White LED lumen testing

[Pinter]

It seems then that although I did not apply any correction factors to my earlier results the inherent methodology resulted in slightly wider beam angles which more or less compensated for the lower absolute lux readings.

jtr1962, great work as always.
Regarding the beam angles and profiles.
Finding out the center of a bare XR-E (or other) emitter is not an easy job as the beam center is not always in the real center. The beam profile gets somewhat inconsistent in the central regions, you can get different intensity curves from the estimated "0" angle, depending on the rotating direction or the orientation you attached the led to the heatsink.
I mean something like this (two thin lines are painting errors on my wall)

This can be a reason of the different beam profiles of the old and new XR-E P4 tests.
The real zero angle position can easily be missed with 1-2 degrees. This can affect lumen calculation also. If the same intensity values are belonging to greater angles, this results in more cumulative lumens.

For a quick test, you can try getting the profile towards the other rotating direction and average them.

[lctorana]

Pardon my ignorance, but WTF is that?

[MattK]

YES I CAN SEE THE DOLPHIN!!!

[Pinter]

Dolphin? Not really. :-) Sorry for not explaining it.

This is a "beamshot" of a bare XR-E emitter on the wall that was taken with a Pana FZ50 digital camera in RAW format. Horizontal field of view is about 53 degrees so it shows the central 25-40 degrees of the beam.

Raw data from CCD (linear intensity values) went under image processing and the intensity falloff was false-colored with ten colors that are repeated.
The min and max intensity range found on the pic was divided into 100 units. 0-9%, 10-19%, ... 90-100% share the same set of ten color, thus showing the intensity change of the beam.

The central section does not consist of the ideal circles, still shows something from the shape of the chip.

[lctorana]

So...

...what is the age of the tree?

[lctorana]

Hi jtr1962,

I've just found something exciting:

http://secure.oatleyelectronics.com/...c0a203da904333

The thought of 25 lumens without a heatsink is very exciting.

Have you seen these? If so, are they any good?

If not, would you like me to send you some for testing? If so, how many?

Cheers,

Bruce.

[jtr1962]

I added 2 samples ( Jeled 210000 mcd 10mm warm white, Jeled 255000 mcd 10mm white) sent to me by CPF member knabsol, 1 sample ( Oatley Electronics 80000 mcd 10mm white ) sent to me by CPF member lctorana, and 1 sample ( Deal Extreme 20mm white ) sent to me by CPF member nein166 to the list in the first post and updated the graphs accordingly. You might need to refresh your browser to see the updated graphs.

[jtr1962]

Cree 7090 XR-E bin R2 (acquired March 2008)

I borrowed an R2 Cree XR-E, bin WG, from CPF member nein166 for testing. The R2 bin is specified at 114 to 122 lumens at 350 mA. The color temperature of the WG bin is roughly 6000K. The results are as show below:







These results are nothing short of amazing! The output at 350 mA is nearly 122 lumens, well above any previous results for power LEDs at that current. Despite the middle of the road Vf of 3.31V, efficiency at 350 mA is still 105.3 lm/W. It remains above 100 lm/W past 400 mA. Even at 1000 mA, efficiency is nearly 75 lm/W. Things get even more interesting at low currents. Under 50 mA, efficiency hovers around 145 lm/W. This represents a wall-plug, or power-to-light conversion efficiency, of around 45%.

Output scales with current in pretty much the same manner as other XR-Es I've tested. At 1000 mA output is over 270 lumens. It approaches 400 lumens at 2000 mA. Cree has continued to raise the bar for LED performance. While we won't see as dramatic improvements as in the past, Cree has continued to squeeze every last ounce of perfomance from its XR-E line of LEDs. I expect we'll have R4 bins and beyond by this time next year.

[MattK]

Awesome work as always!

[IsaacHayes]

Granted 2amp is 2x max current, but 390lm from a single normal sized die is awesome!

