CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   convergence (

raj calay March 25, 1999 09:31

I am solving a 2-d problem of flow in an enclosure representing flow in a building. there are 2 inlets located on l.h.s wall; inlet 1 is close to the bottom and inlet 2 is close to the top wall. There is a temperature difference of 50deg between i and 2 being T2>T1 to generate buoyancy-driven flow in Y direction. Also U2>U1. There is an outlet on the wall opposite (r.h.s)located close to the top wall. Inlet 2 is slightly higher than the outlet. All other surface are set to be stress free. This is a case of flow driven by momentum and buoyancy. So N-S, continuity and energy equations are to be solved. I am using seggregated solver. Even after 100 iterations solution is not converging Convergence history shows solution to become oscillatory quite above the set convergence criteria. Initially the Rayleigh number was set at 1000 (very low). Still the steady state converged solution was not obtained. Relaxation method is hybrid and upwinding is streamline. what is the strategy to obtain s.s. solution?

Joern Beilke March 26, 1999 09:35

Re: convergence
If you just want a converged solution then use a very coarse mesh :)

Otherwise you should think about the physics in your enclosure. The boyancy driven flow in a building with dT=50 Kelvin is not 2d and also not steady state.

So even if you get a converged solution it might be shit at all.

raj calay March 29, 1999 06:22

Re: convergence
The flow is forced + free convection. I understand the equations are highly non-linear which would result in non-convergence regardless of what method I choose to solve. I got a converged solution for Ra upto 5000 but that is no good. because in reality at least Ra is 10E11. I also know that situation is in fact 3-d. But if 2-d s.s. solution is difficult to get, how do you suggest that solving for 3-d will be easy. you 're saying 2-d solution would be shit. But is this is how we solve real problems, simplify complicated system into simple 2-d or sometimes 1-d system by choosing suitable assumptions (both experimentally and numerically)

A converged solution would be atleast numerically correct (to certain limit) solutions of the N-S equation, energy equation and continuity equation (so far accepted physics by everyone to describe flow field).

What will be your strategy? Because this is real problem to solve not an academic exercise. By the way I have tried with several mesh size.

Joern Beilke March 29, 1999 08:34

Re: convergence
It is in many cases much easier to get converged results for 3d cases using the (nearly) real geometry.

If you want to simplify a complicated geometry, try a reduced 3d geometry.

For a building with inlets and outlets you will not be able to satisfy the continuity and momentum equation at the same time if you go from 3d to 2d.

I would start with a relatively coarse mesh in 3d to get some insight in the flow. Then you should know how to improve all the stuff.

Rasputin March 29, 1999 09:22

Re: convergence
You should also monitor the variation of key variables at key points to determine if these points have attained steady state.

It is often the case that the errors are located far from regions of interest and do not effect the points of concern. If these monitored points show no variable variation with succesive iterations then these points will have achieved steady state as far as possible with your current BCs.

raj calay March 29, 1999 11:21

Re: convergence
thank you for your suggestions for monitioring variables at key points. I've started working on 3d model now.

All times are GMT -4. The time now is 22:09.