Transformer thermal modelling

[marts]

Hello to everybody in the user community

I am trying to solve an oil-immersed transformer heating problem using ElmerGUI.

Here is the description and some snapshots of my geometry:

Bodies 2108, 2109, 2110 and 2107 are the four iron cores, with presribed loss density (body force "iron loss")
Bodies 2101-2106, 8000 and 9100-9400 are the High Voltage and Low Voltage copper windings with prescribed loss density (body force "HV loss" and "LV loss")

Body 5764 is the steel corrugated panel tank, modelled as a solid body of small thickness (5 mm)

Body 9500 is the oil between the cores and windings and the transformer tank (a body force "boussinesq" is applied to this body)

Body 5792 is the air box surrounding the transformer geometry

I am attaching the geometry of the active part, the tank and the surrounding air.

geometry.jpg

(64.47 KiB) Not downloaded yet

All interfaces between the cores, windings and the oil are prescribed with boundary condition "active part" imposing normal velocity = 0, as you can see from the .sif file

The same boundary condition is applied in the interfaces between the tank walls and the oil (inner tank) and the surrounding air (outer tank) and the walls of the air box (outer air)

I am using coupling of Heat and Navier-Stokes equations (Heat solver only for the active part and tank bodies and Heat-Navier Stokes for the oil and air)

I am using transient simulation with 20 timesteps of size equal to 2.

I only set an initial condition for the temperature, which is set equal to the ambient temperature.

case.sif

(9.57 KiB) Downloaded 376 times

However, there are convergence problems in the Navier-Stokes solver after step 2 and I cannot get a solution

If I model a similar problem in a simpler geometry (i.e. an iron cube of similar loss density inside an oil box) I get a good solution without convergence problems

Can you provide some hint on how I could ensure convergence? I have limited the non-linear system iterations for both solvers but this does not seem to solve the convergence problem.

I am also attaching the sif file The mesh has been produced by GMSH software. (I am not attaching the elmer files, since they are too big to be uploaded, but I can upload them partially in another post if someone feels they could test them)

Any help would be very much appreciated! Many thanks in advance

[mzenker]

Hi,

what kind of convergence problems do you have?
It may help to
* increase the "Linear System Max Iterations" in the solver section (if you have a message saying something like "ERROR: Failed convergence tolerances"),
* decrease the "Nonlinear System Max Iterations" in the solver sections (even down to 1) AND increase the "Steady State Max Iterations" in the simulation section at the same time,
* decrease the timestep size.

HTH,

Matthias

[marts]

Mathias, many thanks for your quick answer

I have tried all the above

unfortunately convergence problems start to appear at time=2 (no matter how small the timestep sizes, I have reduced them significantly up to 1/10 of the original size) still appear. I get errors of e+02 and above in the Navier-stokes solver iterations after that time, and consequently divergence in the next time steps. I have increased the linear system iterations to 2000 and the steady state iterations to 20 but that does not seem to improve the situation

I think something might be wrong with the problem definition, but I cannot locate what that is...perhaps the complexity of geometry, or something like that?

thanks again for your help

[mzenker]

Hi,

I have no experience with Navier-Stokes simulations. so I can just guess.
It might be necessary to reduce the time steps even further, it depends on how fast your processes are. If you still have divergence, it might also be necessary to make your mesh finer.

HTH,

Matthias

[marts]

Hello again to all elmer developers and users

After several efforts, it seems that by changing the Navier-Stokes solver from BiCGStab to TFMQR I get an improvement in convergence...I can run multiple steps and get reasonable results for the problem. I think that in complex geometries it might be useful to swtich to a less fast fluid-flow solver but with more stable results.

I thought it would be useful to share this information with other Elmer users that might encounter similar problems.