[Photon_Whisperer]

Yes evan9162, that's exactly how I'm doing it. For example, I take the relative lux readings for 0° and 5°, average them, multiply by the area in steradians, and then multiply by the intensity in candelas. After that, I'll do the same with the 5° and 10° values, etc, and add this to my running total. It's interesting that for narrow viewing angle (i.e. 15° or 20°) LEDs the traditional methods of calculating lumens in the main beam grossly underestimates the output. One can see why by examining the radiation pattern of one of these LEDs. Some light falls outside the main beam but this decreases in intensity rather quickly. Still, since the spherical area is larger as the angle increases, this spill still makes significant contributions to the output. After that you usually have a bright ring about 60° off axis. This shows up clearly in the graphs in my spreadsheets. While the intensity might only be 2% or 3% of the maximum intensity, it is spread over a large area and contributes a good 15% to 20% of the total output. Finally, a great deal of light actually makes it out the back because of reflection off the front lens. This could easily add another 10% to 15% to total output. I was honestly amazed by this myself although it makes sense. For example, using the usual way of figuring lumens, and assuming a 20° beam angle, I might have guestimated about 2.3 lumens for the Light of Victory 35,000 mcd LED based on it's peak intensity reading of ~24,000 mcd at 20 mA (the 35,000 mcd spec is at 30 mA according to the manufacturer). I usually multiplied this by 1.25 to account for spill outside the main beam, giving me a final estimate of 2.9 lumens. The actual measurements come out to 3.75 to 3.9 based on a few samples, or over 30% more. I guess I'll revise my adhoc mutliplier to about 1.65 to give a more realistic estimate.

I'll also add that since the dial is just screwed to the tester and easily removeable, I might make another one up complete with a heat sink and mounting area to test Luxeons, Lamina BL-2000s, and other power LEDs. My constant current source can supply up to 2 amps if need be. I can verify my methodology if I get lumen measurements in the mid 30s with my Q-bin Luxeons.

Nice work but may I ask why not just build an integrating sphere? You'll get much better precision & accuracy (especially with focused emitters), and it's much faster.

[jtr1962]

Nice work but may I ask why not just build an integrating sphere? You'll get much better precision & accuracy (especially with focused emitters), and it's much faster.

The method I'm using takes under 30 minutes and is very repeatable. For the very low volume of testing I do it suits me fine. And as a bonus I get beam profile information.

Calibrating an integrating sphere is the real problem. Besides that, all my attempts to build a practical device have been very sensitive to emitter placement. Move the emitter a few degrees, or a fraction of a inch in or out, and the reading changes by over 10%. Even worse, results aren't consistent between different types of emitters. A narrow angle emitter of x lumens will give a wildly different reading than a wide angle emitter of the same lumens. I'll agree if I'm interested in sorting a batch of the same type of LEDs for relative (not absolute) lumens then a simple milk-container type integrating sphere is the only practical way to do it. But building a home-made integrating sphere of reasonable (say 5%) accuracy which works for many different types of emitters is well beyond my capabilities.

If anyone has any ideas on how to make a decent integrating sphere to supplement my usual method I'm all ears. It needs to be accurate, repeatable, and more importantly able to deal with both narrow beam sources and those which emit over an entire sphere (i.e. many indicator LEDs have a significant amount of back scatter).

[jtr1962]

I added 3 samples ( BestHongKong 25000 mcd 5-die white, LED Tech 14000 mcd 100 mA white, and Nichia NSPS500GS-K1 white) sent to me by CPF member TMorita. You might need to refresh your browser to see the updated graphs. I'm also posting the results here:

BestHongKong 25000 mcd 5mm 5-die white (acquired March 2008)
These are BestHongKong's 5-die whites which have 5 dies and a larger lead frame to allow operation at 100 mA. The results were 6.0, 6.5, and 6.9 cd at 20 mA. I assume the 25,000 mcd rating is at 100 mA, so the corresponding results at 100 mA are 24.3, 25.8, and 28.1. The average is 26.1 cd, slightly above spec. Average beam angle was 34.7°, less than the specified 50°. Color temperature was around 7000K, and the beam was very smooth. Average efficiency of the three samples was a decent 72.8 lm/W at 20 mA and 52.9 lm/W at 100 mA. I was expecting somewhat better than that due to the 5-die construction. Apparently they're not using the most efficient dies for these. Average output was 4.07 lumens at 20 mA, and average Vf was a low 2.87V (expected due to the 5-die construction). Corresponding figures at 100 mA were 16.41 lumens and 3.10V. The 16.41 lumens comes reasonably close to the specified 20 lumens at 100 mA. Consistency between samples was excellent-outputs were 3.91, 4.14, and 4.16 lumens at 20 mA. Output scales very well with current as expected (output at 100 mA is 4.03 times the output at 20 mA). Overall these appear to be a decent, well-constructed LED capable of sustained operation at higher currents.



