Enable pixel counting

asked 2020-07-17 06:09:40 -0600

rafael.alonso
933 ●1 ●6 ●12 http://atmoslab.io

updated 2020-08-16 16:28:13 -0600

__AmirRoth__

4456 ●5 ●17 http://bleedinggreenna...

Hi all,

I´m making some tests with the new PixelCounting shading feature of E+ 9.3 but I don´t seem to get it to work.

So far I have tested it on machine with a NVIDIA QUADRO P620 4GB with 512 CUDA core Pascal GPUs, but the error log that comes out reflects as if the PixelCounting was being ignored, as I get a bunch of these:

   ** Severe  ** Problem in interior solar distribution calculation (CHKBKS)
   **   ~~~   **    Solar Distribution = FullInteriorExterior will not work in Zone=ATTECHNIQUE
   **   ~~~   **    because vertex 2 of back surface=ATTECHNIQUEDOOR0 is in front of receiving surface=ATTECHNIQUEEXTWALL1
   **   ~~~   **    (Dot Product indicator=11.1000)
   **   ~~~   **    Check surface geometry; if OK, use Solar Distribution = FullExterior instead.

There is no message on the error log on why PixelCounting is not in use and if I run the file on a different machine (without graphic card) using PolygonClipping I get exactly the same results.

Does PixelCounting require any specific GPU configuration or settings somewhere else?

FIle: https://atmoslab.egnyte.com/dl/CDWVNe... Thanks! Rafael

Comments

@Neal Kruis or @MarkAdams any ideas?

Aaron Boranian ( 2020-07-17 09:02:38 -0600 )edit

@rafael.alonso can you share an IDF file? Also what platform are you working on (Windows, Mac, Linux)?

Neal Kruis ( 2020-07-17 09:37:05 -0600 )edit

Yes, I´ve just added it to the main post.

rafael.alonso ( 2020-07-17 09:45:44 -0600 )edit

add a comment

answered 2020-07-17 13:23:05 -0600

Neal Kruis

4775 ●11 ●17 http://bigladdersoftware.com/

The errors you are seeing are independent of the shading method you are using (they will appear for both PixelCounting and PolygonClipping. These errors indicate that there may be a problem with the geometric input for your surfaces. In fact, if you load your model into Euclid, it will automatically resolve some geometric issues for you.

If you run your file on a computer without a GPU you will see an explicit warning message stating that the simulation will revert to using PolygonClipping. Some computers have an integrated GPU and not a dedicated graphics card. When I run your file on my computer (with an integrated GPU), I can tell that the simulation is utilizing the GPU (and runs considerably faster than PolygonClipping).

Finally, I will point out that you can drastically reduce the number of shading surfaces in your model. For example, for six-sided cube-shaped neighboring building, you typically only need to include walls that you can "see" between a heat transfer surfaces in your model and the sky. A flat roof of an enclosed space will never cast a shadow. Typically, one or two walls of a cube will never shade your model. Ideally, EnergyPlus would have a way to automatically detect these situations and ignore those surfaces when performing shading calculations, but that is not currently the case.

edit flag offensive delete link

Comments

Many thanks for the detailed answer. Then both of my machines are running the pixel counting method, which is great news.

As for the severe errors in from the FullInteriorExterior solar distribution, can I just ignore them and asume that the internal solar distribution is correct? I thought PixelCounting eliminates that limitation and, therefore, the errors.

rafael.alonso ( 2020-07-20 04:01:12 -0600 )edit

PixelCounting should eliminate any limitations related to non-convex surfaces. The errors you're showing appear to be related to different problems in the geometric input (though I didn't look to be sure). If you do see errors related to non-convex geometry with PixelCounting, let me know--it could be a bug.

Neal Kruis ( 2020-07-22 13:44:19 -0600 )edit

Thanks for the helpful response, Neal. I actually have found a case of E+ giving non-convex errors with PixelCounting and I opened a separate topic for it here. It smells like a bug as you said.