Approximating the gross daylight flux through venetian blinds.

Hello Radiance people

1) Anything wrong with this idea?: �If venetian blinds are involved,
to get to the daylight factor with an overcast sky environment,
cut a corner by using a few primitives with complex behaviour to
approximate the gross daylight fluxes through the blinds, instead of
modelling the same using an awful lot of fiddly primitives albeit
with simpler materials.� (More details below.)

2) If you have tried to use BRDFs or other types of materials with
the same idea in mind do you have experiences to relate,
supportive or otherwise?

3) Does anybody know where I can get good data on venetian blinds re
typical performance and typical material surface properties?

4) Details:

I�m usually under pressure to give the right answer yesterday, at all
costs and at no cost. So I went for quick and dirty. I asked around
whether quick and dirty would be fine if in professional hands
with some of the others here.

This generally got a lot of smiling for some reason; personally I can�t
see how being forced into lack of rigour due to expedience could be
cause for amusement in any true professional. I took their smiling as
showing their support for my approach.

The supporting environment was no magnificent computing facilities,
no Unix, no full-scale lab test rigs, no photometers etc., as yet no
Cygwin, Photon Map, or Raydirect/Exergia's radiance
preprocessors for blinds/mktis, or anything more recent than the
compiled binaries from DeskTop Radiance. I haven�t tried using
prismatic materials yet. I�ve also yet to read the 2001 paper on daylight
simulations for a full-scale test office with outer venetian blinds that
Christoph Reinhart kindly sent me, and the stuff re Photon Maps.
Adding onto all that is my general ignorance and minimal Radiance
experience, so you'll agree that it's a truly impressive setup.

Reasonable rigour for my purposes was expectation of a daylight factor
end figure being within 2:1 of �actual�. I.e. the final answer only needed
to distinguish whether the daylight factor was either say 0.5% to 1%,
or 1% to 2%.

The rad files I made grossly simplified the building�s complex facade so
as to more easily do the computing and extract the daylight factor
statistics I needed from the Radiance lux levels.

The daylight problem was a large geometrically complex multi-storey
atrium building, with a triple-glazed fa�ade with 3 kinds of glass and
interstitial venetian blinds, proposed for a built-up city centre location
with complex geometry and lots of materials to model. The design basis
was the usual overcast sky. Not interested in direct sunlight.

Clearly there�d have been some differences from the real world in
assuming the materials and surface reflectances etc., but I'd only
wanted the rough illuminance distributions anyway; not the fine details.

So being neither purist nor pure, I pressed on and:

- modelled a small Test Room A with the triple fa�ade modelled
  accurately in geometric detail, using Optics5 data for each sheet
  of glass, and assumed reflectances and detailed geometry for the
  venetian blinds.

- modelled a Test Room B of the same dimensions, but with a
  pseudo-fa�ade of simple plane elements with BRDFs,

- tweaked the BRDFs until the renderings compared well between
  Rooms A and B under 3 outside environments, with external
  obstructions, materials and ground planes,

- applied the fa�ade and BRDFs of Test Room B to my big models,
  so I can crank out the illuminances and extract the daylight statistics.

Et voil�, yet how far will I be able to trust the end result?

Regards
George Chadwick

···

==
Possibly related stuff found in these threads:
- Gerold Furler - query on BRDF - Venetian blinds
     - Feb 1995
- Greg Ward - Amarpreet Sethi - Barbara Matusiak -Peter
     Apian-Bennewitz -Carsten Bauer - - �Modeling blinds -
     Suggesting photon-map� Oct 2003.
- Greg Ward -Jack de Valpine - Daniel Lash - Alex Summerfield -
     - �Query about accurately modelling glazing� Jan 2004.
- Anthony Farrell - Phillip Greenup - Jack de Valpine - -
     �Prismatic glazing-blind reflector system analysis�
     Feb 2004.

---------------------------------
ALL-NEW Yahoo! Messenger - all new features - even more fun!

Hi George,

That's quite an e-mail. At least you've done your research. However, I don't think you'll get anyone to "bless" your calculation based on your current approach. I won't even hazard a guess as to how close your results are to correct. (I hesitate to make any claims about my own calculations for that matter.)

Given the amount of time you must have spent on this, even on composing this e-mail, don't you think you should spend a little time getting a newer version of Radiance onto your computer? Just a thought.

Here are some others:

- modelled a small Test Room A with the triple façade modelled
accurately in geometric detail, using Optics5 data for each sheet
of glass, and assumed reflectances and detailed geometry for the
venetian blinds.

Does Optics5 do ray-tracing to figure out the spatial distribution of light coming from scattered sources? I had no idea it had such a capability! I really should talk to my compatriots at LBNL more often...

- modelled a Test Room B of the same dimensions, but with a
pseudo-façade of simple plane elements with BRDFs,

Using BRTDfunc, or what?

