DIY spectrometer

3 posts / 0 new
Last post
agenthex
agenthex's picture
Offline
Last seen: 5 years 10 months ago
Joined: 07/14/2010 - 04:49
Posts: 3070
Location: Merica
DIY spectrometer

Reading this makes you smarter: http://lesswrong.com/

Don
Don's picture
Offline
Last seen: 2 years 4 months ago
Joined: 01/12/2010 - 16:32
Posts: 6617
Location: Scotland

Interesting idea, but I can't see it being much use outside a didactic context. I would have killed for one of these when I had to teach science.

 

Calibration would be a huge pain. While the software extracts quantitative data, I'd doubt the value of it with the typically hideously noisy sensors on phones. I'd not even bet on the consistency of cheap CMOS sensors.

 

I once spent 3 months extracting cadmium from sewage samples into chloroform in a poorly ventilated lab, then analysing the resultant product by spectrophotometry. Despite severe chloroform intoxication when I ran the stats on the results, I've never seen cleaner statistics. I thought they were way too good to be true - you don't get +/- 2% errors with manual wet chemistry no matter how good your technique. Thanks to the chloroform, my technique wasn't all that good. But my boss got the same awesomely good stats.

 

So, as a result of our finding a lot of cadmium and cyanides and other nasty stuff in a sample from a sewer in a residential area, someone was going to jail. The "fingerprint" of nasties in this stuff was characteristic of an electroplating outfit pumping seriously dangerous waste down a residential sewer. Which would have meant tankering the stuff to the opposite side of town and pumping it down a drain in the middle of the night. I knew this because I'd spent 24 hours sitting at the end of a pipe (in a sewage works) taking samples every 30 minutes.

 

So we get in touch with the Water Board's lawyers to find out how to play this with a preliminary finding of seriously toxic material being tankered across town and pumped down a residential sewer.

 

So the samples get divided into 3 parts. And I get to spend another 24 hours at the end of a pipe, this time with a witness present as i take and label the samples.

 

One part we analyse, one part got analysed elsewhere, and one part got kept for any legal challenges to the prosecution that was likely to ensue.

 

So we send off the samples. Unlike my beautiful error limits, these samples came back with essentially random numbers attached. So I mapped them against my data and they bore absolutely no resemblance to mine.

 

It was unfortunate that I'd done the samples in the same order each time as the results turned out to have been solely depended upon how long the spectrometer had been switched on for. And that was a $50,000 device that had just been recalibrated by Perkin Elmer who made the thing.

 

I've never been that fond of spectrometers since then!

 

The numbers from my light tests are always to be found here.

https://spreadsheets.google.com/ccc?key=0ApkFM37n_QnRdDU5MDNzOURjYllmZHI...

agenthex
agenthex's picture
Offline
Last seen: 5 years 10 months ago
Joined: 07/14/2010 - 04:49
Posts: 3070
Location: Merica

I opened up the executable quickly this morning. It looks like it takes a "sample" (measured) jpg vs a calibration one, which kind of only makes sense for the specific chemical composition measurements they're doing. Measuring output like we are would just use one of those.

 

However, the biggest problem is that calibration is just guesswork on the part of the user. You would think the SW could do it automatically if you calibrate colors against something since it would then know what color is which (ie which part of the spectrum) and center the plot accordingly. So the precision of the measurement is ok, but accuracy is going to be +-50nm or so.

 

The "cell phone" feature/name is just a marketing point.

 

When I planned for a spectrometer (and by plan I mean an excuse for wasting time on the internets) I was thinking of either doing the image analysis in VIPS or something, but that required actual work.

Reading this makes you smarter: http://lesswrong.com/