Please wait...
×

Error

Posted Tue, 11 Jun 2013 15:04:03 GMT by Stephan Löbmann Student
Hello,
I am using FEFLOW for my diploma thesis to implement heat transport processes during a geothermal use. For evaluation of the geothermal use I used a ground source heat pump. I need an optimized computational model since I will perform many simulations due to a parameter analysis. Now I am having the problem that one simulation with a ground source heat pump needs more than 8 hours, but for a simulation without ground source heat pump FEFLOW needs just 0.5 hours.

Is there perhaps a way in the settings to reduce the simulation time significantly for such cases?

Following my scenario description:
• Generated mesh with 1000 total elements (Polygon graduation 5 und point element graduation 10)
• Use RICHARDS-equation in unsaturated zone
• Flow velocity: 1,89 m/s
• Final simulation time: 365 d, starting at 0 d

Borehole heat exchanger properties:
• Inlet temperature: 5 ° C
• Flow rate: 30 m³ / d
• Length of the heat exchanger over the entire thickness of ca. 40 m
• BHE Geometry: Double U-shape

Thanks for your help!
Best
Stephan
Posted Tue, 11 Jun 2013 22:17:44 GMT by adacovsk
Wow, 8 hours. I thought my models that take weeks to run on overclocked CPUs were long. :-).

Anyway, to improve performance, consider using multiple computers if you're doing multiple simulations (that would mean more licenses = $$$); overclock your computer; reduce the amount of elements (or at least maintain the resolution closer to your important areas) whereby using triangle.dll for mesh generation is the best option usually; reduce the size of your domain (reduces the amount of elements); and take larger timesteps (pay attention to your Peclet number); avoid unsaturated problems if you can (these tend to take far longer than fully saturated problems); and potentially use the 'SAMG' solver for the solver type, if you have a multi-core processor and cooling it isn't a problem and you're running a steady state.

One important thing to note is that the more time you spend designing your supermesh, the better resolution you'll get around your areas of interest and the less resolution will be necessary in far-field areas. I typically try to minimize the amount of elements to the maximum while using Triangle.dll while minimizing delaunay triangle violations and the number obtuse triangles (which can typically be fixed after you generate your mesh if it isn't too refined around your areas of interest).

Adam
Posted Thu, 04 Jul 2013 07:30:34 GMT by Denim Umeshkumar Anajwala
With respect to BHE simulation, you can think about using the semi-analytical Eskilsson-Claesson approach rather than the fully numerical Al-Khoury method.

Set the iteration number per time step to 1, but make up for it with a much stricter error criterion (maximum error norm, small value) than the default one to force a short time step. Additional iterations tend to not improve the solution within the BHE, thus shorter time steps with just one iteration are a better approach most of the times.

You must be signed in to post in this forum.