below my comments “inline”
I am looking to evaluate glare on an annual basis for a case with a fabric roller blind with a high openness factor. This would involve evaluating instances where the sun is in direct view of the occupant through the shade fabric.
I’m currently considering the following indicators/simulation methods (other suggestions are appreciated):
- DGPsimplified;enhanced using rtrace for the luminance distribution images and either 3 or 5 phase method for the illuminance part.
I would definitely NOT recommend this - Using Ev or a poor resolution of the sun will definitely lead to wrong results (underestimation of glare situations). Ev would be only capable, if the diffuse transmittance of a farbic would be that large giving an effect. But these fabrics are typically not sold. E.g. the most sold one in Europe have typically 2-3% direct transmittance and less than 6% diffuse transmittance. With this, Ev and metrics based only on EV will not work at all.
- DGP based on luminance distribution from 5 phase method image base simulation.
The 5-phase-method would require a high resolution of a) your BSDF and b) of the sky or a separate modelling of the sun. And “high” means in the order of 1 degree. To my knowledge, this is not yet established. But Greg and LBNL are working on that and are in exchange with Lars, David and me.
Daysim (and DIVA) are slow, but are capable running it, but not based on BSDF, but on geometric modelling and works also with “my” fabric model you mentioned (see also below).
- GlareEv; a metric using direct and total illuminance based on research by Konstantzos and Tzempelikos. The metric was derived for conditions where the sun is visible through fabric shades. Here I would also use the 5 phase method.
While writing the cross-validation paper we have looked also into that metric, although we did not include it in the evaluation since it would have violated our rule not to use any developement data (we were using the dataset for the development of GlareEv) - it seemed not to be as robust as many other metrics.
My problem is related to modelling the fabric for this purpose when I only have data from the CGDB available, so a low resolution BSDF (145 klems) and overall transmittance and reflectance at normal incidence. To my understanding using a Klems-resolution BSDF will distribute directly transmitted light over a large solid angle and this would cause an underestimation of glare (correct?).
yes, this is right. The sun disc should be modeled “somehow” correctly, keeping basic physics.
I’m therefore considering using a different material type for the direct/solar part of the simulations but this raises the question of which material to use and how to supply it with meaningful inputs. Direct to direct transmittance of fabric is strongly dependent on incidence angle so I want to be able to describe this accurately.
I went through Peter Apian-Bennewitz’s review of materials in Radiance but I’m still a bit puzzled. These are the options I am currently considering:
- TransData with a data file describing the angular dependence of transmittance and reflectance using values I find at discrete incidence angles from the measured BSDF (the behaviour of the fabric seams quite symmetrical so I think doing this in a one-dimensional way would suffice).
- Transfunc or BRTDfunc. Using a function file to compute off-normal properties from a study by Kotey et al or the shade function that Greg shared here.
- A two layer approach presented by Jan Wienold in ‘Annual glare evaluation for fabrics’. Here he takes one glass layer (transmissivity = 1) and uses its refractive index to model the cut-off effect of the fabric where its transmittance goes to zero at higher incidence angles. As a second layer he uses a BRTDfunc to describe the angular transmittance.
I was hoping to get some views on some of the following questions:
- Am I correct in concluding a low resolution BSDF will lead to an underestimation of glare from sun shining through a fabric?
- On this forum I read that there is some fundamental difference in the way TransData and Transfunc behave as opposed to the behaviour of BRTDfunc. I see that Greg recommends BRTDfunc in many cases relating to frabric like materials but I don’t fully understand why. Could anyone clarify?
- From Apian-Bennewitz’s paper I understand that transFunc, transData and BRTDfunc can perform badly in describing a CFS’s behaviour under sky conditions without sun. Considering that I am only planning to use this material only for determining the direct/solar component in glare assessments, am I correct in concluding that this is not a big problem for me?
- Does anyone know where I can find some examples of how to use BRTDfunc and transData?
- Does anyone know if Jan Wienold’s Radiance fabric models are available somewhere?
I uploaded it here, together with the paper you are mentioning. It should be self-explaining.
- Is it correct that for the Wienold’s two layer approach you need to use a mixFunc?
No. The outer layer is mainly there to “pull down” the diffuse transmittance. the cal- file controls the direct part.
- Your thoughts on my proposed approaches would also be very welcome.
Your post comes 2 months too early… I have developed a new annual method which bridges the time gap between the current need of a fast evaluation method (e.g. for the EN17037) and the general solution for an effective temporal and spatial glare assessment, where Stephan Wasilewski, a PhD student of mine, is working on.
This new method of mine is already validated (but not yet published) for fabrics (and the BRTDfunc-model using a cal-file) , but I’m still extending and validating it for other complicated cases. The advantage of that method is, compared to Ev based methods or the new method from Nathaniel, the high accuracy and that it works also for complicated cases (e.g. fabrics, specular reflections from metal surfaces, external buildings…). Compared to the “old” gen_dgp_profile-tool implemented in DIVA (which is also accurate), the time effort is dramatically reduced from hours to a few min (e.g. the fabric case with a shoe box takes 1.5min for hourly data on my MAC). The disadvantage of this method is that the extension to a grid-based analysis is a rather limited path - so the time effort gets easily large in case you want to evaluate a full grid. There are also other limitations, due to the method, which I can share later, but for the fabric case they can be neglected. The method itself will be compatible with the daysim- and with the n-phase-methods.
In case you are using a linux or mac environment, I’m willing to share with you the current code (please send me then an email directly so we can clarify this, it would take me also a bit to clean it up to share), which is not yet embedded in a c-code but in bash/awk scripts. But it is still in an “alpha-stage”, so not robust against any wrong parametrization. The plan for now is to finalize the tool and its validation, then to put it into a c-code to make it available also on Windows (will be around end November I guess).
But, as mentioned before, the the daysim approach (“gen_dgp_profile”) definitely works as well “out of the box” - it is just very slow.