Stuck? 
support@mecway.com
support@mecway.com
Mecway
FEA
Convergence by a trick.
Hi Victor,
In structural static analysis (any dimension) probably for any type of analysis would be quite useful to do a trick I have seen in Ansys 16+. If we get a model and we analyse it, finding high gradient of any values, we could use New Local Refinement for two purposes (if possible).
1. Select a radius to modify mesh in high gradient area (exists but just for imported geometries, could be useful with FE Model created manually).
2. Select a radius to isolate such area (multiple areas) where we would read boundary nodes values (from first simulation) as an input to those isolated patches of FE Model and analyse just the patches. That would massively reduce analysis time and reduce computational cost instead of rerunning the whole model? All meshing would still be a manual job and selection of the patches but e.g model of 100 000 DOF would be reduced to 500 maybe (depending on the model) and if result would still be unsatisfactory we could triple the mesh density and run model of 1500 DOF instead 101 500 DOF.
It just sounds like half way through of automatic convergence with pseudo adaptive meshing but in my opinion would be much quicker as after first run of relatively well defined model we could just focus on the questionable areas.
Am I dreaming or this could be possible?
Regards,
apadzak
In structural static analysis (any dimension) probably for any type of analysis would be quite useful to do a trick I have seen in Ansys 16+. If we get a model and we analyse it, finding high gradient of any values, we could use New Local Refinement for two purposes (if possible).
1. Select a radius to modify mesh in high gradient area (exists but just for imported geometries, could be useful with FE Model created manually).
2. Select a radius to isolate such area (multiple areas) where we would read boundary nodes values (from first simulation) as an input to those isolated patches of FE Model and analyse just the patches. That would massively reduce analysis time and reduce computational cost instead of rerunning the whole model? All meshing would still be a manual job and selection of the patches but e.g model of 100 000 DOF would be reduced to 500 maybe (depending on the model) and if result would still be unsatisfactory we could triple the mesh density and run model of 1500 DOF instead 101 500 DOF.
It just sounds like half way through of automatic convergence with pseudo adaptive meshing but in my opinion would be much quicker as after first run of relatively well defined model we could just focus on the questionable areas.
Am I dreaming or this could be possible?
Regards,
apadzak
Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Comments
2. That sounds like substructuring. Not something I give a high priority to since it's an optimization that could complicated to use.
1. Sounds like a good option.