Learning Forum

[sardar.svn]

(Appologies, thinking I had posted on wrong forum, I flagged my own question, so re-posting here)

Good day,

I am solving mixing of high volume fraction of particles in water (transient, Eulerian, DDPM, kwsst, Coupled, injection). The issue is that simulation solves and converges (as per both residuals and monitored variables) for the first flow timestep, but when the flow timestep is bumping to the second one, continuity and k, e residuals fly up to 10 orders of magnitude, and then there is this Experiencing convergence difficulties warning before I get floating point exception and simulation is terminated.

My background effort to fix the issue has been the following, however, I have tried them one at a time for most of them to see individual effect each could have had, but unfortunately in vain:

• Solving the flow to a stable state before injecting
• refining the mesh ( Watertight/ Polyhedral or Polyhedral Hex Core - Orth: min. 0.3 - avg 0.95  |  Skew: max 0.7 avg 0.04  |  AspR: Max 19.9 - avg 1.75),
• solving for timesteps as small as 0.002 [s],
• compromising on injections as small as parcel count of 2 (while keeping DDPM active),
• manipulating URF's of momentum, pressure, k and e for as low as 0.2 (pressure) and 0.5 for the rest, as the manulal says
• increasing number of DPM updates per flow timestep and reducing DPM source URF to make sure DPM sources are fully consumed in each timestep.

Since tank top is open in reality, I define zero-shear wall and escape as boundary condition for top open surface and DPM, which I believe should not be a deal. I still have not fumbled with Flow Courant Number and the DPM tracking length scale.

But there is nothing else I can either find online or think of on my own.

Can anyone help please?

[Prashanth]

Hello,

What is the parcel volume? Is it smaller than the grid size?

[sardar.svn]

sorry for my delay

Using volume injection in tandem with DDPM, ensures that the smallest cell in the entire domain contains five parcels.

[Prashanth]

It is difficult to point to a solution without taking a look at the case. Most of the times, excessive overlap of particles causes the issues. Also, how are you modeling partice-particle interactions? If needed, use a separate particle time step.

[sardar.svn]

Thanks for following up.

I suspect my initialization being incorrect, I use hybrid, and would like to have your openion on that too please.

But apart from that, I am using no DEM, because, despite keeping DDPM on, I am using low particle VF, just to keep on the safe side. (Or is it low VF of DDPM that leads the model to divergence?). And I am already using separate timestep for particles. :*(

[Prashanth]

Its recommended to use standard initialization practices and set reasonable initial conditions. I would also uncouple the particles from the flow to troubleshoot the setup.

[sardar.svn]

 

Uncoupling, sounds good. thank you.

As for standard initialization, would it be reasonable to give inputs based on a few first timesteps solved via DPM instead of DDPM? This way, provided that divergence is due to bad initialization, I might come closer to a better start.