Make a poor integrating sphere, get poor results. Help?

LED Flashlights – General Info
1

edit: Updates in post #12

Sorry for the length, I tried to bold the important text.

So yesterday I was all excited to whip up an integrating sphere (box) from a little Styrofoam box I had laying around, total cost for the whole project: $0.00 I used the method where I did not cut a hole in the box for the light meter, I just sanded the styrofoam a little thinner under the sensor. As such, I did not put a baffle in the box since the light meter cannot see the light source through the wall.

I took all of the non-modified lights that I had, and put fresh (rested) cells in them. I tried to find either manufacturer or reviewer-measured lumens values that I trusted. Then I took readings for all the lights and plotted “measured Lux vs lumens” of these “known” lights to try and get a correlation. And I did get a correlation, what appeared to be a good one, in fact. (see attached pic)

But then I tried measuring a couple modified lights, and I’m not sure it’s giving me accurate results. Both on good 30Q cells:

  • P60 quad, 2xXP-L Hi V2, 2xXP-L Hi V3, measuring 3.96A draw – 1205 Lumens
  • Convoy C8, XP-L HD V6, MTN FET+1 DD driver, spring bypasses, SMO reflector, AR glass, current above 5A (I forgot to recheck it) – 1358 Lumens

Per djozz testing, that bare XP-L HD V6 should be about 1900Lumen at 5.5A. Obviously I lose some lumens OTF, but it shouldn’t be 30, right? And I was calculating ~1700-1800 emitter lumens for the quad, a 40 “under-performance”.

I’m thinking a couple things, and would like your input:

  • The box size is too small: Interior dimensions are 10cm x 20cm x 13cm. The hold size is 5.7cm diameter. Interior surface area = 1180cm^2, hole size = 25.5cm^2, so the hole is about 2.2% of the surface area. Also, I used white card stock as custom sized skirts for each light.
  • The box is too thin: It is 10mm thick, am I losing significant amounts of light out of the box on higher power lights? I covered the exterior walls of the box with cardstock because under kitchen lights, the luxmeter would read 2 or 3 lux due to room light leaking through the box walls.
  • The light meter is faulty: I’m using the common and seemingly decent HS1010A.
  • I don’t have enough reliable high power lights to get accurate “known” values above a couple hundred lumens
  • My expectations for my modified lights were wrong: – maybe my 4Amp P60 quad really only does 1200Lumens, and my 5.5A XPL V6 C8 really only does 1350Lumen?

Your thoughts and input are appreciated.

Here is a screenshot of my attempted calibration. Note that the majority of measurements are under 100Lumens, and only a handful are between 200 and 1000Lumens. (edit: the C8 in the pic is not stock, it was the modified light I was measuring, discussed above).

Here is the integrating box, HS1010A, and stock lights then I was testing.

2

That calibration line looks pretty convincing, still a few remarks:
-loosing some light through the walls is not a big deal, as long as it is a fixed percentage.
-the entrance hole is pretty big compared to the box size, expect some measurement error from that, even if you use the skirts.
-I think that it does matter that light from the entrance hole can directly shine on the area where the meter is, even if it measures through the wall. A better postion for the meter may be next to the entrance hole. But in practice a flashlight will not easily shine on the meter area because the beam angle is limited, but a bare led would.

About expected OTF results in a flashlight: it will be hard to obtain the led cooling that I achieve in my bare led tests, that may be a factor involved in a lower output than expected from bare led tests. Further, the djozz-lumen is very consistent but not necessarily correct, I never did an absolute calibration (=rather expensive).

3

Well from what I’ve seen, the djozz-lumen seems to correlate pretty well with what the Cree datasheets predict. I know you’ve got to throw that disclaimer in there whenever anyone like me quotes you as fact, but don’t sell yourself short either.

Is there some rule-of-thumb (generic rule) for flashlight opening/interior surface area ratio? Testing a light with/without the skirt made a noticable difference, so I suspected the hole might be kind of big.

I don’t think that heat was a factor in my “low” measurements - the lux value at instant-on for the P60 quad was 36500, and the 30-second value was 36300. (That light, despite being in a plastic P1 host, is remarkably thermally capable at 4amps - I have ceiling bounce tested it for 5 minutes on high with less than 5% drop in lumens.)

I know OTF lumens are less than LED, but 30-40%, really? If so, wow - I didn’t know it was that much.

One of the things I like about EagleTac is that they publish their “Technical Specifications” sheets, which includes both LED lumens, and ANSI-FL1 lumens. Looking at the spec sheet for the EagleTac P20A2 mkII (which I have, and used as part of calibration) the biggest difference from LED to ANSI was 27.4. I flipped through a handful of other spec sheets and didn’t find any others with that big of a discrepancy, most were 15-25 losses from LED to ANSI.

4

From djozz’s measurements here, a triple optic lost 27% of the output, which is close to your measurement (1280/1700=.72).

