SimScale CAE Forum

Residential Convection Fail


Please see the attached .jpg with a screenshot of my project.

I’m experimenting with my students trying to do a thermal convection simulation based on Milad Mafi’s project in the public folder.

Here’s mine:-

I’ve made the internal volume of a house in Fusion 360 and it imported easily into Simscale. The first simulation had all the walls of the house at 273 degrees kelvin and the heater in the living room is a box with its walls at 773K (like a wood heater without a fan). This gave us some results however I was unable to get tracer particles to show.

To try and show tracer particles I thought I’d change the simulation (see the second one) so that there is a velocity inlet being one of the sides of the heater and an velocity outlet, one of the windows at the back of the house. The simulation failed three times and I tried a couple of things to see if it was about my boundary conditions having the wrong directions or something.

The solver log mentions a problem with mass inflow and mass outflow. Do I need to have these perfectly balanced?

Would you please take a look and see if there’s something that could be changed to make this work?


Thanks for your detailed post @arnold1 !

It might take a while until I can jump into this - @vgon_alves & @Get_Barried, do you mind having a quick look at Arnold’s project and see if there is anything you would change in the setup?




Hi @arnold1,

Great project and good looking CAD! So for starters I wouldn’t put my outlet as a mean velocity. While this might seem like the solution is not being forced, you may still potentially run into continuity errors as you probably have encountered. A potential fix would just be to set the outlet as a pressure outlet at gauge pressure (i.e pressure outlet, 0).

Additionally, I would insert some sort of result control that probes the temperature at say a couple points around the house so that it is easy to confirm that the result has converged and reached a steady-state. You wouldn’t want to post-process the result with data that hasn’t converged.

Hope this fixes the issue, and let me know!




You’re a gem, Barry. The pressure outlet allowed it to complete.

Now, is there a tutorial on temperature probes? I don’t see that in Simulation Control, are they custom boundary conditions?
I have a sketchy idea of convergence but my method here is to follow a procedure and pray that understanding dawns.

Thanks, Arnold


Hi @arnold1,

Cheers. Glad the solution worked.

Tutorial wise, there isn’t any that I am aware of. Here are a couple of screenshots to help you out.

So basically after selection of the result control to probe points, you then need to set the position of the probe (in the house of course) and the position of the point is defined under Geometry Primitives and must be input in x-y-z coordinates. How to deduce those coordinates is the hard part as you want to probe “interesting” points rather than randomly or everywhere.

From what I can guess, interesting points would be just near the heat source (but not too close) above the heat source, near the outlets and at other out of the way locations like corners in far off rooms etc etc. You’ll have to determine yourself what is best to check but the idea is to ensure that you can monitor as many adverse areas as possible. Looking at your geometry, I would say maybe a point in each room coupled with the ones I’ve stated, which brings it to about 10 points or so. Again totally flexible and you won’t need to be so precise as it is a laminar simulation which will not generated turbulence effects that can cause significant temperature fluctuations that you would otherwise have to try and monitor.

Side note, if you do decide to probe points, keep all your coordinates in a excel file. Will help in keeping your sanity if your probe data points goes missing for whatever reason here on SimScale. I believe there is a way to input the excel coordinates directly into the probe points but I’m not sure if its implemented yet. So its good to keep it around for future use whenever that is ready.

Another option to this (probably the easier/faster method) is to just let the sim run, check the convergence plots and ensure they are not converging anymore, post-process the data and get a couple of vertical and horizontal temperature slices. If you see any strange regions or bubbles of temperature, then you probably need more simulation end time to allow the solver to smooth those areas out. Again, maybe subjective so do post your results here after you’ve post-processed them if you’re unsure.

Hope this helps.



P.S Here is a good article regarding convergence. It is still very subjective depending on what your criteria is or what you are trying to find out.


Thankyou for your help, I’m sure I’ll be able to follow this. I’ll have a go at it tomorrow and I’ll come back with some pics when I have some.



Here you go @Get_Barried
The probe graph shows the points, I think this one is comparing the velocity of the air at each point, number 11 (the flyer) is the one closest to the open window at the rear of the house.
I was expecting to have to assign temperature to each point in advance but it seems that its all there and you can choose temperature or pressure in the results.

The coloured plot has a cutting plane through the heater and the warmed air looks like I might expect.

Thanks again, I’ll be passing this on to my students.



Hi @arnold1,

Brilliant. Glad things worked out. You might want to increase simulation time as it looks like it has not converged with reference to the point probes. My criteria for convergence is steady-state results with a maximum deviation of 1% for an arbitrary time length. However, if this is a preliminary study, then it is good to go.

Cheers and hope your students enjoy your content.



Just one more question if I may Barry, this is a roughy to get things working, and I plan to vary a few things when I get with my students but the idea of time is what I want to get my head around next. It looks like we’ve got a simulation time of 1000 seconds. I’ll try 5000 seconds and hopefully the convergence would improve with that change?

Still, in 1000 seconds I should be able to show tracers. I’ve tried, see pic below and nothing seems to appear.


Hi @arnold1,

The simulation is a steady-state. Time doesn’t really exist in this context. It is more like iterations rather than actual time. Imagine the solution actually interacting with the entire domain already, just that you need the solution to continue to calculate in order to produce a definitive, ideally numerically settled result. Hence, my notion of iterations rather than time.

Time that you are probably thinking of and are familiar with is in the context of flow traversing a medium and interacting with it at various timesteps, is only for transient simulations. That also comes with a whole mess of complications such as Courant number where we cannot allow information to flow faster than the actual source of that information (i.e the traversing fluid in motion) if not we reach a information discontinuity which obviously results in an error.

5000 timesteps may be too much, so to optimize a little, start with double the end time (2000s). It should be enough, if not just increase a little more to say 3000.

It should indeed. The post-processor is a little new to me as well. I will try to see if I can get streamlines out. Just a matter of selecting the right thing . Let you know with some steps once I get it.

Also feel free to ask questions anytime. Its good for me to try to articulate what I may or may not know. Great for both of us!