LED Tech 14000 mcd 5mm 100 mA white (acquired March 2008)
These are LED Tech's 100 mA whites which have a larger lead frame to allow operation at 100 mA, and probably also multiple dies. The results were 2.8, 2.8, and 3.2 cd at 20 mA. I assume the 14,000 mcd rating is at 100 mA, so the corresponding results at 100 mA are 11.0, 11.2, and 12.3. The average is 11.5 cd, somewhat below spec. Average beam angle was 77.1°, nearly equal to the specified 80°. Color temperature was around 7000K, and the beam was very smooth. Average efficiency of the three samples was a very decent 86.8 lm/W at 20 mA and an excellent 63.8 lm/W at 100 mA. These are better numbers than the very similar BestHongKong LEDs because they're apparently using better dies. Average output was 5.12 lumens at 20 mA, and average Vf was a low 2.95V (expected due to the multi-die construction). Corresponding figures at 100 mA were 20.06 lumens and 3.15V. Consistency between samples was OK-outputs were 4.71, 5.05, and 5.60 lumens at 20 mA. Output scales very well with current as expected (output at 100 mA is 3.92 times the output at 20 mA). Overall these appear to be a decent, well-constructed LED capable of sustained operation at higher currents.



Nichia NSPS500GS-K1 5mm white (acquired March 2008)
These are Nichia's latest and greatest 5mm whites. The results were 36.0, 39.2, and 39.8 cd at 20 mA. I think these are speced for 44,000 mcd, so the average of 38.3 cd is only slightly out of spec. Average beam angle was 13.4°, a little less than the specified 15°. Color temperature was around 7000K, and the beam was very smooth. Average efficiency of the three samples was a excellent 96.7 lm/W at 20 mA, so these take the new efficiency crown. Even at 100 mA efficiency manages to remain at nearly 50 lm/W. Average output was 6.15 lumens at 20 mA, and average Vf was 3.18V. Corresponding figures at 100 mA were 21.43 lumens and 4.34V. Given the rather steep increase in Vf with current, these don't appear to be multi-die as some here have speculated. Consistency between samples was OK-outputs were 5.75, 6.19, and 6.52 lumens at 20 mA. Output scales very well with current despite the apparent single-die construction (output at 100 mA is 3.48 times the output at 20 mA). This isn't much worse than the two types of multidie LEDs just tested. It looks like Nichia has another winner here, and we have a new efficiency champ.

[saabluster]

Awesome job as usual! I'm really glad you test these all the way up to 2A instead of stopping at manufacturers specs. I would be willing to donate an R2 to test as well if you can push the test even further. I want to see just what these things can handle. PM me if you are interested.

[jtr1962]

Awesome job as usual! I'm really glad you test these all the way up to 2A instead of stopping at manufacturers specs. I would be willing to donate an R2 to test as well if you can push the test even further. I want to see just what these things can handle. PM me if you are interested.

The new current sources I made actually go all the way up to 10 amps. When I build a pair of them a few months ago I was thinking of the future, figuring that the 2 amp limit of my old tester eventually wouldn't be enough. Turns out with the release of the P7 I was right.

Anyway, I'll be happy to test your R2 if you want. I'll basically push until either the output stops increasing or the thing blows, whichever comes first.

PM sent with my shipping address.

[saabluster]

The new current sources I made actually go all the way up to 10 amps. When I build a pair of them a few months ago I was thinking of the future, figuring that the 2 amp limit of my old tester eventually wouldn't be enough. Turns out with the release of the P7 I was right.

Anyway, I'll be happy to test your R2 if you want. I'll basically push until either the output stops increasing or the thing blows, whichever comes first.

PM sent with my shipping address.

I sent you a PM then realized others might like to know what is transpiring. So here is a copy of what I sent.

"Thanks for the quick response. My intent is to know the absolute limit of these things. This way there will be no guessing or wondering how much they can handle. Of course I understand this means I will not be getting the R2 back. I'm doing this in the interest of science.
As part of this "pushing the limit" I would like to see how it does with the best heat-sink we can provide it, be that solid copper or a heat-pipe, and solder it down directly. I can't remember what you're protocol is exactly when it comes to heat-sinking endeavors so if this is above and beyond what you normally do I can mount it for you and send it that way. Let me know."

