Diminishing returns for overdriving Leds

Prompted by the comments in Bike Light 20W+ 7 Leds
I ran a spreadsheet on running 6 x 3Watt Luxeon leds at 700mA versus running them at 1A for a fixed size heat sink (1.11 C/W resistance)

The results are
1A 14.15% more light for 51% more power
versus 700mA

The spreadsheets takes into account the loss in output due to increased junction temperature and the fall in Vf due to junction temperature.
The spreadsheet assumes a constant current drive of either 700mA or 1A
and a 25C ambient and a Vf of 3.5V at 700mA 25C

Based on this result I will be turning my Tesla down to about 700mA per led
matthew

Yup, overdrive the lm/watt goes down.

However, in a single led light that needs to throw, then it is sometimes accepetable.

I've done the same calculations based purely on datasheet information and got similar results. However, once I got a light meter and actually bracketed the output vs. current, I found that output scaled much better than the datasheet information and calculations indicate. In my real world testing, I measure about 25% more output going from 700mA to 1000mA. That was using a star, which adds some thermal resistance. Bare emitters mounted to the heat sinking surface will result in even better current scaling, around 27% or more.

Did you let the leds reach steady state temp. before you measured the output?
I noticed in my Tesla it takes more the 15mins to reach steady state due to the thermal mass of the head.
The figure of 27% looks like what the Luxeon III output versus current graph reads at 25C junction temperature, but you are still pumping 50% more power even for this result.
matthew

Yes, it is very important to include the head temperature in the numbers, as it is reality.

Good to see some folks are beginning to finally understand this.

Yes, they reached as steady state as they were going to - the star was attached to a heat sink with a fan, holding it steady at room temp. The junction temp has reached its operating temp - it only has a thermal time constant of around 20ms, so any time after that it will only increase in temperature if the main heat sink increases in temperature. At 700mA, Tj=58C; at 1A, Tj=77C.

In my spreadsheet I assume natural convection only, not many torches are force cooled with a fan. Assuming the head gets to 20C above ambient (ie 45C perhaps a little optimistic) then 700mA Jt was 72C, 1000mA Jt was 96C

However, this is a difference of 24C between the 700 and 1000mA while your measurements were 19C difference. 24C to 19 is to close to explain the differences between calculation and your measurement.

So either you are getting more lm/W at higher currents then the spec says or the temperature effect is less.

Given the wide output variations on Luxeons within a given bin it may be that some are more efficient at higher currents then the spec suggests.
Not sure we will know without some controlled batch testing.

So in summary your measurements give us hope of getting 27% more output for 50% more power input (when fan cooled), while the specs only promise 15% more output for 50% more power input (using natural convection). Either way it is the law of diminishing returns.
matthew

Seeing your test results, I am going to increase my current setpoint from 700mA to 800mA

Now if I can just find a way to get rid of the 16Watts of heat...
The batteries (6xAA NiMH) run down in an hour or so, so I only need to dump the heat for a short time.
I could wrap a cold pack around the head to keep it cool. I might even get more out of the leds as the output increase below 25C.
I could freeze the cold pack while the batteries are charging.

I know!! Make the barrel 3D length and add a dummy D cell at the front filled with cold pack fluid.

Wait !!!! I have a charger socked in the torch so I don't remove the batteries
I could add another socket and use a thermopile to 'recharge' the cool cell while the other batteries are charging. Now that might just work.....
matthew

Isn't that 1.11C/W a way too optimistic? Calculation for my 20W light showed about 4C/W thermal resistance on natural convection (approx. 100g heatsink weight, black anodized surface). It went under 1C/W only when fan was used.

mpf said:

Prompted by the comments in Bike Light 20W+ 7 Leds
I ran a spreadsheet on running 6 x 3Watt Luxeon leds at 700mA versus running them at 1A for a fixed size heat sink (1.11 C/W resistance)

The results are
1A 14.15% more light for 51% more power
versus 700mA

Click to expand...

Now that we have bright LEDs I think efficiency is the main point to be worked on, which is why I tend to order light sdriven at spec or even underdriven ... as the gain in runtime outweighes the little brightness lost in most cases. This is especially true for the heat-suffering LuxV IMHO.
bernie

Pinter said:

Isn't that 1.11C/W a way too optimistic?

