After musing about it for a while, I got some advice from Kevin, of Lambda Lights, on designing a copper heat sink for my LED E10 bulb project. Kevin's elegant solution to heat sinking is to solder the LED's thermal pad directly to a copper nub poking through a slot milled in the PCB. However, he ultimately concluded that there would be little point to machining pillars out of copper for an LED light bulb application as the heat has nowhere to go beyond the surface of the copper (ie. it is not attached to a larger heat sink or in contact with a housing or chassis of any kind). My enthusiasm and greenness beat out Kevin's sage wisdom and I went ahead and had a prototype copper heat sink made for a Cree XLamp XP-G. I had them machined by a company in Shenzhen that did a great job for a great price (less than one tenth what a couple of USA prototyping services quoted me). I reasoned that adding a few fins would aid in shedding heat to the surrounding air, so the design became fairly baroque:
|Copper LED heat sink for Cree XP-G. The fins add about 2 square inches of surface area.|
In a nifty kind of internet 'digital to analogue' conversion, my design went from virtuality to reality in 10 days:
For the first time I used Seeed Studio's (yes, there are 3 Es there!) Fusion PCB service, which turned out to be cheaper and faster than than BatchPCB, although the board quality doesn't seem to be quite as nice. The main impetus for using Seeed was that they can mill internal slots in their boards, whereas BatchPCB cannot.
I soldered the LED on to the board and heat sink with a heat gun, which proved to be a challenge; I quite literally blackened one side of the PCB and melted off the dome of the LED! The LED survived, although I don't know if its performance is at all affected by its rather rough (and out of spec) treatment. This experience prompted me to finally invest in a hot air rework station, which is currently in the mail.
|Cree XP-G soldered directly to copper heat sink. Overzealous heat gun use|
After a few days of admiring my creation, a couple of things arrived that actually allowed me to test its performance. After wanting one for years, I finally caved and got a Fluke multimeter. It's the 'Not For Sale Outside of China' 17B model available through DealExtreme. Genuine Fluke for a about one third of a similarly equipped US market model, although the warranty is automatically voided by export. Most relevant for this application though is that it has a thermocouple thermometer, so I can measure the temperature of the heat sink by direct contact. With a bit of thermal grease, I was able to wedge it into the space between fins and make good contact with the heat sink.
|Thermocouple on LED heat sink with thermal grease|
|New Fluke 17B multimeter!|
In the same package was a cheap lux meter to let me measure the relative output of the LED being driven at different currents at different temperatures.
|Rudimentary lux meter set up. Response time is impressive: meter is showing lux from the camera flash!|
My excitement was tempered almost immediately by two realizations: First, my design puts the LED way too far above the vertex of the parabola of the reflector, negating any advantage of using the original vintage optics. That was kind of a dumb oversight! Second, as Kevin predicted, the heat sink just gets hot; there isn't enough air flow to cool the heat sink sufficiently, even with the addition of fins. Running at 1A it takes about 20 minutes or so for the heat sink to plateau at almost 100°C. That is considerably hotter than my goal of keeping the heat sink at or below 80°C. Luminous flux decreases as the junction temperature of the LED increases, reducing efficiency and LED life.
I measured the lux at different currents, reading the initial lux as well as the lux once the heat sink had reached a steady state temperature. Readings were taken at 1 meter in an otherwise dark room (lux = 0) without any secondary LED optics (and a missing LED dome!). Ambient temperature was about 22°C.
As you can see, a higher steady state temperature results in less light output. About 20% reduction at 1A, 13% at 0.75A and 9% at 0.5A. So, I think there will be little point to driving these LED bulbs beyond 0.75A, which is about what an upgraded GH6 Dynohub can put out. Of course, I have to change the design to drop the LED down to the approximate level of a bulb filament so that it's in the optimal position within the parabola of the reflector. This necessarily will be a smaller heat sink (lower mass and surface area), so it will heat up faster but I expect the steady state temperature shouldn't be too much higher. Waiting for the new design in the mail to find out...