[jtr1962]

I sent you a PM then realized others might like to know what is transpiring. So here is a copy of what I sent.

"Thanks for the quick response. My intent is to know the absolute limit of these things. This way there will be no guessing or wondering how much they can handle. Of course I understand this means I will not be getting the R2 back. I'm doing this in the interest of science.
As part of this "pushing the limit" I would like to see how it does with the best heat-sink we can provide it, be that solid copper or a heat-pipe, and solder it down directly. I can't remember what you're protocol is exactly when it comes to heat-sinking endeavors so if this is above and beyond what you normally do I can mount it for you and send it that way. Let me know."

Rather than respond via PM since my inbox is almost full I'll answer here. I'm using a pretty large microprocessor heatsink for my testing. It should do fine here. Normally I use passive cooling (heat sink gets warm but not hot at 2 amps), but for this test I was planning to use forced air cooling just to see what the R2 is capable of under optimum conditions. We should break 400 lumens before 2 amps, and possible 500 or more lumens past that. Maybe I could go even crazier with this and try thermoelectrics if the R2 survives the initial round of forced air cooling. The real limits here will probably be how much current the bond wires can take. Decent cooling should make junction temperatures a non-issue, even at well over 2 amps.

[saabluster]

Rather than respond via PM since my inbox is almost full I'll answer here. I'm using a pretty large microprocessor heatsink for my testing. It should do fine here. Normally I use passive cooling (heat sink gets warm but not hot at 2 amps), but for this test I was planning to use forced air cooling just to see what the R2 is capable of under optimum conditions. We should break 400 lumens before 2 amps, and possible 500 or more lumens past that. Maybe I could go even crazier with this and try thermoelectrics if the R2 survives the initial round of forced air cooling. The real limits here will probably be how much current the bond wires can take. Decent cooling should make junction temperatures a non-issue, even at well over 2 amps.

I've got to get some work done so I will talk to you again Sunday or Monday about this in more detail.
Anyone want to guess the amperage it blows at and the peak lumens? This is going to be real interesting.

[jirik_cz]

[ViReN]

Awesome work jtr1962

[chimo]

These are Nichia's latest and greatest 5mm whites. The results were 36.0, 39.2, and 39.8 cd at 20 mA. I think these are speced for 44,000 mcd, so the average of 38.3 cd is only slightly out of spec. Average beam angle was 13.4°, a little less than the specified 15°. Color temperature was around 7000K, and the beam was very smooth. Average efficiency of the three samples was a excellent 96.7 lm/W at 20 mA, so these take the new efficiency crown. Even at 100 mA efficiency manages to remain at nearly 50 lm/W. Average output was 6.15 lumens at 20 mA, and average Vf was 3.18V. Corresponding figures at 100 mA were 21.43 lumens and 4.34V. Given the rather steep increase in Vf with current, these don't appear to be multi-die as some here have speculated. Consistency between samples was OK-outputs were 5.75, 6.19, and 6.52 lumens at 20 mA. Output scales very well with current despite the apparent single-die construction (output at 100 mA is 3.48 times the output at 20 mA). This isn't much worse than the two types of multidie LEDs just tested. It looks like Nichia has another winner here, and we have a new efficiency champ.[/FONT]
[/FONT]

Nice work, as always. I have a question on the GS. The beam profile on this LED is asymmetric due to the rectangular die. Did you modify your testing to account for the different beam profile?

I think I put the two die question to bed here. Cheers,

Paul

[jtr1962]

Nice work, as always. I have a question on the GS. The beam profile on this LED is asymmetric due to the rectangular die. Did you modify your testing to account for the different beam profile?

I just checked the beam profile. The asymmetry is barely noticeable to my eyes. As it turns out, I took beam profile readings on the short side of the rectangle. If anything this reduces the efficiency numbers. Also note that most of asymmetry is in the initial ±20° but this portion of the beam only accounts for a little more than half the lumens. In other words, my failure to account for the asymmetry probably affected the overall results by less than 10%. It's likely then that the lumens per watt of these are in the low 100s at 20 mA instead of around 97.

I think I put the two die question to bed here. Cheers,

Yes, and increasing die size looks to be a good way to bump efficiency. Looking at the efficiency versus current chart, if Nichia used a normal size die then current density would be doubled, and efficiency at 20 mA would only be in the high 70s to low 80s. Conversely, as can be seen by the leveling off of efficiency under 10 mA, doubling the size of the present die would only result in an efficiency gain of around 10%.

