View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0001634 | OpenFOAM | Bug | public | 2015-03-24 16:39 | 2015-03-24 21:34 |
| Reporter | Assigned To | henry | |||
| Priority | normal | Severity | minor | Reproducibility | always |
| Status | closed | Resolution | no change required | ||
| Platform | GNU/Linux | OS | Ubuntu | OS Version | 14.10 |
| Summary | 0001634: MPPIC packing models: Missing term in denominator | ||||
| Description | In both the explicit(.C) and implicit(.C) packing models the denominator is missing a factor deltaT*F.Sp(). In explicit.C the code is included, but commented out. In the implicit model it seems the factor is missing altogether, it is also not added later when the corrected velocity is applied to the parcel. The missing factor is given here: P. J. O’Rourke et al., Chemical Engineering Science 64:1784-1797, 2009, equation 39. | ||||
| Tags | No tags attached. | ||||
|
|
How does the behavior of the model and the results changes when you add the factor you say is missing? |
|
|
Most likely the impact is quite small when the time-step is sufficiently small, but one could say the same for the denominator of the momentum transfer term, and in that case it is implemented. However, I am checking consistency with the papers, in the header file of explicit.C the 3D mp-pic paper is explicitly mentioned as reference. It seemed to me though there is a reasoning behind the commented-out code in explicit.C (in implicit.C the code is missing). |
|
|
Please run the test-cases with any changes you propose to ensure the stability and results are not compromised. |
|
|
Since I'm still learning OpenFOAM I will not able to do this soon, also I don't know what your criteria are for giving an OK when rerunning the tutorials, other than it looks OK physically in ParaView. My point was simply to point out an inconsistency in the formulas. Thereby I would like to point out that the tutorials are quite questionable anyway for MP-PIC (again, I don't know what your criteria are for judging them as OK). For instance, when running the column tutorial for MP-PIC for t>3 seconds you clearly see the particles stacking in the lower cells. See for instance this forum thread: http://www.cfd-online.com/Forums/openfoam-solving/150345-mppicfoam-column-tutorial-incorrect-bottom-patch-handling.html |
|
|
That term is designed to make the calculation of the packing force more implicit. There are a large number of stability games to play with MPPIC. The referred implementation (Barracuda) uses interpolation procedures which don't generalise to a polyhedral mesh. That's all we know; the implementation is closed-source. We (and the developers of MFIX) found that the algorithm's stability was dependent on these interpolation methods. We had to modify the algorithm to make it stable without the benefit of structured-mesh interpolation. If you have a look at MPPICCloud.C and MPPICParcel.C, you will see that the movement of the particles itself is predicted with the usual force terms, and then corrected with the predictions of the collision models. This was found to be the most stable way of doing the time step. It was also assumed that the prediction stage already provided suitably implicit particle properties from which to calculate the packing force. The comment is a leftover from testing, and I recall it making no difference whatsoever. The packing force is horribly non-linear, and tends to be either completely negligible or totally dominant. Calculating the drag, gravity, etc ... force terms again seems like a needless expense. |
|
|
Regarding the second point, MPPIC makes no guarantees as to the distribution of particles within a given cell. The sub-cell particle distribution at close pack is a function of the interpolation methods used to calculate the packing force (and other sub models). The interpolation procedures in OpenFOAM were devised without a reference; they are crude and can generate numerical artefacts like the ones in the column case. If you can provide a reference to better methods, that would be great, but I am not aware of any that have been published for polyhedral meshes. |
|
|
To demonstrate the suitability of any changes you propose it would be best to run test-cases for which there are experimental results and compare with those e.g. the Goldschmidt case. |
|
|
Dear Will, Thank you for the explanation. Indeed I already found what you told me, namely the split approach of usual forces and collision models. I found myself that for close pack the explicit approach for particle stress is too crude and provides little shockwaves propagating, the implicit model works better in that case. For non-packed situations (i.e. fluidized beds) I guess the explicit method is sufficient as local overpacking will not be such a big problem. Regarding your comment about the sub-cell distributions, the MP-PIC authors also talk about the issue in their 2D paper, where particles tend to form vertical bands in steady state solutions. But of course MP-PIC is not really built for steady states like a packed bed. As you mention packing force is highly non-linear, so indeed I would say the extra term in the denominator is not very relevant, so in that sense I guess I would be okay to close the issue, as neither the explicit nor the implicit model include this term currently. If I manage to figure out the code a bit more I will do a comparison in the future with experimental results. My apologies for my high bug submit rate, but it seems not many users have actively looked into MP-PIC yet, judging by the limited number of posts on the forum :). |
|
|
Dear Henry, (slightly off-topic, sorry for that) "To demonstrate the suitability of any changes you propose it would be best to run test-cases for which there are experimental results and compare with those e.g. the Goldschmidt case." I am wondering though, in the Goldschmidt tutorial no pressure gradient force is included, although it was agreed earlier that there should be, and definitely this impact is not small; how can this have matched with experimental results? |
|
|
Feel free to test this or any other cases with or without the pressure gradient force and compare the results. Have you run the Goldschmidt with the pressure gradient force? How different were the results? Were they more accurate with the pressure gradient force? |
|
|
I will do this in the future :) |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2015-03-24 16:39 |
|
New Issue | |
| 2015-03-24 16:59 | henry | Note Added: 0004469 | |
| 2015-03-24 17:19 |
|
Note Added: 0004470 | |
| 2015-03-24 17:32 | henry | Note Added: 0004471 | |
| 2015-03-24 17:58 |
|
Note Added: 0004472 | |
| 2015-03-24 18:02 | will | Note Added: 0004473 | |
| 2015-03-24 18:09 | will | Note Added: 0004474 | |
| 2015-03-24 18:16 | henry | Note Added: 0004475 | |
| 2015-03-24 18:30 |
|
Note Added: 0004476 | |
| 2015-03-24 18:39 |
|
Note Added: 0004477 | |
| 2015-03-24 19:03 | henry | Note Added: 0004478 | |
| 2015-03-24 19:10 |
|
Note Added: 0004479 | |
| 2015-03-24 19:13 | henry | Status | new => closed |
| 2015-03-24 19:13 | henry | Assigned To | => henry |
| 2015-03-24 19:13 | henry | Resolution | open => no change required |
