why sensor inside a closed box still got irradiance reading?

Dear Radiance Experts,

I found that the sensor located at the center of a simple closed box still
got irradiance reading, though very small.

The enviornment file used is:
!gensky 6 20 +15.0 -a 1.0 -o 103.0 -m 105 -c -b 22.8600006104 -g
0.20000000298
skyfunc glow sky_glow
0 0 4 1.0 1.0 1.0 0
sky_glow source sky
0 0 4 0 0 1 180
skyfunc glow ground_glow
0 0 4 1.0 1.0 1.0 0
ground_glow source ground
0 0 4 0 0 -1 180

The rtrace command used is:
rtrace -I -aa 0.15 -ab 5 -ad 2048 -ar 128 -as 256 -h -w -oov -u- scene.oct <
sensors.txt > results_ir_RGB.txt

There should be no radiation falling on a point inside a closed space,
right?

can you help to advize me why Radiance still produce non-zero results in
this situation?

Ji

Hello Ji.

Dear Radiance Experts,

I found that the sensor located at the center of a simple closed box still
got irradiance reading, though very small.

[...]

The rtrace command used is:
rtrace -I -aa 0.15 -ab 5 -ad 2048 -ar 128 -as 256 -h -w -oov -u- scene.oct <
sensors.txt > results_ir_RGB.txt

I'm wondering why you use '-ab 5' if the box is closed and has no
openings. But if you can spare the time for the calculation there is
no harm in it.

There should be no radiation falling on a point inside a closed space,
right?

Right. But the box is only a geometrical box and not a physical one.
If your box has only a single polygon layer as "wall" then you might
see the result of interpolation along the edges here. Try to enclose
your box in another larger box and do your tests again.

When you model a real test scene try to model the inside and outside
of a wall separately. That should avoid errors due to interpolation in
your results.

Regards,
Thomas

···

On Wed, Oct 20, 2010 at 8:06 AM, Ji Zhang <[email protected]> wrote:

Hi Ji,

Thomas is essentially correct. The lack of thickness to your walls means that the occasional ray can "escape" out the corners of your box, permitting light to leak in during your calculation. Using a sphere rather than a box, you'll find your value goes to zero. Or, adding a second box to enclose your first one also fixes the problem.

These kinds of errors are inherent to floating point calculations, and people have written entire Ph.D. theses addressing the problem.

Best,
-Greg

···

From: Thomas Bleicher <[email protected]>
Date: October 20, 2010 7:04:29 AM PDT

Hello Ji.

On Wed, Oct 20, 2010 at 8:06 AM, Ji Zhang <[email protected]> wrote:

Dear Radiance Experts,

I found that the sensor located at the center of a simple closed box still
got irradiance reading, though very small.

[...]

The rtrace command used is:
rtrace -I -aa 0.15 -ab 5 -ad 2048 -ar 128 -as 256 -h -w -oov -u- scene.oct <
sensors.txt > results_ir_RGB.txt

I'm wondering why you use '-ab 5' if the box is closed and has no
openings. But if you can spare the time for the calculation there is
no harm in it.

There should be no radiation falling on a point inside a closed space,
right?

Right. But the box is only a geometrical box and not a physical one.
If your box has only a single polygon layer as "wall" then you might
see the result of interpolation along the edges here. Try to enclose
your box in another larger box and do your tests again.

When you model a real test scene try to model the inside and outside
of a wall separately. That should avoid errors due to interpolation in
your results.

Regards,
Thomas

Does this imply that best practice in general, if one is measuring levels and perhaps even in general, is to give walls actual thickness? I've been chasing rsensor problems lately, and I wonder if that would help.

···

On 2010-10-20 09:41:04 -0700, Greg Ward said:

Thomas is essentially correct. The lack of thickness to your walls means that the occasional ray can "escape" out the corners of your box, permitting light to leak in during your calculation. Using a sphere rather than a box, you'll find your value goes to zero. Or, adding a second box to enclose your first one also fixes the problem.

--
Randolph M. Fritz • [email protected]
Environmental Energy Technologies Division • Lawrence Berkeley Labs

Yes, walls should have thickness. They do in real life, and it definitely helps avoid light leaks of all sorts.

-Greg

···

From: "Randolph M. Fritz" <[email protected]>
Date: October 20, 2010 10:09:04 AM PDT

On 2010-10-20 09:41:04 -0700, Greg Ward said:

Thomas is essentially correct. The lack of thickness to your walls means that the occasional ray can "escape" out the corners of your box, permitting light to leak in during your calculation. Using a sphere rather than a box, you'll find your value goes to zero. Or, adding a second box to enclose your first one also fixes the problem.