Overall I'm just glad to finally see a jump in 5mm LED efficiency. The best ones were stagnating in the low 80s for the last two years. Now if all the lab improvements make it into production, we should see LEDs about 50% better than the Nichia GS fairly soon.

[chimo]

I just checked the beam profile. The asymmetry is barely noticeable to my eyes. As it turns out, I took beam profile readings on the short side of the rectangle. If anything this reduces the efficiency numbers. Also note that most of asymmetry is in the initial ±20° but this portion of the beam only accounts for a little more than half the lumens. In other words, my failure to account for the asymmetry probably affected the overall results by less than 10%. It's likely then that the lumens per watt of these are in the low 100s at 20 mA instead of around 97.


Yes, and increasing die size looks to be a good way to bump efficiency. Looking at the efficiency versus current chart, if Nichia used a normal size die then current density would be doubled, and efficiency at 20 mA would only be in the high 70s to low 80s. Conversely, as can be seen by the leveling off of efficiency under 10 mA, doubling the size of the present die would only result in an efficiency gain of around 10%.

Overall I'm just glad to finally see a jump in 5mm LED efficiency. The best ones were stagnating in the low 80s for the last two years. Now if all the lab improvements make it into production, we should see LEDs about 50% better than the Nichia GS fairly soon.

Well that's good news (higher efficiency). I just checked an ArcAAA that I modded last night with a GS and an ArcMania 40mA converter. The hotspot rectangle was about a 2:3 ratio. Cheers,

Paul

[saabluster]

I'm using a pretty large microprocessor heatsink for my testing. It should do fine here. Normally I use passive cooling (heat sink gets warm but not hot at 2 amps), but for this test I was planning to use forced air cooling just to see what the R2 is capable of under optimum conditions. We should break 400 lumens before 2 amps, and possible 500 or more lumens past that. Maybe I could go even crazier with this and try thermoelectrics if the R2 survives the initial round of forced air cooling.

My main intent with this is to see what it can survive with a reasonably doable(could actually be used in a flashlight) heatsink setup. I can not think of a better way of keeping the junction temps down than soldering the Cree directly to a heat pipe. This is also, I feel, completely doable in a flashlight. Here is a picture of exactly what I would like to give you.
This heat-pipe is of the flat variety and will allow me to solder the Cree down easily. The Heatpipe will then be soldered to a 1/8" piece of copper that you can then attach to your heatsink setup.

I don't have to tell you how much better heatpipes are than even copper. It is of the utmost importance that the heat that comes through under the die be extracted as fast as possible.
As I said I'm mainly interested with what we could actually do in a flashlight. As such I think that precludes active cooling. However it would be interesting to have the additional data just for our edification.
So here is my proposal. Run the test as you normally do but with the heatsink I discused. When you reach 2A and for every additional 100mA or so take a measurement without the aid of active cooling and then cool the heatsink and take another measurement. So that at each increment above 2A your will have two lumens measurements. An easy way to do the cryo-cooling without having to change the rest of the setup would be to use a "can of air" upside down and spray it onto the heatsink. I don't know if this is what you had in mind already. I don't think a peltier would be as easy or as effective to use here.

Thats it. Let me know what you think.

[nein166]

... I don't think a peltier would be as easy or as effective to use here.

Thats it. Let me know what you think.

jtr1962 tell him about your freezerator

[saabluster]

jtr1962 tell him about your freezerator

Yes! Do tell!

[jtr1962]

My main intent with this is to see what it can survive with a reasonably doable(could actually be used in a flashlight) heatsink setup. I can not think of a better way of keeping the junction temps down than soldering the Cree directly to a heat pipe. This is also, I feel, completely doable in a flashlight. Here is a picture of exactly what I would like to give you.
This heat-pipe is of the flat variety and will allow me to solder the Cree down easily. The Heatpipe will then be soldered to a 1/8" piece of copper that you can then attach to your heatsink setup.

I don't have to tell you how much better heatpipes are than even copper. It is of the utmost importance that the heat that comes through under the die be extracted as fast as possible.

