Quote from: wattMaster on March 06, 2016, 09:50:59 AM

The problem comes when you need to cool the LED. 

C'mon, that is not that difficult, unless you do something impossible.

With LED modules you should never forget the rating means really the absolute maximum, defined as (loose citation from ISO)
"ABSOLUTE MAXIMUM RATINGS are limiting values
of operation and should not be exceeded under the worst
conditions. These values are chosen to provide acceptable
serviceability of the device. The equipment manufacturer
should design so that initially, and throughout life, no
absolute maximum value is exceeded. If exceeded, even if
the device continues to operate, its life may be considerably
shortened."

In real life and practical designs it means with most LED's the usable power loading is about half of the rating or even less.
With IR LED's, the limiting factor would be the current capability of the bondwire. Fortunatelly the cathode uses to be the substrate, so the only bondwire is on the anode side and there the current is flowing from the bondwire into the chip, what makes the electro migration somewhat slower.

You need a diffuse light, that makes the thing easier: If you need 100W, use 4x "50W" modules or even better 8x "25W" ones and feed them by the 100W total.
Or you may just get the big lot of the 5mm LED's (they are among the cheapest ones, yet offer some output focus) and connect them on a common board. With that keep them with at least 10mm distance from each other and keep the package at least 8mm  from the board to allow thermal expansion without causing strain on the LED assembly (not respecting this is the main reason for early failures with most assemblies using this LED format).
If you do not want the narrower beam, then go for some SMD format, like 5050 or 2016 or so. Better stay away from the "0805" or "1206" or similar resistor-like formats - they do not offer anything to compensate for thermal expansion, so the soldering tends to crack. Even with these keep them 10mm apart (the 2016; the 5050 even further, because they are of higher power). With this you will get the power input of about 0.2W per square inch of board space and that should be handled well with the LED's/board itself, without any explicit heat sink.

Well, I was thinking of an array of 100 of those 100W infrared LEDs with no space between together, And driven at full power(Absolute maximum rating).
Everybody I have seen drives 100W LEDs at 100W.

Everybody I have seen drives 100W LEDs at 100W.

Then everybody is then surprised to see it failing after short time and then spitting about "trashy LED's"...

And good luck with the 10kW heater, keeping the bases at 25degC...

What about incandescent source driven to  the right temperature + filter ?

What about incandescent source driven to  the right temperature + filter ?

I was just thinking of that, But it would not need a filter, Because infrared is out of the visible spectrum, The bulb would be heated to not glow, But emitting infrared.

There are also lots of bad 100 Watt LEDs with uneven chips. Here's a good video of bad ones:
https://www.youtube.com/watch?v=NjKgPLeJ79Q

And someone made a 1Kw LED flashlight:
https://www.youtube.com/watch?v=-JVqRy0sWWY
The good thing is, He says it can only be used for about 1 min. to keep the LED's cool.

An overkill solution for cooling 10Kw of LEDs would be phase change. 

It would still need a filter i think - If you want the peak at wavelength X, then the "tail" of the spectrum will go higher and might cover the visible light. 740- nm thats allready well visible red, but 800nm range may still be visible. On the other side no filter is perfect either, so if you can suffice with imperfect filter, the 800nm range visible light maybe won't be as bad problem either

I like Clive's videos

LEDs are non linear loads. The higher voltage you apply to the LED array, the more LEDs will light up, and (sorta..) clamp the voltage to the diode forward voltage - Making the effect of resistive leakages less significant, and evening out the heat dissipation per diode. As long as you dont overdrive the harder-driven (i.e. the better) diodes ino the array, i think it would be still fairly useable - i.e. at a power level at which all chips allready light up more or less evenly but not yet full power

At 1KW (actually allready at much less than that) I would go with inverter and 1KW HPS..

I like Clive's videos

At 1KW (actually allready at much less than that) I would go with inverter and 1KW HPS..

Me too!

There are also lots of bad 100 Watt LEDs with uneven chips. Here's a good video of bad ones:
https://www.youtube.com/watch?v=NjKgPLeJ79Q

Definitely this is not about die matching, but about dies with defects in the active area.

It makes a feeling these are actually assembled from chips rejected as deffective from normal production. I see no other possibility for so many of them being dark (normally without any testing I would expect the defect rate in percents, so less then 5 pieces from the complete 100xLED array).
The background of the very low price of some e-bay offerings...

Or someone had destroyed them (by "testing" without proper heat sink and so the LED's get overheated. The rather compact patches of dark LED's would suggest that area being the hottest.
Or even high inrush current from the output capacitor of some ballast - Leaving the ballast powered and connecting LED's to it, instead of connecting/disconnecting the LED's only when the ballast is not energized. The consequence is, the ballast charges the output capacitor to it's open circuit limit, but connecting the LED's means unlimited discharge current.
So morons attempting to "test everything prior delivery" without actually knowing what they are doing.

Consequence of all of these is an extra leakage (demonstrating itself as a resistor parallel to the affected LED), but usually not that noticeable when supplied close to the rated current. Even when it may appear to be not important for the functionality (and with this state it really will be not important), the leakage will most likely worsen over time, so it will fail way earlier than the rated life and that is a real reliability problem. Regardless of what is the cause of the defects, if it is manufacturing or some damage afterwards.