Click to expand...

Yes it is. That is what I would like to have.
At the moment it is more like 1.55C/W and the head and body get up to 53C at 25C ambient after about 20mins. Thankfully by then the batteries are down a bit and it levels out and then drops off.
I used 1.11C/W to be generous and try and calculate the 'best' result.
matthew

Did you measure the Vf drop after 20 minutes? You can roughly calculate the die thermal difference from that. Just courious of your data.

Pinter said:

Did you measure the Vf drop after 20 minutes? You can roughly calculate the die thermal difference from that. Just courious of your data.

Click to expand...

I did a second test to measure the voltage with the head removed from the barrel.
Using -2mV/C the differences after 20min the leds were between +88C and +147C compared to starting temp (25C)
Not sure this is a very good estimate of the junction temp compared to thermal resistance and wattage calculations.
Temp was 62C and the wattage was 15W for 6 Leds which would suggest a junction temp of 13C/W * 15/6W + 62 = 92.5C
The head (separeted from the barrel) had not reached steady state but this should not effect the calculations except for the time constant of the probe.
Assuming 2min for the probe time constant add about another 3C to the total.
So about 95.5C well below the 113C to 172C calculate from the voltage drop.
matthew

Starting at t=0.1 sec the led die is not 25C but 25C+13C/W*Power. At t=0.1 sec only the heatsink that holds ambient temp.
Unless you are able to measure Vf within microseconds, the initial Vf corresponds to this higher temperature.

I've found the Vf drop vs. Tj coefficient to be unreliable at best. I've attempted to bracket that value using 10 Luxeon Star samples all from the same bin, using a temperature controlled peltier unit and constant current power supply. My results were all over the place, from 1mv/C, all the way to 4mV/C. I don't think measuring temperature based on any delta-Tj is a reliable method.

Pinter said:

Starting at t=0.1 sec the led die is not 25C but 25C+13C/W*Power. At t=0.1 sec only the heatsink that holds ambient temp.
Unless you are able to measure Vf within microseconds, the initial Vf corresponds to this higher temperature.

Click to expand...

Good Point. But that just puts the temperature estimates way off.
Allowing an initial 3W = 39C rise in the die gives final temps of
125C to 186C compared to estimated thermal resistance calc of 95C

If the -mV/C measurements are correct, I should have a number of dead leds, but I don't.

Kiessling said:

Now that we have bright LEDs I think efficiency is the main point to be worked on, which is why I tend to order light sdriven at spec or even underdriven ... as the gain in runtime outweighes the little brightness lost in most cases. This is especially true for the heat-suffering LuxV IMHO.
bernie

Click to expand...

Bernie, in your opinion, what's the "ideal" driver current for a LuxV?

mpf said:

Good Point. But that just puts the temperature estimates way off.
Allowing an initial 3W = 39C rise in the die gives final temps of
125C to 186C compared to estimated thermal resistance calc of 95C

If the -mV/C measurements are correct, I should have a number of dead leds, but I don't.

Click to expand...

Some notes.

1.) 125C - 186C die temp can be realistic. Probably will not kill your leds immediately, only affects their lumen maintenance.

2.) Leds are more efficient now. Although (a rather old) Luxeon Application Brief says that you should calculate with power Vf*If, in reality you can calculate with Vf*If*0,85. In 2005 for good bins at least 15% goes for light, rest is heat. This is a minus in your calculation, as only 12,7W heat is generated.

3.) You only calculated with the thermal resistance of the emitter itself. Thermal path consist of other minor paths. Such as emitter-to-heatsink (thermal compound), and heatsink-to-body thermal resistance. These can add another 1-2C/W thermal resistance. So this is a plus.

4.) The mass of your setup can also affect your assumptions based on 20 minutes data. In my simulation the mass of the heatsink was only 100 grams.
After 10 minutes it reached 70% of its maximum deltaT.
After 20 minutes it reached 90% of its maximum deltaT.

So maybe your calculated estimations are valid only for e.g. 1 hour time frame, not for 20 minutes.