- tweaked the BRDFs until the renderings compared well between
Rooms A and B under 3 outside environments, with external
obstructions, materials and ground planes,

This is the part that has me most worried. Getting a reasonable match in a few different scenarios is no substitute for good modeling.

- applied the façade and BRDFs of Test Room B to my big models,
so I can crank out the illuminances and extract the daylight statistics.

It should be possible to arrive at a reasonable BRTDfunc approximation to blinds based on a particular slat spacing, angle, and reflectance. It would take me a week or so to do it, though. I wouldn't trust plasfunc or any of the other types, because they don't include a pure transmitted component, which you will certainly have with venetian blinds. In short, I think for the time being at least, you should stick with a mkillum-driven calculation, or rely on the basic Radiance interreflection calculation with some large values for -ad and -as if all you want are a few data points per time step.

That's my two cents. I hope the others have something more to say, because I know there are users out there with more experience with daylight simulations than I have....

-Greg

Hi George,

Newer version of Radiance:

I'm embarrassed by your raising the subject of
me getting a newer version of Radiance onto
my computer; that's been on my mind. I do
need to catch up but to my shame have done
nothing. Richard Morris at Arup Lighting in
London helpfully suggested Cygwin as
the way I can get more in line with everybody else
and up to speed despite being a Window$ user,
and pointed me to Francesco Anselmo's site.

That's a good suggestion, though you might want to wait until Francesco compiles the latest 3.6.1 release, which just came out yesterday. That way, you'll be sure to have the most recent additions. I haven't used Cygwin (or PC's) myself, but I guess it works. FreeBSD is my favorite option on Intel hardware -- much easier to manage than Linux for most things, though of course Linux comes with a friendlier desktop these days. If you can afford it, an Apple running OS X is a very good investment. The time you save in administering it will more than compensate for the higher price tag, and the G5 performance is quite competitive with Intel and AMD machines these days. Plus, you'd get to run all your Office applications and Photosphere to boot!

Optics5 data:

My apologies for being unclear on using Optics5
data: all I did was take the rad file outputs from
Optics5 for each proprietary type of glass and
put that data into my Radiance materials files,
nothing more. How sophisticated Optics5 output is,
I have no idea. I guess in an ideal world
Optics5 would spool off a customised Radiance
BRTDfunc for each glass type or combination of
glass and framing in a glazing system, but
somehow I don't suppose it actually can. I am
guessing what you get are averaged values
derived from hemispherical test data which you
use as simple transmission and reflection
factors. I can't see the incident angle figuring in
the output anywhere. Maybe someone would
like to put me right.

You didn't post your reply to radiance-general -- just to me. Did you mean to post it? I will post this reply in case others want to pipe in. I don't have any experience with Optics 5, other than files people have shown me. From what I can tell, it does a pretty rudimentary job modeling glazings in Radiance -- not even accounting for Fresnel effects on reflection and transmission. The glaze script I wrote for Visarc does a more reasonable job in this regard, but lacks the extensive database of Optics 5. I misunderstood and thought you had used it to compute the BRDF for the blinds, which I didn't think it could do. I guess from what you say, it can't.

Mkillum-driven calculations:

Thanks to your advice on using mkillum-driven
calculations, it's now full steam ahead with that.
As I'm using mkillum for the first time, I am taking
it as read that there will be a way I can use an
illum to account for all those room cases
with different external obstructions and
significant reflections arising from the rooms'
external contexts.

Yes, that is correct. Mkillum simply acts as a "go-between," mediating and accelerating the window contributions in your model. As I said in my first reply, it doesn't help so much if all you're doing is calculating a few illuminance values, but if what you want is a nice polished image, I've found it to be indispensible. I only wish it were a little easier to use.

BRTDfunc:

There would be a huge cheer from me if
you were to publish a BRTDfunc with macro
performance approximating reasonably to
venetian blinds.

Unfortunately, a good cheer might not be quite enough motivation, but I'll give it a look.

I do take right on my chin your point about
my getting a reasonable match in a couple of
arbitrary cases being no substitute for good
modelling. I’m still interested in any real
world venetian blind physical data, such as
anyone may have, not just so as to provide
capability for validation of whatever the
BRTDfunc may turn out to be, but also
generally for other studies.

I know Kostantinos Papamichael did a fair amount of research on venetian blinds for his Ph.D. You should be able to find some of his articles. Go to the following URL and type in Papamichael. That name shouldn't create a lot of search collisions:

  http://www-library.lbl.gov/lbnl_reports/sf

A thought: maybe it would be too much to go
for in one function, but it seems to me that
perhaps a useful general BRTDfunc for surfaces
with pseudo-venetian-blind macroperformance
properties would need switches for the user to
specify say a world vector normal to the slat axes
and a slat angle relative to the surface normal.
Then if possible maybe such a BRTDfunc could
be made to work for vertical blinds too. Maybe I’m
wrong, but my subjective impression is that real
blinds have direct and diffuse transmissions which
are both asymmetric w.r.t the azimuth angle of the
incident ray relative to the slat axis in the surface
plane, the diffuse distribution for each ray case
being itself asymmetric w.r.t azimuth angle. Ouch.