As I said I'm mainly interested with what we could actually do in a flashlight. As such I think that precludes active cooling. However it would be interesting to have the additional data just for our edification.
So here is my proposal. Run the test as you normally do but with the heatsink I discused. When you reach 2A and for every additional 100mA or so take a measurement without the aid of active cooling and then cool the heatsink and take another measurement. So that at each increment above 2A your will have two lumens measurements. An easy way to do the cryo-cooling without having to change the rest of the setup would be to use a "can of air" upside down and spray it onto the heatsink. I don't know if this is what you had in mind already. I don't think a peltier would be as easy or as effective to use here.

Thats it. Let me know what you think.

The main problem with doing it the way you pictured is that I can't put the LED on my test jig to get the beam profile. In other words, I won't be able to get lumen measurements, only intensity measurements. And while it's true the heat pipe has very low thermal resistance, the real controlling factor will ultimately be the heat sink. The heat pipe won't make the heat sink "better". In fact, the heat pipe will add a small amount of thermal resistance into the equation as opposed to just mounting the LED on the heat sink as I always do. Heat pipes are used mainly to carry heat from tight spaces where larger heat sinks don't fit to open areas where they do. A possible use might be to carry heat from a microprocessor to a large passive heat sink on back of a PC case. Or in the case of a flashlight, to carry heat from the LED to the body. However, the size of the flashlight body is ultimately the controlling factor in how hot things get.

Anyway, I think the best idea for now is to send an unmounted, never soldered LED. You can send the heat pipe and plate setup if you want. I can determine the thermal impedance of it and see how much it would raise operating temperatures as opposed to just mounting the LED directly on my heat sink.

I'll definitely be doing two sets of measurements as you said-one without forced cooling and one with.

Anyway, those are my thoughts for now. If anyone sees any flaws with my thinking please let me know.

Yes! Do tell!

I'll upload some pictures of my "freezerator" later. I also wrote a description of it on another forum a few years ago, so I'll have to find my post and cut and paste the text here.

[jtr1962]

OK, here's the promised pictures of my thermoelectric temperature chamber (or the freezerator as nein166 calls it). These are from 2004. I've since added white LED interior lighting.

Overview of the entire chamber:



Here's the chamber part in case anyone is curious (inside measures 10"x10'x14"):





Multipaned plastic viewing window:



Cold sink (note temperature sensor IC on chamber ceiling):



These are copper heat sinks that I made. Altogether at full load they dissipate over 500W of TEC power plus up to 200W of pumped heat (much less as the temperature drops), for a total of over 700W. And they do so while keeping the hot side of the TEC only about 3° C higher than the tap water temperature. Since this is intermittantly used, I don't recirculate the water via an external air-to-air heat exchanger. Rather, I just throw it away, or send it to the garden in the warmer months. I only use about 30 gph, so it is more cost effective to do so.



Here's a picture of the control panel with unit in operation (note -55.63°F actual chamber temperature). The auxiliary display shows 47.6°F, which is the heat sink temperature (tap water in February is around 41°F). I use a two stage setup. 6 modules in the first stage cool the hot side of 3 modules running at lower current in the second stage. I had problems with condensation at one point, and some of my modules are performing a little less than optimally. Still, I can reach under -55°F in the winter months. I'm thinking of replacing the 70W first stage modules with 85W ones. This will probably let me reach roughly -65°F.



Those two large main power supply transformers are cooled by the copper plates with water tubes. You can see some large filter caps right under the transformers.



Here's a top view:



The large toroids on top are inductors for the power supply. The tubes you barely see in the middle of the picture cool the power supply MOSFETs.



The fan is just to ensure a slight airflow over the large black voltage regulator heat sink and the aforementioned inductors. It only comes on when the chamber runs at or near full power. With these two exceptions, everything is water cooled. The MOSFET cooling tubes are more visible here.



Here's the display boards shown from a front view (sorry about the blurring):



Closeup of the control panel:



The chamber can also heat by running the second stage modules in reverse. I need to limit my temps in heating mode to about 200°F because the chamber fans and some of the plastics used to line the chamber can't take much more than that. BTW, when under no load the second stage cold plates get down to about -80°F, and that numbs your fingers in a few seconds (yes, I touched it). Just for kicks once when I was testing I put a TEC on the second stage cold plate, and a temperature probe on the TEC's cold side. I covered it with insulation and let the whole thing cool down to -80°F or so. I then powered the TEC. Of course, the aluminum cold plate it was sitting on heated up pretty fast (a relative term since it likely never got above -20°F during this whole procedure), so any temps I reached wouldn't be maintainable, but I did see the temperature probe reach about -110°F before it started creeping up again.

