FieldFromDumpFile

[rwendner]

By now we have a number of options to dump SVs and fields of the mechanical as well as the diffusion problem.
Thank you!

At the moment all of our formulations for creep, aging, etc. are based on one-way coupling.
During calibration but also parameter studies the HTC problem typically does not change. Yet, we have to run the full analysis every time.

I suggest to use the dumpfiles of the multiphysics problem in a similar way as the interp.dat in the past.
We simply need a FacetFieldFromDumpfile and a ParticleFieldFromDumpfile to significantly accelerate many of our simulations.

Mohammed and Giovanni are in favour of this feature request.

[rwendner]

Can you please comment ?

[zhouxinwei]

Can you please quantify the time spent in the diffusion problem compared to the time spent in the mechanical problem?

[rwendner]

This is hard to tell and depends on the problem at hand.
In the case of large scale drying specimen (e.g. beam, slab) with the required mesh refinement the cost of a "good" discretisation of the diffusion problem is significant, especially since the diffusion problem is limited to a single threat.

Many of our investigations (creep, shrinkage, aging,...) require many runs (dozens) of the mechanical problem for calibration and parameter study without the need to rerun the HTC part.
Multiplied with the number of runs even small time savings matter.

In our opinion (myself, Mohammed, Giovanni) , this is worth the effort to save and reimport fields.

Furthermore, the ability to reload any SV field opens many opportunities for future research.

[zhouxinwei]

Even for a fine mesh, the mechanical part will takes more time as well. We should really look at the time ratio spent on the diffusion problem vs the mechanical problem. From what I understand, the diffusion problem is solved every so often, most of the time is in the mechanical part. If in a simulation, the diffusion solution takes 5% of the total time, there really isn't much to gain.

[alnagm]

Dear all, i did check the computational cost, for example, if the loading time is late (182 days for Bryant creep simulations) we needed 40 ms to solve the first 182 days then 30 more for the rest of the 500 days period this means we have to run the htc for 40/70=57% !!!! of the run. The time used in HTC is not only the time consumed while both HTC and mechanical parts are both active which is true is a small fraction, the main issue is the lead time we have to spend before loading which is at least for usual tests is 28 days. For Bryant at 28 days, i spend 25 ms for the first 28 days then 45 for the remaining 500 days which is 25/70= 35.7% of simulation time that can be eliminated just by loading them directly.

So, simply, since we have to run with small time steps in the beginning of the HTC solution, this is an additional cost during which we never apply the mechanical loading. Thus, if we can get this saved and uploaded, the mechanical part will run after only 0.1 ms, because we can upload the whole 182 days in that 0.1 ms because no forces are applied so kinetic energy will be zero. But if we don't separate, the 182 days will force us to run for 40 ms just to compute the HTC conditions at time 182 days from casting. Then start calibrating.

So, i strongly agree it is a very important time saving feature

Thanks,