3D modeling with tomography data

6 years 11 months ago #60 by osilva
Hi Florian and Vanessa,

What do you mean by "VERY VERY long time"? Note that the fact of only calling to the PhreeqC libraries should increase the simulation time respect to the Comsol simulation by at least one order of magnitude (approximately). We have run some iCP simulations (transient flow and reactive transport considering only equilibrium reactions) with 100,000 elements that took about 1 week. The simulation time could get even higher under situations of abrupt concentration gradients.

I didn't find any attached file with your post. Please, send us the json and Comsol input files by email. We will go into your problem to see what is happening and come back to you ASAP.

All the best and happy Xmas

Please Log in or Create an account to join the conversation.

6 years 11 months ago #61 by Flo
Hey Orlando,
thanks for your quick answer!
The quite unprecise statement "VERY VERY long time" means around 1000h (roughly estimated; the first 3 time steps of 1000 in total took 3h). the simualtion may speed up with time, but we were just surprised.
We attached the json file but obviously it did not show up...sorry for that!

I will send the files tomorrow again!

all the best,

Please Log in or Create an account to join the conversation.

6 years 10 months ago #79 by Alvaro Sainz
Hi Florian and Vanessa,

What made the simulation in iCP so slow was the structure of the Comsol input file. Each physic module was solved in a different “Study”. As the physics are linked it is more efficient to solve all the physics with different Solvers but in the same Study.

In your case the problem was that the whole flow problem was recalculated each time step to provide the fluid velocity to the solute transport, instead of using the information of the flow already computed.

Furthermore, we recommend to have a sequential structure where the first solver estimates the flow, the second the transport, and the last one save the results (even when it’s not required for this simple chemistry). In this way the information needed for each physical module it’s already solved and saved by Comsol, making the transfer of information more clear and fluent.

Another way to reduce the computational time are the settings of the chemical step: number of Threads, Libraries and Slides. Jorge has posted about it here .

We will send you back the .json and comsol file with the new structure. With this configuration the computational time for each step has been reduced from 50 to 5 min in our 8 cores machine.

Happy new year!
Alvaro & co.

Please Log in or Create an account to join the conversation.

Time to create page: 0.184 seconds
Powered by Kunena Forum