Does this imply that best practice in general, if one is measuring levels and perhaps even in general, is to give walls actual thickness? I've been chasing rsensor problems lately, and I wonder if that would help.

Whoa, back to the drawing board, or at least the modeling program. Oh, well. Prof. Ganter at UW has a really-doesn't-like-surface-models rant, and every now & again I find myself thinking he has a point.

Randolph

···

On 2010-10-20 10:13:07 -0700, Greg Ward said:

Yes, walls should have thickness. They do in real life, and it definitely helps avoid light leaks of all sorts.

Thank you, Greg, Thomas, and Randolph!

We've tried according to your suggestion by using a box with "thickness",
i.e. the walls of the box are composed of two layers of polygons, with the
normal of the external layer pointing outward, and the nomal of the internal
layer pointing inward.

However, it seems the irradiance reading of the center of this box still
varies depending on the thickness of the wall, or the distance between the
two layers of polygons.

The variation seems irragular as well, i.e. we got zero reading for the
centre for one thickness, but we then got non-zero reading when the
thickness is reduced or increased.

Is it still related to the float point calculation matter?

Or, should we simply ignore this issue because the non-zero reading is still
fairly small and can be rounded to zero if the precision required is not
that high?
(e.g. for an overcast sky which will produce a horrizontal illuminance of
10,000lx the illuminance reading of the center of the box is 0.00249131lx)

Thanks!

Ji

···

On Thu, Oct 21, 2010 at 1:13 AM, Greg Ward <[email protected]> wrote:

Yes, walls should have thickness. They do in real life, and it definitely
helps avoid light leaks of all sorts.

-Greg

> From: "Randolph M. Fritz" <[email protected]>
> Date: October 20, 2010 10:09:04 AM PDT
>
> On 2010-10-20 09:41:04 -0700, Greg Ward said:
>
>> Thomas is essentially correct. The lack of thickness to your walls
means that the occasional ray can "escape" out the corners of your box,
permitting light to leak in during your calculation. Using a sphere rather
than a box, you'll find your value goes to zero. Or, adding a second box to
enclose your first one also fixes the problem.
>
> Does this imply that best practice in general, if one is measuring levels
and perhaps even in general, is to give walls actual thickness? I've been
chasing rsensor problems lately, and I wonder if that would help.

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

Ji.

Have you tried Greg's suggestion to enclose the scene in a sphere already?

For practical purposes you can ignore the small error introduces by
the interpolation. Any reasonably lit space will show illuminance
values which are several magnitudes above that. However, you should
still verify that it is indeed the interpolation that causes your
readings.

If you get zero values with an enclosing sphere that's a good
indicator for it. You can also render a hemispherical image towards
the ceiling and the floor. Interpolation errors should show as bright
spots along the edges.

Regards,
Thomas

···

On Sun, Oct 24, 2010 at 10:18 AM, Ji Zhang <[email protected]> wrote:

Thank you, Greg, Thomas, and Randolph!

We've tried according to your suggestion by using a box with "thickness",
i.e. the walls of the box are composed of two layers of polygons, with the
normal of the external layer pointing outward, and the nomal of the internal
layer pointing inward.

However, it seems the irradiance reading of the center of this box still
varies depending on the thickness of the wall, or the distance between the
two layers of polygons.

The variation seems irragular as well, i.e. we got zero reading for the
centre for one thickness, but we then got non-zero reading when the
thickness is reduced or increased.

Is it still related to the float point calculation matter?

Or, should we simply ignore this issue because the non-zero reading is still
fairly small and can be rounded to zero if the precision required is not
that high?
(e.g. for an overcast sky which will produce a horrizontal illuminance of
10,000lx the illuminance reading of the center of the box is 0.00249131lx)

Thanks!

Ji

On Thu, Oct 21, 2010 at 1:13 AM, Greg Ward <[email protected]> wrote:

Yes, walls should have thickness. They do in real life, and it definitely
helps avoid light leaks of all sorts.

-Greg

> From: "Randolph M. Fritz" <[email protected]>
> Date: October 20, 2010 10:09:04 AM PDT
>
> On 2010-10-20 09:41:04 -0700, Greg Ward said:
>
>> Thomas is essentially correct. The lack of thickness to your walls
>> means that the occasional ray can "escape" out the corners of your box,
>> permitting light to leak in during your calculation. Using a sphere rather
>> than a box, you'll find your value goes to zero. Or, adding a second box to
>> enclose your first one also fixes the problem.
>
> Does this imply that best practice in general, if one is measuring
> levels and perhaps even in general, is to give walls actual thickness? I've
> been chasing rsensor problems lately, and I wonder if that would help.

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general