I don't know how efficient these power LEDs are... so approximately how much applied power is converted into light and how much is dissipated as heat? Almost all 10W is dissipated or only small portion of it?
- The large majority of energy in is turned to heat. Assume 100% for heatsink dimensioning. See below.
I cannot compute the required heat sink,
I have a 74 x 94 mm heatsink with 12mm high fins and half of the surface has place for a fan, so no fins there. (Summarised)
For practical purposes assume that all energy is dissipated as heat.
100% efficiency for white light is approach 400 lumen/Watt and an LED of that sort will be unlikely to be better than about 80 l/W[1] (and may be quite a lot worse depending on who made the emitters and when) so MAY be 20% efficient, could quite possibly be 10% and may be worse still.
The difference between needing 100% and 80% heat sinking =
10W / 8W = 25% more
so insignificant in practice.
For longest life you want lowest heat sink rise.
So say 30C heatsink rise and the less the better.
30C rise for 10W = 3 C/W.
Without a fan this is a very substantial heatsink.
With a fan this is doable.
You can easily measure operating heatsink temperature or even finger test.
If you can hold the heatsink indefinitely (say 10's of seconds) when running and stable then it is OK (about 55C max).
With LEDs cooler is always better for longer lifetimes.
[1] 2018 Edit: These days the efficiency of white LEDs is typically 100lm/W or better; some LEDs like STW8Q14D-E3 can archive 200lm/W or more (depending on color temperature und binning). Thus it might be reasonable to select a smaller heat sink, especially for space constrained applications.
Christmas 2018 :-) : I'll add to nqtronix's above update. The very best LEDs can now indeed achieve over 200 l/W, but you need to be extremely careful how they are operated to achieve this. The various graphs in the STW9Q14D-E3 datasheetthat nqtronix cited as an example allow you to explore just what these conditions are.
Efficiency drops with increasing junction temperature & input current.
Efficiency rises slightly with falling power input.
Power & temperature are not necessarily correlated as heatsinking alters their relationship.
Most of the above cited example LEDs specs are given at 65 mA drive. A non exhaustive look through the datasheet showed that you may reach 200 l/W output at reduced current and with good heatsinking with best case samples of best case binnings. At say 200 l/W and about say 30 mA and 2.8V forward voltage, power in is ~= 84 mW, light power out is about 40 mW and heat energy is around 40 mW+. The thermal resistance Tth-jS is 10 C/w so with 40 mW to dissipate the heatsink interface is only Rth x Pth = 10 C/W x 0.04W = 0.4C :-).
Adding a 10 C/W heatsink (large for this zie of LED in most applications) means that Tjunction is only about 1C above ambient. The given output graphs are mostly at 25C, so at typical room temperatures and modest input 200 L/W should be achievable in some cases. Even reducing the heatsink to say 50 C/W would result here in a junction temperature rise of a few degrees C above ambient.
It is a long question, but better than a short one, as you've shown your own research.
1) Solar cells. If you're stacking your own ones, stack 9 of them and get the 4.5V of the original circuit.
2) Battery charging. Batteries are the only thing you've left out of your spec. This is an area where the circuit design relies on cutting a lot of corners. In theory it might be out of spec, if you were to put 4.5V at 280ma through AA NiMH cells indefinitely. In practice, you don't get full sun all day, you'll be using it indoors, and you're not going to get optimal power transfer from the cells, so this isn't going to cause problems.
3) Diode. It's just a regular diode, not a zener. Current through it is actually determined by the battery and right hand side circuit, not the solar panel - the transistor is off when the panel is generating electricity. The original 1N914 will be fine. 1N4004 will also be fine.
4) Resistors: not a precision component here, use whatever meets your cost constraint. 5.1k for 5k is fine.
5) Wire: not critical. Your ebay link looks suitable. Thinner is better for the toroid.
6) Transistors: stick with the exact part numbers. Design may rely on specific parameters.
7) LED: again, this circuit relies on cheating. Normally a white LED won't run from two NiMH cells. The joule thief part provides a boost converter that gives small pulses of higher voltage. It doesn't have the capacity to provide a lot of current at that voltage. In combination with the pulsing this means there should be no risk of damaging it.
(A proper analysis of this circuit would be good, if nobody else supplies one I'll do it in a few days).
Best Answer
If you received actual 3 W LEDs and were to run them at 3.5 V, 700 mA without a heatsink, then yes, you will sooner-rather-than-later kill them.
You have three options.