Really like the high quality design of the paper! €250 for a kit + shipping/assembly time does seem like a heavy price that would take a long time to pay for itself if not used in a centralized commercial facility. I like to imagine in the future we'll all have circular economy infrastructure built directly into open source hardware appliances in our living spaces, no need to 'take out the trash'
This looks really neat, although I share other commenter's feeling that the price was surprisingly high.
As a minor edit, the word "breakout board" is used in a non-standard manner I think. It usually applies to small pieces of circuit board that hold a single integrated cicuit or other component with many/small/fiddly pins, and route them (break them out) to typically 0.1"-spaced pins. This makes the IC or other component accessible for breadboarding and/or hand soldering.
A circuit board that holds multiple such components with tracks routed between them for connections is not a breakou4 board IMO, that's just a circuit board.
This is pretty cool. It looks like the cost of LEDs is by far the largest blocker to scaling this or making it a consumer product (if that's even intended). I suspect it's because they're esoteric frequencies are high precision. Is this precision needed to discriminate between different plastics or does it just increase the accuracy of a match?
i work with IR emitters (LEDs) (similar to this project but with several cameras), in the SWIR range a single LED can easily cost 10-20$. they tend to have very low efficiency at these wavelengths too, typ around 1mW/sr (120 degrees) at 50mA. a good SWIR setup with a camera and led illumination will quickly be 20-50k.
I guess what he is implying is, that the LED is the biggest cost factor on the BOM of such a product. If you have the goal of making such a product available to every household, would it be a feasible product if the bill of materials is say 40 USD already? How much would people pay for such a thing?
Of course in electronics everything gets massively cheaper with scale.
This is super neat. I always wondered why do the people have to 'sort trash' when computers or systems where this trash ultimately goes can do it better than even humans can. I assume at least.
I am aware that not all plastic is recyclable but still. I think the system of getting everyone to do a job that can/should ultimately be automated is so strange. Unless it's impossible to do?
Recycling facilities everywhere use tech like this (near infrared) at an industrial scale [0], and sorting robots that use AI is a rapidly growing industry [1]
I remember a few decades ago reading a documentary about PC cases shipped to facilities in southeast Asia where little kids would melt it with a lighter and smell the fumes to determine what type of plastic it was. Of course finding an article from the circa 2000 internet today is nigh impossible, but hopefully things have changed by now.
We're not even recycling ten percent (per [1]) of post-consumer plastic waste, and most of the "recycled" you see in advertising is hogwash - what is counted are scraps from along the manufacturing process that get recycled for that metric.
We should be focusing on actually enabling any recycling, not nitpicking about efficiency gains.
The biggest issue with plastic recycling is probably that products like packaging are optimized according to marketing criteria (colourful/shiny/coated) and adding a little bit of convenience for consumers (glue to reseal after opening).
This way things end up being made from multiple components which aren't easily separable, thus difficult to (if at all) recycle.
Making simple packaging where the whole thing is made from a single (widely used, standardized) variant of plastic and instead of being coated or multi layered, things like labels ought to be fixed in a way that allows easy separation (easily soluble glue).
But it's not gonna happen soon, since it'd basically come down to throwing away all those neat "innovations" the packaging industry has developed. Also, if there's one thing retailers care about it's products that look shiny and colourful and entice people to buy them. They're hardly going to give that up.
> Also, if there's one thing retailers care about it's products that look shiny and colourful and entice people to buy them. They're hardly going to give that up.
Wrap the plastic in paper. More and more products here in Germany do this.
This is interresting device, but i am not sure whether the concept based on measuring light reflectivity is valid, because same plastic can be molded in molds with different surface structure. May it be mold with mirror finish, or matt sandblasted finish. Also color pigments can be added into the plastic. These things highly affect amount of light reflected from the plastic. And i cannot find how exactly they do their measurements. Would be nice to get more insight.
Looks like it's reflectance spectroscopy. The specific polymers in the plastic will have signature absorption lines that will be there no matter the color or overall reflectivity of the surface, as long as enough light reaches the sensor.
This is great! I wish we could have wider band/multispectral sensors in regular phones by default, feels like there could be a bunch of use cases. Maybe too niche
Optical power of the illumination decreases with the inverse square of the distance. Optical power of the reflected light decreases with the inverse square of the distance.
The fluorescence in the plastic is not very bright (how bright?), and it's competing with ambient light (use at night?). Need to do everything you can to increase the signal to noise ratio.
I'm imagining two ways to tackle this.
1. Illuminate a wide area and use a camera sensor.
To compensate, you need very very bright light sources. It's going to be difficult to make a multi-spectrum, precise, very bright, lightweight light source. Probably strobe the lights to manage heat and so each exposure can compensate for changing ambient light. Probably need a way to measure outgoing light or otherwise live-calibrate based on the changing outgoing brightness.
For the camera you need something that doesn't have an IR-filter. You need to use IR lenses (not glass). Preferably use a scientific/astronomy sensor optimized for extremely low-noise and long integration times. The very low-noise sensors are not ideal for drone use (vacuum chambers + cryo cooling = large + heavy + high power). Probably also need a very high dynamic range, even if you're running at night. Use a filter wheel to swap which wavelength the sensor is looking at for each exposure.
Then you're going to need a lot of signal processing to handle the video. You'll need the drone to be extremely stable so you can stack long exposures, or use some night-photography exposure stacking AI, or both. The challenge is finding the tiny signals within all the noise.
2. Illuminate a narrow area and use a photodiode.
Use a set of lasers at different wavelengths and a photodiode+lens system which looks where the laser is pointing. Now you're looking at a tiny spot at a time and you need a system to scan the laser around. However, it's much easier to make a single very sensitive photodiode amplifier and you can focus your illumination in a smaller, more manageable spot. The signal processing is easier now, too.
Maybe just image processing from a regular camera feed is the way to go then. It's not like beach plastics are the same color as their surroundings in regular light.
Why stop there? Put a robotic claw on the drone and you have an autonomous plastic collection system. Then you can scale up to hundreds of drones and clean all the beaches
You could use a phone as the UI, but the detection requires emitting and sensing very specific wavelengths of light that phones don't emit or sense, so there'd still be a hardware component.
From what I know (having owned two Xiaomi Phones with the 'Remote Functionality') it is a basic IR LED (just like a universal remote). The LED is seated on the top of the phone (like a regular remote) so capturing that light with the camera for analysis would be quite incosistent imho. But worth a try?
* Thesis: https://repository.tudelft.nl/islandora/object/uuid:1fa997b7...
* Wikifactor: https://wikifactory.com/+plasticidentificationanywhere/break...
* EasyEDA Schematic and PCB (GPLv3): https://easyeda.com/jerzeek/nir-spectroscope-final-pcb