The stock S2 reflector with non-AR glass lost 17%. Your estimated loss in your C8 is more than that. The difference could come from lower-than-expected current, variance in emitter outputs, differences in lux meters. Or your integrating sphere could have some directional differences that could cause a more focused beam to read less than broader beams.

5

Ambient room lighting will bleed through the styra-foam and into the box. If you don’t block it out completely it can be a source of variation.

6

In my relatively bright kitchen (I think I measured about 300lux at countertop level?)where I built the light, I was registering zero lux with the meter installed in the box and cardstock covering the light entrance hole. I figured if I made it impervious to outside light in the bright kitchen, I could use it in dimmer rooms without worrying about that source of variation.

I ran the tests in a more dimly lit area (~40lux at table level), and the lux meter still read zero, as should be expected. So I don’t think light bleeding into the box was the issue.

And even if was bleeding in, wouldn’t it be a constant offset? For example, say 2 lux are bleeding through the styrofoam: If I’m measuring a moonlight mode that should register 20 lux, then my reading will be 22lux. If I’m measuring my quad, instead of the proper 36,300lux, I will get 36,302lux, right? The ambient light doesn’t scale with the power of the light I’m measuring.

7

I think the box itself is the issue. Not light leakage or other things.
I don’t know the technical terms here but I don’t see how it could be accurate light to light.
Most of the light is bouncing from back to front, with primarily the beam width angle determining how much light reaches the sensor on the side. The light can’t bounce around the box appropriately be smoothed and then be captured by the sensor, due to it being a square/rectangle.
This seems fundamentally wrong compared to what actual spheres attempt to do.
Just my 2cents so please don’t confuse it for a nickel. :money_mouth_face:

Hats off to you for the attempt and the valid results you have gotten. :+1:

8

Maybe the term you are looking for is like “Aspect Ratio”?

Good point.

I thought about that as I was making the box, but I forgot it as I pondered the results - maybe a cube box of equal length sides would be closer to “good enough”, but maybe the fact that the aspect ratio of my box is 2x1 messes with the reflections just too much.

9

I never made a square lightbox but many others have, including ‘integrating milk cartons’ that are reported to work pretty ok. This may not be the biggest worry.
Did you btw sand the inside of the box to a true matt finish? If not, that will help integration quite a lot.

10

I did not sand the interior, I will do that and report back.

What do you think about taking a curved peice of card stock and placing it in the long end of the box? Would that impove the “aspect ratio” some, and maybe also improve the integration?

(Couldn’t find a better image)

11

I'm no expert a I cn only tell you the results I get from my own box which is set up a little different but, I've put probably 30 known lights in it.

The only one I have to comapre to your measurements is the ThorFire G10. My highest reading is 877 and that is a 30Q stright from the charger just to see any differences. If I let ot run on high for about a minite that settles to 847.

I know that's no help. Just a reference. It drove me nuts until I got enough light sources to trust it..

12

I did two things last night:

  • Sanded the interior surfaces of the I.S.
  • Acquired several more high power (1000-ish lumen) lights

Here is the new calibration. I feel more comfortable with this, especially with the larger sample size of known lights.

I don’t plan to get into the light review business, so I don’t need to know the “true” absolute lumen value of any light. For my primary purpose - measuring relative effects of my modifications - this should be sufficient I think.

I still may upgrade to a styrofoam sphere sometime in the future for increased accuracy, but mainly curiosity if I would see any noticable improvement.

Also, I feel that I need to chose probably two lights that I used for this calibration, and keep them as reference lights. I feel it is important to check the reference light as a confirmation of the calibration before each use of the sphere.

13

Would you expect to see a linear or power correlation between lumens and lux in an integrating sphere?

With the data points that I have available, both options could be potential fits.

Under about 1500lumens, the maximum difference between the two fits is 30lumens, (4%) (at 750lm).

Over 2000 lumens, the difference is more substantial: 150lumens, 7%, and continuing to grow larger as the values increase.

14

It should be linear. Your sphere appears to work pretty well. The only question, IMO, is whether there is significant dependence on beam width.

Your P60 quad measurement is consistent with djozz’s measurement that TIRs can lose around 30% of the total output.

Your C8 measurement, on the other hand, is not consistent with the 15-20% loss associated with reflector lights. One possibility is that wider beams illuminate the wall right around the lux meter and narrow beams don’t. This situation would cause an observed decrease in the measured lumens for narrow beamed lights. You could test this by measuring your C8 and moving the beam direction around by ~30 degrees or so. If the lux changes a lot then that could be your answer.

15

I should have pointed out specifically: sanding the interior led to a new calibration.

With the new calibration (the linear one, at your suggestion) I get:

  • 1308lm for the 4amp P60 quad, and
  • 1561lm for the fully modded C8.

Those numbers seem more reasonable to me.

I will double check, but it didn’t seem that sensitive to beam angle.