I am working on the turbulence simulation. The scaled residuals of x, y & z velocity before going upward 1 e-5. The solution seems not to be converging. I am using K-E turbulence model to capture turbulence characteristics. Same issue is occurred in RSM model.
I used under relaxation factors as follows.
1. mass flow rate = 0.5
2. Pressure = 0.1
3. Density = 1
4. Body Forces = 1
5. Momentum= 0.5
6. Turbulent Kinetic Energy = 0.5
7. Turbulent Dissipation Rate = 0.5
8. Turbulent viscosity = 1
SIMPLE scheme is used for pressure coupling velocity, with least square cell based gradient , standard for pressure and all discretization are first order upwind . I hereby attached the sample picture of residual plots.
Kindly suggest what are the possible issues.
Dear Muhammad, convergence is a complex thing. In my opinion, every problems has a different condition to considered it has converged. I use normally reduction of four or five orders of magnitude for residuals, with exception for energy. For energy I use reduction of scaled residual up to E10-8. However a good option is to observe the behavior of several parameters against the iterations. may be, temperature, pressure, heat transferred, friction factor or any other. When the relevant magnitudes are keeping constant and goals of scaled residuals has been get I decide to stop simulation.
You could try
1- improve mesh quality
2- use different turbulent model
3- increase number of iterations
Dear Muhammad,
Is your task transient or stationary? In some cases you cannot reach convergence once the phisics of the task is not properlly defined.
If you solve transient task pay attention to time scale as some of turbulent structures can casue flow disturbances not coherent with time step you have defined.
All the best!
Damian
Goodmorning,
after verification of the mesh and the model of turbulence, you can decrease in the factors of Under-relaxation for to have a convergence.
Dear Muhammad,
1-First you specify your problem and geometry you working on
2-Try to simplify the problem to 2-D or even 1-D if you do not have analytical solution or computational treatment solution or experimental measurements.
3- Try different grid sizes starting with coarse mesh to fine .
4- Convergence will be accepted if the numerical solution approaches to one of the above results starting from trend of 1-D or 2-D or analytical solution and the most proper acceptance if the predicted results are close to experimental data .
Dear Muhammad, which Fluent release are you using? Try to use realizable k-eps and "coupled" scheme for pressure-velocity coupling...
If your flow is steady, try to use "pseudo-transient".
Are you using 2nd order spatial discretization?
You can also try to solve the flow first and then add the turbulence equation.
Dear Muhammad,
Convergence depends upon various factors. It also depends on how much convergence is required/ sufficient for your type of problem.
Straight away going to the Under relaxation factors is not the right methodology.
Follow these steps, if required
1. Check the mesh quality, improve/ allign according to the expected flow pattern.
2. Review Boundary conditions. Check for the required wall functions
3. Better initialization, find initial solution using 1st order discretisation, and ultimately having a stricter criteria. Adoption of changing of Courant number( monitor the solution throughout)
4. If nothing works, then opt for changing of URF
First of all improve the quality of the mesh and than run it with different turbulence models
- Badges
- Science topic