[Davitt]

Hi everyone,

I'm using COMSOL and I'm running a CFD analysis of a wind turbine. I did a 2D model of it and ran the simulation and it worked out well.

Now I'm trying a 3D model with alot more DOFs and the solution is super slow. I actually don't mind that the time-steps take longer but the real issue is that the size of the time-steps are extremely small.

For example: a normal time interval would be around 0.15s but the time step I'm getting is around 0.00015 or less. and it gets even smaller as the fluid flow magnitude increases.

Also, I'm running the simulation on a i7-4930K, 3.4GHz hexacore processor with multithreading, 32GB of RAM, 64 bit OS...but for some reason COMSOL only uses about 3GB of RAM when running the analysis.

Does anyone using COMSOL know how to redress this situation?

Any help will be greatly appreciated..

Thank you,
Davitt

[mprinkey]

Originally Posted by Davitt

but for some reason COMSOL only uses about 3GB of RAM when running the analysis.

I haven't used COMSOL, but there is no reason that a simulation will use more memory than it needs to just store a few timesteps worth of DOFs and whatever intermediate working sets (sparse linear equation matrices, Krylov space vectors, etc.) it needs. In most CFD codes I've used and developed, the memory working size is roughly proportional to the computational cell count, or equivalently, the DOF count. In FLUENT, it was roughly 1 kB per cell, at least for the segregated single phase solver. If your have a few million elements in your model, then you are probably using about the right amount of memory.

Reasons for small timesteps can be numerous. Again, I am not familiar with the algorithm that COMSOL is using, but if you are setting a CFL number and the algorithm is adapting the timestep to meet that criteria, then you likely have one or more small cells that are seeing very high velocities and the code is responding by cutting the timestep. Also, if you are doing FSI or even just moving mesh, there may be mesh motion limits that are making the timestep so small--maybe because an element would become too skewed or even negative volume if a larger time step was chosen. Also, it maybe depend on the turbulence model that you are using and the mesh near the blades, etc.

Sorry I don't have more specific recommendations, but I would say to check your grid. You may not have a worse 3D grid than your did in 2D, but 2D simulations can be more forgiving about such things. The other thing to consider is what the real characteristic times of your physical system are--it is possible that there timescales in the 3D system that are not there or are much larger in the 2D system and didn't surface before. If you find that there are critical physics happening on the scale of your current small timestep, you have to accept that and wait.

[Davitt]

Thanks for very prompt and informative reply mprinkey...you've given me some very important points to consider. I'm going to look into them and make some modifications. I'll let you know what happens.

Thanks again
Davitt

[kaya]

Hi

I am not sure what is a DOF in COMSOL. Is it the DOF of the turbine structure? I don't think so but thats what just popped out when I googled. If thats the case, your time step can't be longer than the half period of the finest structural mode you're resolving, which are usually at very high frequencies.

If you're just running stiff rotating rotor computations, in case you have a very fine grid, your time step would be dependent on that. Imagine a constant CFL condition, if you double the mesh you'd have to half the time step to keep the same CFL.

Ah also for a wind turbine operating at tip speed ratio = 8, you get about 70m/s tip speed at an average inflow velocity and this is quite bit of a movement around that super small sized meshes. I am not sure which control parameters you have there but maybe you can try reducing the rotor speed somehow or coarsen the mesh and see.

[Davitt]

Thank you for your input Kaya...

I believe DOF refers to the degrees of freedom of every finite element in the analysis domain. So it should refer to DOF of turbine structure as well as the surrounding fluid, ie. all the domains of the study.

Also, what do you mean by 'the half period of the finest structural mode'? The turbine is actually a savonius turbine with a tip speed ration of about 0.8 so it does not achieve very high speeds so I believe the frequency may be low.

Thank you for your suggestions.
Davitt

[kaya]

imagine a ruler at the edge of a table, when you hit it it starts vibrating up and down and what you're looking is the first flapwise mode. same goes with wind turbine blades. say the frequency of the movement is 1Hz, to be able to capture the movement you need a sampling dt max 0.5 sec.

Although I have no idea about strucctural modes of vertical axis turbines, you said your dt is 0.00015 i don't believe that you have a structural mode at the range of thousands of hertz... so perhaps your dominating dt limitation is from the flow solver but I think you should shot down the structural solver first and play with your mesh.

[Davitt]

Hey guys...so I revisited my mesh and made it a bit course. Now when I start the calculations the time step starts ok then reduces and reduces to the point where it's about 0.00001s and now it just stays at one time step or it's probably getting even smaller and the increment is too small to measure.
There's only 122,000 DOFs...can the quality of the mesh cause the small time steps?

[Jehosh]

I have found with a very coarse mesh, time steps need to be very small to converge. On the other hand, with a very fine mesh it becomes very expensive to run all the calculations. There is a bit of a sweet spot in the middle. Part of the CFD art.