All Over But the Funding: Open Hardware Spectrometer Kit

New submitter mybluevan writes "The Public Laboratory for Open Technology and Science is putting together an open hardware spectrometer kit on Kickstarter. The kits are built using an HD webcam, discarded DVD, and a couple other odd bits. They've also put together a kit for your smart phone and open-source software for desktop, Android, and iOS. Need to analyze the contents of your coffee, the output of your new grow lights, or a distant star on a budget? Just build your own spectrometer, or pick up the limited edition steampunk version." Besides making cool hardware, they'd like to "build a Wikipedia-style library of open source spectra, and to refine and improve sample collection and analysis techniques. We imagine a kind of 'SHAZAM for materials' which can help to investigate chemical spills, diagnose crop diseases, identify contaminants in household products, and even analyze olive oil, coffee, and homebrew beer."

Re:Pretty cool ...

[dns_server]

One of the steps of building the kit is removing the IR filter from the webcam.

I freeken love this.

[Anonymous Coward]

I did my PhD on a topic that involved building some gear related to spectroscopy. A good deal of my work involved cludging together optical parts from seemingly unrelated sources. As such I just love this sort of thing. Unfortunately I have to agree with the above commenter that the precision is just not going to be good enough from a webcam for real chemometric measurement. However with a little more investment the precision could be improved. The real difficulty will come from the need to calibrate the thing. You need to know exactly where each wavelength of interest is centred ** exactly** on the sensor at the time of measurement and this is not a trivial task. Unfortunately proper calibration is so often just not achieved... and that's not even beginning to consider the issues of background/ stay light, Johnson noise etc. I suppose this design could be used for crude detection like the article says rofl.

Lots of work has been done here

[Anonymous Coward]

Lots of people have been working in this field. The most impressive results are achieved by the astronomy community. link [stargazing.net] It is possible to produce a home made spectrometer that gets useful results. Some of these are capable of resolution sufficient to identify chemicals. These are sophisticated and often use a peltier cell to cool the CCD in order to reduce noise. link [fzu.cz]

I did a project whose aim was to produce a cheap spectrometer to match paint colors. link [conestogac.on.ca] The problems I found were:

1. Cheap webcams are quite noisy
2. Cheap webcams are not at all linear
3. For dark colors, sensitivity is a big problem
4. The spectrum of the light source varies depending on which angle you view it from.
5. Organizing the data is perhaps the biggest problem of all

My own engineering trade-off was sensitivity vs. resolution. To get spectra for dark colored paints, I widened the slit which reduced resolution. That, as far as I could tell, was reasonable because I wasn't trying to identify chemicals and the spectra from paints weren't particularly sharp.

The folks in TFA have a site where people can upload spectra. That's fine but a huge database of spectra is not too useful. The spectra have to be organized somehow. Here's an example [utexas.edu]. In fact the problem can be quite daunting [boingboing.net]

Re:Pretty cool ...

[thegarbz]

I have converted consumer junk to remove IR sensors on many devices from simple webcams to high end DSLRs. It is simple.

The hardest part about the process on cameras is working with tiny ribbon cables and sourcing replacement optics if you wish the resulting camera to operate again (since removing the filter completely will result in your camera becoming nearsighted.

Neither of these are a problem with webcams, and the only issue you're likely to have is that some webcams have their filter glued onto the CCD rather than just placed on and retained with a small amount of pressure from the case.

Re:I freeken love this.

[Anonymous Coward]

This is the same AC that you replied to. Yeah, every time one little part got changed on our $100,000 units the whole thing had to be re-calibrated for response factors. Then they would break down all the doggam time. Fibres would break or a microchip might come unseated and I wasn't allowed to touch the insides. Not to mention the cost of the reference reflectance tiles being a little bit more than the average bathroom ceramic.

As for the crackpots and bad calibrations, I thought my head would fall off from shaking it in disbelief at some of the results people touted as 'significant'. Just because data can pass a t-test doesn't mean that R^2=.4 is useful(!) Especially when there's no future proofing the the method. And don't get me started on some classification results. Poor linearity is just not a good excuse to revert to just setting threshold.

Cringe-worthy

[threeplustwo]

I don't want to sound all negative here, but... I don't have a choice, do I? A visible light spectrophotometer will not "detect toxins", no matter how much you try to make it open-source or crowd-sourced. The very concept of identifying compounds by visible light absorbance is very much flawed. Thing is, *most* molecules will not absorb visible or near UV light in a way that is specific enough. Real Chemists (TM) traditionally use the so-called IR fingerprint region for this purpose. This region is from approx. 700 to 1500 cm-1 (about 6 to 20 uM - that is 6000 to 20000 nanometers). A special detector is needed for these wavelengths. The one we have in our lab is cooled with LN2 and costs south of $15K. We also have a UV-Vis spectrophotometer, which has its own purpose. That purpose is not "identifying toxins", or analyzing any unknowns. Now, on to my next point. Identifying molecules is challenging, because they are very, very, very mindbogglingly small. Chemists have been grappling with this challenge for a long time. There are many spectrometric methods out there, including IR and UV-Vis (briefly discussed above), near-IR (900 to ~1800 nm, useful for *some* fingerprinting), and NMR (60-1000 MHz, very informative, bit needs a BIG magnet). Spectral data for many molecules of interest has been compiled into readily accessible databases, and is easily accessible. Some of the databases are proprietary/pay-per-view: https://ftirsearch.com/features/libraries/sea407.htm [ftirsearch.com] Some are semi-public: http://riodb01.ibase.aist.go.jp/sdbs/cgi-bin/direct_frame_top.cgi [aist.go.jp] And some are government/public: http://webbook.nist.gov/chemistry/ [nist.gov] The people who started this project do not seem to grasp of the very basic concepts of chemistry, nor did they do any research on the subject. Reading a Wikipedia article on UV-Vis would have been a good start. What is even more disconcerting is that the fundraising effort behind this cardboard spectroscope has been a success. One just has to hope that nobody buys this to screen their food for "toxins", or to teach their kids chemistry.

Re:Cringe-worthy

[jywarren]

Hey threeplustwo -- actually there is pretty good literature on laser fluorescence spectroscopy of polyaromatic hydrocarbons in the near-UV to visible range, you should check out some of the Public Lab research notes on the subject: http://publiclaboratory.org/notes/warren/7-18-2012/fluorescence-oil-spill-residue-diverse-spectrometer-use [publiclaboratory.org] with longer exposures we are able to get a clean read on the fluorescing spectrum. And even in the shorter term (before these harder uses are better developed and more rigorous) there are plenty of applications that are already feasible and useful. Check out the use cases highlighted in the KS Updates -- one guy used it to detect brighteners in laundry detergent, others are using it to empirically test grow lamps in aquaponics, etc etc.

Re:Blood sugar test?

[threeplustwo]

Non-invasive glucose measurement is one of the holy grails of portable spectrometer design. As it happens, it is most practical using NIR and IR. There are products that will be on the market soon. The kit from TFA won't do it, of course. Spectral resolution is too low, and it looks at all the wrong wavelengths.

Re:Pretty cool ...

[Anonymous Coward]

The bit depth of the sensor is more or less irrelevant. Spectral accuracy is a function of their diffraction grating and the size of the sensor: it's a spatial effect, not a detection/encoding one.

The bit depth does matter for sensitivity and dynamic range. For sensitivity, the dark noise of the sensor is also important, since you can effectively average for a very long time.