Yes, it should be possible. Of course, a BRTDfunc for venetian blinds won't account for all that goes on outside a window, because Radiance does not sample the directional diffuse component of its BRDF types. This is where mkillum and the photon mapping add-on will have the upper hand.

-Greg

I gave it a little more thought, and I'm note sure how useful it would be to have a BRTDfunc that approximates the aggregate behavior of venetian blinds. The only context where this might prove useful is in performing daylight analyses where you don't care about external reflections and the users doesn't care to visualize the space. Because a BRTDfunc doesn't include the directional-diffuse portion in its indirect component, it wouldn't get an accurate contribution from light reflected off external surfaces. The best it could do is get a reasonable value for the indirect solar due to light hitting the blinds directly. Mathematically, even this would be difficult -- I'm guessing it would take me a day or two to work up a reasonable approximation. As far as the appearance goes, one would see a semi-translucent material if one was to visualize your model with this material, which wouldn't end up looking like blinds at all.

In summary, I don't think this would be much of an improvement over a mkillum solution. Does anyone else have some thoughts on this?

-Greg

George,

This paper may be helpful:

ANALYSING RADIATION TRANSPORT THROUGH COMPLEX FENESTRATION SYSTEMS
N.S. Campbell and J.K. Whittle
http://www.ibpsa.org/proceedings/bs97/papers/P069.PDF

> The design basis was the usual overcast sky. Not interested in direct > sunlight.

> Clearly there’d have been some differences from the real world in
> assuming the materials and surface reflectances etc., but I'd only
> wanted the rough illuminance distributions anyway; not the fine
> details.

I appreciate that a design basis is just that - and we don't have much
choice if it is what the client wants. However, I can't hold myself
back from wondering just how useful is a daylight factor evaluation for
the scenario you describe. Surely, the blinds are there also to control
direct sunlight? Perhaps even it is their main purpose.

-John

···

-----------------------------------------------
Dr. John Mardaljevic
Senior Research Fellow
Institute of Energy and Sustainable Development
De Montfort University
The Gateway
Leicester
LE1 9BH, UK
+44 (0) 116 257 7972
+44 (0) 116 257 7981 (fax)

[email protected]
http://www.iesd.dmu.ac.uk/~jm

Hello,
There`s a new paper that is related, but can be accessed online only (as far
as I know):

M. Andersen, M. Rubin, R. Powles, J.-L. Scartezzini, Bi-directional
transmission properties of venetian blinds: experimental assessment compared
to ray-tracing simulations, Solar Energy, In Press.

It discuses mainly the method to measure BRTDs, but not much about the
application to simulations. If somebody has more information about this
research, I would also be interested to know.
On the other hand, to simulate blinds, would it be possible to combine a
BRTD with a linear function that accounts for the slats? I mean, can the
BRTDfunc be dependent on the object coordinates? just an idea, I don`t know
if it makes much sense.

Regards,
Santiago

···

-----Original Message-----
From: [email protected]
[mailto:[email protected]]On Behalf Of
Reinhart, Christoph
Sent: Saturday, October 30, 2004 1:10 AM
To: 'Radiance general discussion'
Subject: RE: [Radiance-general] Approximating the gross daylight flux
through venetian blinds.

Greg, I think that it would be REALLY useful for
daylight-coefficient-based
daylight simulations if someone (meaning you:)) came up with a
BRTDFunc for
blinds. This way one could simulate interior annual illuminance profiles
with the blinds opened and closed. This is already possible if
one simulates
the blinds geometrically in gory detail (genblind), but the
simulation times
are horrendous. I assume the calculation time penalty would be less severe
if one used "BRTDFunc" instead? To my knowledge, mkillum cannot be used
with a daylight coefficient approach since mkillum is tied to a
specific sky
condition?

Christoph

I haven't read this paper -- thanks for pointing it out. I know that Marilyne Anderson is continuing her measurement work now that she's at MIT.

Regarding your suggestion with the BRTDfunc, if I understand you correctly, you are suggesting that it might be possible to simulate the apprearance of slats in the model. I agree, but I think that might defeat the purpose of using such a model, as we would be reintroducing irregularities in the light coming through that would then require heavier sampling to resolve. That is the current problem with venetian blind models -- it's not that Radiance can't handle the complex geometry, quite the contrary. It is simply that the slats themselves cause a large variance in the penetrating radiation, to the point where many ray samples are needed to resolve the agregate (i.e., integrated) behavior.

I hope I didn't misunderstand what you were saying.

-Greg