Cookies

This website uses cookies for analysis, personalized content and ads.

Artificial intelligence in our sorting process

What if a machine could sort e-waste at a single glance? And what if that could become a reality in the very near future? Together with researchers from ID Lab (UAntwerpen), Recupel is currently developing software that will soon enable us to automatically recognise small devices like mobile phones, home appliances, radios, computers and DVD players within our waste mountain. The next step: sorting type by type. How is that possible? With Artificial Intelligence (AI).

Artificial Intelligence and the 40,000-tonne mountain of e-waste

Today, processing plants still sort electrical and electronic devices the old-fashioned way, using the human eye. This offers the advantage of ensuring that the sorting is done precisely and the recycling goes smoothly. A producer can also (for example) ask for devices of a particular brand or type if it needs specific materials or parts to make new ones.

But sorting through 40,000 tonnes of e-waste per year is a lot of work. Really a lot. So we asked ourselves how technology might speed up this process, and we quickly began working together with ID Lab. This imec research group from UAntwerpen has vast experience with Artificial Intelligence, and that technology appeared to be the solution. Today we’re in phase 1 of the smart recognition system, where we’re testing out the system and finding ways to make the recognition more precise and reliable.

How do you sort e-waste accurately with AI?

Sorting has to be done carefully. Kind by kind. Brand by brand. Type by type. But can a camera eye do that without human intervention? The first thing we had to do was make the camera smart. And we did that by using all the photos of e-waste from the past 5 years to build up a memory store for the camera. Brace yourself: this involved over 1 million photos. Special image recognition software soon has to not only distinguish different types of small e-waste, but also specifications such as the brand, type and year of manufacture. That’s how it’s done today with the human eye.

And does AI work in practice as well?

The first test phase with the residual category of e-waste (mobile phones, radios, DVD players, etc.) demonstrated that the camera placed over 90% of the devices into the right category. That is an excellent basis for further refining the software in the coming weeks and months. We are shooting for 100% accuracy, where the camera will be just as infallible as the human eye.

When does the AI project go live?

The test phase is currently in full swing and we expect to be able to start working with the recognition software at the end of 2019. And that means everywhere, not just here at home. Other countries process e-waste the same way we do. We’re looking with partners at how foreign processors too will be able to sort more efficiently thanks to this AI application, ´Made in Belgium´.

esc