19:37

Can you post an image of the flow you are getting?

andyross33 August 27, 2013 13:24


I decided to take your advice that I've seen you post on many other threads and re-read the CFX convergence documentation. I realized that my physical timescale was way too low. The residence time within my domain is about 3600 s and I'd been using a timescale of about 1e-3. As soon as I had bumped this up to within the range of the average residence time, things worked a lot better. The flow was no longer reversed and the model was far more stable.

Thanks for consistently hammering this point home, eventually people like me will get it :)


All times are GMT -4. The time now is 12:34.