Light filter around 592 nm

[beerbaron]

Hi, we currently have a wide range of LEDs (582 to 602 nm) we need to keep only the ones around 592 nm ±1 nm. This is very hard to see the difference by eye (selection will be made in china so worktime is not an issue). So what I tough was that we should order a light filter that will only let pass the right wavelenght.

Lens should be 10cm x 10cm.

Am I aiming in the right direction ? Or should I use another method?

Thanks
BeerBaron

 

[IRstuff]

+/1 1 nm requires a interference filter, which is not cheap. Additionally, though, since LEDs will have wavelength drift over temperature of something like 0.3 nm/ºC, specifying such a tight wavelength seems to be problematic.

Moreover, given the wide range you've specified, it would seem to be a very expensive cherry-pick.

TTFN

 

[beerbaron]

Yes it seems like it's going to be expensive... would it still be in the 1k-2k range or am I dreaming pink elephants here?

So what range would be a cheaper filter technique?

BeerBaron

 

[IRstuff]

For small filters, not that expensive, but the transmission is pretty horrible:

http://www.knightoptical.co.uk/acat...rsNarrowbandrange3nmbandwidth25diametermm.htm

While is possibly reasonable, cost-wise for the filter, the overall concept still seems problematic. Your products have an overall range of 20 nm, for which you're screening for even 3nm, is 1/7 of the total product, assuming uniform distribution. What are you going to do with the 86% rejected parts?

Why do you need such a tight wavelength for LEDs?

TTFN

 

[beerbaron]

Well, let's just say that these LEDs are use as a fire simulation and that "Green" or "Red" fire is not very realistic.

We woudln't lose 87% of our LEDs since it's probably more a gaussian distribution. We buy the LEDs from a manufacturer that claims 592 ±.5 nm but it's far from the truth since I can find 2-5% of completely different ones.

We are willing to throw away 5% of our LEDs without too much problems.

Regards
Frederic Boucher

 

[IRstuff]

If that's the case, why not use LEDs in the 630 nm range?
Not sure what your operating temperature is, but a 20 degree downward shift would shift the wavelength well below your acceptance criterion.

TTFN

 

[IRstuff]

Some more thoughts. Assuming a wrapped labor rate of $30 per hour and 3 minutes per diode tested and 33% fallout, you'd effectively add $4.5 to the cost of each diode, which seems to be quite a bit.

Moreover, unless you run a filter/no filter comparison, you can't be sure why an LED fails OR passes.

Overall, I'm unclear why you have an upper limit. Presumably, 602 nm should be a perfectly acceptable red.

I'm not even clear why 592 is necessarily bad. Aren't your LED potted in red plastic to begin with? Again, I would suggest that you get LEDs that have a distribution farther into the red, just to avoid this whole exercise.

TTFN

 

[beerbaron]

I know that going to a redder LED would simplify my life. As a matter fact I plan on building a business case with arguments toward a longer wavelenght but as you probably know... the engineering department is often the slave of the sales department. So I got to consider all options.

LEDs would be processed with a breadbord, 100 at a time (15 minuntes to load a breadboard). It's being done in a factory in China so it would be more like 0.001$ per LED from a workforce point of view.

Regards
Frederic Boucher

 

[cev]

You should realize that the interference filter test may be difficult to pull off with untrained labor. The center wavelength of interference filters varies like cos(theta), where theta is the zenith angle of incidence. You get a 1-nm shift for 3.3 degree misalignment and 3-nm shift for 5.8 degree misalignment.

Because of this, the geometry and spatial extent of your proposed breadboard may be a critical factor.

Curtis

 

[IRstuff]

Since, presumably, deeper red is acceptable, you might consider finding a cut-on fliter, i.e., a filter that passes everything above 590 nm. This may still be challenging, because the human eye's logarithm response makes it a terrible light meter, so devices that might look acceptable with the filter may still have too much green.


As another alternative, I would suggest a machine vision camera with some software to electronicaly look at ratio between the red and green channels. This could potentially make it fully automated and remove the subjective decision making required from the operator.

TTFN

 

[beerbaron]

Thank you both for your inputs.

Curtis, I never tought of that problem but it is a real one if I go with a filter. I would need to make some kind of strict breadboard.

TTFN, both idea are good. A machine vision would be better implemented at the LED manufacturer since it could represent big investments.

Is a lowpass frequency filter potentially easier to make ?

Best regards
BeerBaron

 

[IRstuff]

It's EASIER than a narrow bandpass, HOWEVER, making a few nm-wide transition band is non-trivial:

http://www.newport.com/store/genproduct.aspx?id=141133&lang=1033&Section=Spec

or

http://www.newport.com/store/genproduct.aspx?id=141135&lang=1033&Section=Spec

probably OG590

TTFN

 

[beerbaron]

With a low pass filter would I still a frequency shift if the incident angle is not 90 degrees?

BeerBaron

 

[IRstuff]

It'll be less sensitive because it wouldn't be interferometrically based. The typical wavelength filter is done by playing with film thicknesses, which are less sensitive to angle changes.

TTFN

 

[MikeHalloran]

Rather than manually plugging the LEDs into a 10x10 array in a test fixture, and then trying to evaluate which ones seem 'on' through a filter over which I have to scan my eyeball to retain a normal incidence, and then remove them from the fixture and pack the good ones in ... I don't know what,

I'd try to set up the system so the LEDs are delivered to the inspection station on tape and reel, inspected while still in the tape, then rolled up and sent on to the assembly station, with the bad LEDs missing from the tape.

I.e., inspecting them one at a time, quicker test duration, no manual handling, better net throughput. I wonder if someone already makes and sells such equipment, or something adaptable to do this job?



Mike Halloran
Pembroke Pines, FL, USA

 

[IRstuff]

I think the point was that labor in China was absurdly cheap, while capital equipment is as expensive here.

TTFN

 

[beerbaron]

Exactly, and implementing an automated system would make more sense if we would be producing the LEDs but we are buying them. In a bag not even on a reel so can't imagine how expensive it could be to create such an automated system.

BeerBaron

 

[IRstuff]

Well, my druthers would still be the camera approach, because I don't think that filter technology alone can give you a sufficiently clean go/no-go decision, particularly with the human eye. Lighting, filter slope characteristics, filter cuton repeatability, operator proficiency, glint, etc. will all contribute to inaccuracies and non-repeatability of the process.

While the camera/vision software approach requires some development, the approach would be substantially more robust and repeatable.

TTFN

 

[MikeHalloran]

I wasn't thinking of a fully automated system, but I did assume the parts were on tape. Given that, you could rig up something similar to an old film editor, manually indexing the tape through a single test station. The test station would need an unequivocal go/nogo indicator, because that cheap labor is also unskilled.



Mike Halloran
Pembroke Pines, FL, USA