Besides testing electronics, I did this simply as a fun, long term project. I started the chamber in 1995. It took me over three years to get everything working right. It started out as air-cooled but I couldn't get much under 0°F. By small increments I made refinements and got lower temperatures. This included changes in both the heat sinks and also in making the chamber leak heat less. My first attempt at liquid cooling didn't go well-I used aluminum plates instead of copper. This meant using epoxy instead of solder to hold everything together. Epoxy and water don't mix long term. Hence, water-damaged TECs. Also, scaling developed inside the channels which compromised the efficiency of the heat sinks.

Besides changes in the physical setup I tried two makes of TECs. The second make was much better in terms of lower temperatures and durability. I also redid the power supply to get the efficiency up to about 95%. And then there was the problem of maintaining the set temperature. You can't get stable temperatures by just turning the TECs on and off. I designed a PID (proportional-integral-differential) controller which cut TEC power just enough to maintain temperature as the set point was reached. The unit can maintain stable temperatures within about 0.02°F of setpoint. While the chamber uses well over 500 watts at minimum temperature, it uses much less when set to maintain higher values. For example, to maintain 0°F I only need about 40 watts. To hold -20°F will use maybe 100 watts. Operating as a refrigerator (32°F) uses just a few watts. Of course at higher settings I can cut my water flow from 30 gph to 1 or 2 gph since I'm removing far less heat.

BTW, commercial temperature range is -40°C to 85°C (-40°F to 185°F). That's what I usually need to test over. I'd like to be able to get the chamber extemes to match the military range eventually, which is -55°C to 125°C (-67°F to 257°F).

It's a shame that no better thermoelectric materials than bismuth telluride have been commercialized in the 40+ years TECs have existed. I've heard researchers have found a few promising candidates. Since my chamber only requires about 25 watts of cooling power to maintain -67°F in theory if a TEC operating at Carnot efficiency existed I would only need about 10 watts of input power. I could make such a chamber using simple air cooling as the heatsink would only need to deal with 35 watts. I'd even be happy with TECs running at half the Carnot efficiency. I could hold -67°F with ~20 watts, and approach -200°F with less than 200 watts. There is plenty of incentive to make better TECs to eventually replace compressor-based cooling. I think we'll see a major breakthrough within the next decade

[Gryloc]

jtr1962,

Wonderful project! That is some amazing work; I see that it took plenty of time and dedication. Can I ask what sort of insulation you used? I am familiar with Zircar's Microsil, but that is used in extremely hot temperatures. I made a small furnace that could maintain 800C from around 150W. It is too bad that Peltiers aren't as efficient as removing heat as it is creating heat. I, too, would be thrilled to see the technology advance further. You wouldn't know off the top of your head what the efficiency is for the better Peltier devices compared to the average compressor-based coolers would you? I was just curious.

So, are you able to mount an LED in that chamber to test how the latest efficiency LEDs fare in low temperatures (to see the percentage increase in brightness). Though it would not be the most impressive as it would be with an older emitter, it would still be nifty to see a Q5 or R2 XR-E driven at 2A in your chamber. Maybe even a wildly overdriven Seoul P7 (ooh at 8A). I still dream of having access to some liquid nitrogen for casual experimentation. I heard from someone somewhere on the CPF years ago that someone stuck a standard 5mm high brightness red LED emitter in some liquid nitrogen and it supposedly lit up like a road flare. Maybe that experience was based on observations from non-flashaholic eyes. I do not know what the conditions were, however, and how the LED fared in the extreme environment.

I understand that heat pipes are configured to have certain operating ranges (where either above or below the range would cause the vapor to freeze or vaporize before returning), but could you find one that could work in your cold environment? What if you could mount an emitter right outside the insulation (with the heat pipe going through the insulation if you do not mind), and mount a heatsink to the inside of the chamber? You might be able to tilt everything so you can get lumen measurements, but it is not a huge deal. Actually, can you take lumen measurements before hand, then use the change in intensity due to the extreme cold to estimate the lumen output? I thought that was how you took lumen measurements to scale according to current input. I was just thinking...

Thank you for sharing your mighty cooling rig. I must have missed the first time that you presented this beast. <(Brr, I'm cold!)


