26/05/2013 I've spent a bit of time tidying up the meshing in the NSW to remove the mesh warnings. I seem to have succeeded for the wall on its own but with other components in the build mesh warnings may return. It seems that small changes to the mesh appear to have a large effect on the number of warnings received and their location. Before tidying up the mesh none of the affected cells look bad, so I suspect that the warnings are due to a fussy mesher rather than being real issues. (And a pedantic programmer for spending several hours trying to remove them!)
25/05/2013 Committed comi code to Launchpad Repository via Bazaar. Revision 24. (Hall Model 92)
24/05/2013 Today we have upgraded to OPERA 16.
In preparation for OPERA 16 I have been running the model through OPERA 15R3 to find potential volume meshing issues. There are suprisingly few given the trouble we had six months ago which is a positive indicator as to how much the model has developed, and how many bugs have been found. One bug that the 15R3 mesher found was that the Quadrupole air cutout was a little small creating some odd shaped air-blocks. This is non-critical but has been corrected. I have also put an additional cut plane in (after model 91) between the EMR and solenoid Air to help meshing in this region. This will leave just a couple of cells with mesh warnings that I need to take a closer look at again, but having viewed these cells previously there doesn't look like there's anything particularly troubling about them.
20/05/2013 Given that Klaus has been successfully running the hall model on version 16 of the OPERA I'm tempted to wait a couple of days for the release of OPERA 16 before I start giving the solver a full run of hall models to solve.
20/05/2013 I noticed a small geometric error in a cutout of the East Wall which has been corrected. This error will have no effect on the solution, but the correction may help meshing.
20/05/2013 We feel that we have finally gotten to the bottom of the problem with the models not solving. Klaus ran some models with the default OPERA tenten curve and these models have solved much more quickly. Having tried this on my hall model this seems to prevent the hanging seen in the solver. Additionally the hall model appears to solve relatively quickly on my system - 51 hours for a full model!
To recap the problem lies in the BH curve that we were using for the tenten material. Our BH curve has some additional points on it over the OPERA default one, however these points were making mu_r non-monotonic over a range of values. This combined with recent introduction of a finer mesh in parts of the model appears to have triggered these problems in the solver.
Thanks to Holger for putting us on this path!
16/05/2013 I know it's a bit mixed up but see model page for additional information as there has been a lot of models run to try and get to the bottom of the solver problem. Not there yet.
14/05/2013 Model 83 seems to be running ok. As of this morning it is on iteration 5. I will assume that this will run to completion.
Now that I have sorted the model out I am now going to create a simpler model for Klaus to use. This will be model 84. This model will contain, NSW, SSW, Floor Web, Q9-Q7 including bases and baseplate, Solenoids Step IV, Solenoid Mode 240MeV/c, VP & USTOF plates, EMR, Beamdump, Linac Wall, DSA, Walls and finally the floors.
Committed comi code to Launchpad Repository via Bazaar. Revision 23. (Hall Model 83)
12/05/2013 Model 83 Now running - see model notes.10/05/2013 I have found several volume mesh errors in the model. This may have been present in the model for a while, as I have known the modeller to pass models with single meshing errors through to the solver before. This could be easily missed as the mesh creation and database generation is automated. This error has occurred because of the change in thickness of the South Wall when the SSB was built, but I suspect the recent meshing changes have brought it to light. (i.e. it may not have presented as an error previously.) I'm not sure at the moment whether this is the suspect for the failure of model 79....only more tests will tell.
On the subject of the falling over of model 79, if further tests prove inconclusive I shall try changing the SSW material to air. There are some strange features in the mu_r curve of our AS1010 data, and this is suspicious.
08/05/2013
Models 81 and 82 are increments on model 80, once again an attempt to find the cause of the solver error.
For further details of what is running in the model see the models section.
Update: 09/05/2013 Model 81 got past iteration 2. Model 82 is now running.
07/05/2013
On the presumption that model 80 will get past iteration 2
I have prepared another model, (Model 81) using the new meshing code for the NSW and the old meshing code for SSW.
Once again a test to see if it gets past iteration 2 in the solver.
Update: 08/05/2013 Yes it did get past iteration 2. Model 81 is now running.
06/05/2013 The rerun of model 79 seemed to hang on iteration 2 in the solver. I'm now trying to rerun this model (model 80) with the old meshing code for the NSW and SSW. Let's see if this gets past iteration 2.
03/05/2013 I believe that I have now found all of bugs that were causing the problems reported by Klaus. I am now trying to re-run model 79.
02/05/2013 Klaus has reported that there are some air volumes that are set to default potential so they will run at reduced potential in the model. This is a bug as these air volumes need running at total potential. This will be corrected for before I try and rerun model 79.