The machine is 140 x 45 cm and including the hopper 180 cm high due to the legs it’s on (to be able to put a bucket underneath). It’s weights 160 kg.
We were told it needs an air compressor that can deliver 200 l/min air, but now using it for a while we learned a far smaller and thus cheaper compressor of let’s say 50-100 l/min air would have been enough.
The price including shipping to the Netherlands and insurance was US$7100. Here we had to pay 110 euro import duties and about 300 euro for the paperwork. We bought a 500 euro compressor, but a smaller model of 100-150 euro would have been enough. So all together we’re about 7000 euro poorer. The smaller model would have been 1000 euro less.
So not an investment for coffee roasters or farmers in economically more challenging countries.
The service so far is very good. When we couldn’t work out some things we got twice live video support from China for 3-4 hours.
I hope the photo comes through. Its not the best photo …it’s me hiding behind it. And one that shows the air compressor…
@aqdennis that is amazing! I’m jealous! My wife’s business is not big enough to justify that investment. Maybe we’ll ship all coffee to you for sorting before it comes to us
Thanks for sharing the photos and info.
@matthias sorry to hear things haven’t all been going smoothly for you.
I know you spent a lot of time looking at equipment etc but I wondered if you ever came across this…
Works on micropython and tensorflow lite. I haven’t read all the details but it looks exciting to me!
I hope one day we can get a prototype of this thing going. Such a fun project.
There’s a Brazilian machine called Seletec ST-4 which can handle ~120kg/hour and costs about 5,300 USD (26,000 BRL). More details on their products page.
I couldn’t find photos of ST-4, but a seller guaranteed me the ST-12 has the same skeleton/size: 0.955m x 0.520m x 1.800m. I’m attaching some pictures and a brief tech spec I found (may help in design).
There’s also a video of a ST-6 operating (can handle ~240kg/h and costs ~6,000 USD).
Thank you for the pictures and the specs of that machine. As far as I know, none of us here has known it before. The design, esp. of the user interface, could provide us with inspirations for creating a simple optical sorter.
I managed to submit an application for the coffee sorter project to Prototype Fund Hardware yesterday, shortly before the deadline as usual If they fund this (9500 EUR), there will be enough resources to create a working machine and all relevant documentation. Let’s see … in about a month we’ll know more
Well… it seems to a bit late in reply but recently we launched our green coffee sorter and it operates a bit different from color sorting but utilizing image recognition technology instead. Currently, the sorting process works around 5~6kg/hour. The machine is quite compact itself, weighing at 50kg only. If anyone is interested in it, feel free to take a look…
Some of the key differences from color sorting is that we focus on Continuous Learning and Improvement:
Leveraging advanced data analysis and AI model to enhance its recognition accuracy over time with defective bean photos
Comprehensive Defect Identification:
Unlike color sorting machines, the AI image recognition coffee sorter conducts physical inspection bean by bean hence the carry-over rate of the good bean from the rejected outlet is considerably low.
Examines each green coffee bean with two 4K cameras from a 360-degree perspective, specializing in identifying defects such as subtle insect-damaged, fungus-damaged beans, broken beans that might be overlooked by color sorting machines.
Thanks @avercasso! So, your company did beat our project here for producing the first AI powered coffee sorter … you’re welcome Of course we’re interested in some more tech details, if you are free to disclose them. Especially the input size of single-bean images fed to the deep learning image recognition algorithm (assuming you mean that by “AI”).
Also, a few suggestions that come to mind already:
Your machine’s product page says the machine has programs for “Natural, Wet, Wet Hulled”. You might want to add one for roasted beans as well. For example @aqdennis reports above that he does a second round of sorting after the roast, as that allows to sort out any unripe beans that were not recognized during the first sorting. They appear much brighter than the rest after roasting.
There’s a mistake in there. The unit of power is Watt, or kilowatt, but not kWh. So instead, the values should be “≦ 50 W” and “≦ 100 W”, respectively.
If they fund this (9500 EUR), there will be enough resources to create a working machine and all relevant documentation. Let’s see … in about a month we’ll know more 
Of course we’re interested in some more tech details, if you are free to disclose them. Especially the input size of single-bean images fed to the deep learning image recognition algorithm (assuming you mean that by “AI”).