-Tony

[jtr1962]

Wonderful project! That is some amazing work; I see that it took plenty of time and dedication. Can I ask what sort of insulation you used?

Thanks for the compliments. This really was an "endless" project. I may even upgrade it as I discussed when I have the time. It would be nice to see -65°F, perhaps even -70°F if I'm really lucky. Regarding the insulation, I used 2 layers of 2" thick Celotex insulation board. This is basically polyurethane foam. Probably better insulations exist these days, but not in 1995. Also, it was readily available at Home Depot, and cheap ($20 for a 4'x8' sheet). I think the insulating value is pretty similar to Microsil. I heard evacuated stainless steel panels filled with aerogel are nearly perfect insulators but well out of my price range. The chamber is pretty well insulated. I think I figured the thermal impedance is roughly 3.3°C/W. In other words, to maintain -55°F in a 65°F room only takes about 20 watts of cooling power (actually the cold sink fan adds another 1.5 watts to that total).

It is too bad that Peltiers aren't as efficient as removing heat as it is creating heat. I, too, would be thrilled to see the technology advance further. You wouldn't know off the top of your head what the efficiency is for the better Peltier devices compared to the average compressor-based coolers would you? I was just curious.

I know there's a lot of R&D devoted to improving Peltiers but so far nothing has made it out of the lab. The efficiency of Peltiers relative to compressors depends upon the size of the system. For something like a room AC or a refrigerator compressors are probably 2 to 4 times more efficient at the temperature differentials involved. For small cooling applications like picnic coolers Peltiers are probably better because the the frictional losses in a compressor are a greater percentage of the total power input.

So, are you able to mount an LED in that chamber to test how the latest efficiency LEDs fare in low temperatures (to see the percentage increase in brightness). Though it would not be the most impressive as it would be with an older emitter, it would still be nifty to see a Q5 or R2 XR-E driven at 2A in your chamber. Maybe even a wildly overdriven Seoul P7 (ooh at 8A).

For cooling LEDs it makes more sense just to mount them on a cold plate rather than put an LED-heatsink assembly in the chamber. I actually did a test a few years ago with an amber Luxeon. Link to thread. There probably wouldn't be as much gain with whites but even a 25% gain in efficiency with an R2 translates into ~130 lm/W at 350 mA.

I still dream of having access to some liquid nitrogen for casual experimentation. I heard from someone somewhere on the CPF years ago that someone stuck a standard 5mm high brightness red LED emitter in some liquid nitrogen and it supposedly lit up like a road flare. Maybe that experience was based on observations from non-flashaholic eyes. I do not know what the conditions were, however, and how the LED fared in the extreme environment.

My guess is the thermal shock wouldn't be too good for it but I'm sure it could deal with the temperatures just fine if gradually cooled. There's probably limits on how much increase in efficiency you'll get at low temperatures. The amber Luxeon was already failing to show much improvement under -40°F.

If Peltiers improve significantly then we should be able to reach liquid nitrogen temperatures with a two-stage device. I'd be happy even to reach dry ice temperatures.

I understand that heat pipes are configured to have certain operating ranges (where either above or below the range would cause the vapor to freeze or vaporize before returning), but could you find one that could work in your cold environment? What if you could mount an emitter right outside the insulation (with the heat pipe going through the insulation if you do not mind), and mount a heatsink to the inside of the chamber? You might be able to tilt everything so you can get lumen measurements, but it is not a huge deal. Actually, can you take lumen measurements before hand, then use the change in intensity due to the extreme cold to estimate the lumen output? I thought that was how you took lumen measurements to scale according to current input. I was just thinking...

No need to go through all that. I just mount the LED on my test jig, do a beam profile, and then get absolute lux readings from 1 meter to calculate lumens. Once that's done, I can just take relative lux readings at different currents/temperatures to see what effect that has on lumen output. I'm sure the heat pipe scenario you mentioned won't work well at all. Far better to just mount the LED right on a cold plate.

Thank you for sharing your mighty cooling rig. I must have missed the first time that you presented this beast. <(Brr, I'm cold!)

I never talked about it in detail here. I did on a few other forums I used to frequent but that was 3 or 4 years ago. Most of the description above was just a cut and paste of what I wrote on the other forums. Glad you enjoyed it! Now go get yourself some hot coffee.