Artificial Intelligence and the 40,000 ton e-waste mountain
Today processing companies sort electronic and electrical devices with the naked eye. This offers the advantage of accurate sorting, which results in smooth processing. Consequently, a manufacturer can easily ask for all devices of a specific brand or type when in need of certain materials or components.
Still, sorting 40,000 tons of e-waste per year is a big ask. A really big ask. So we started wondering how technology might accelerate the sorting process. We got cracking together with ID Lab. This IMEC research group at UAntwerp has extensive experience with Artificial Intelligence and that technology was just what the doctor ordered.
How does AI enable the accurate sorting of e-waste?
Sorting is a high-precision affair. Sorting by class. By brand. By type. But can a camera do this without human intervention? First of all we had to make the camera smart. We did this by uploading all the pictures of e-waste we took in the last 5 years as a reminder for the camera. Mind you, we’re talking about 1 million pictures. Special image recognition software not only had to be capable of distinguishing various kinds of small
e-waste but also detailed specs such as brand, type and model year. That’s exactly how human eyes do it today.
And does AI work in practice?
Absolutely. In the initial test projects with residual e-waste (mobiles, radios, DVD players, ...) the camera assigned 90% of all devices to the right category. That gave us an excellent basis to fine-tune the software. In the next phase we’re aiming for 100% accuracy so the camera will be just as infallible as the human eye.
When is the AI project going live?
Laboratory simulations are in full swing and we expect to deploy the recognition software at all our sites by late 2019. And by all our sites we don’t just mean in Belgium. Other countries process e-waste the same way we do. In conjunction with partners we are looking to see how foreign waste processing companies can also sort their waste more efficiently thanks to this AI application ‘made in Belgium’.