When robots and AI sort textile waste

“The ability to sort textiles automatically by material and colour is an important step towards being able to recycle used textiles to a high standard in the future,” says Vittoria Troppenz, Head of Circularity at SOEX. According to Troppenz, it was foreseeable even back then that the volume of textile waste would increase, and that automated sorting would become more important, especially for textiles that were no longer wearable. 

SOEX uses a combination of three technologies to improve textile-waste sorting: near-infrared spectroscopy (NIR) analyses the molecular composition of the textiles and checks whether they are single-origin or mixed textiles, such as polyester/cotton or acrylic/cotton. Then, an RGB sensor sorts the textiles by colour, while an object recognition system – still under development – separates them into different product groups, such as trousers, T-shirts or dresses.

To ensure precise material recognition, SOEX has trained an algorithm based on machine learning, which, according to Troppenz, has a recognition accuracy of up to 95 percent. “Our system can distinguish between 90 different natural and synthetic materials and material combinations,” she says. Currently, the system processes 1,500 items or around 500 kilograms per hour for chemical or mechanical recycling. In addition to textiles from classic container collections, SOEX also processes textiles from the retail sector via its subsidiary I:Collect. These textiles include garments, home textiles, such as bed linen and towels, and workwear. Troppenz has no doubt that SOEX will continue to rely on its in-house sorting technology despite the provisional insolvency proceedings that were initiated recently: “A change of focus towards high-quality textile-waste recycling is vital for the success of a circular textile economy in the future.”

"Fibersort was a response to the fast-fashion problem, which leads to large amounts of waste from low-quality clothing", says Jean-François Gryspeert, Sales & Business Developer at Valvan. 

According to Gryspeert, Fibersort bridges a gap between textile sorters and recycling companies by identifying the fibre composition of textiles. "With manual sorting, one has to rely on touch and label. However, studies show that label information is incorrect in as many as 50 percent of cases", explains Gryspeert. Fibersort automatically recognises fibre composition – "from baby socks to bed linen". Building on this technology, Valvan developed ‘Trimclean’ in 2020, an AI-aided solution for even better material recognition. Gryspeert vividly recalls the unexpected source of inspiration for Trimclean: "One of our engineers was visiting a nearby company that makes French fries. There, he saw how visual inspection was being used to sort out substandard fries. This led to a brainstorming session on using AI to process visual information when sorting textiles". According to Valvan, Trimclean is now so good that it can even recognise and automatically sort out patches, seams and prints on textiles.

Now investment is called for. "What we need in the coming years is money. EPR, or Extended Producer Responsibility, will become reality in 2025, and success now depends on attracting investors to expand textile recycling", says Gryspeert. 

The industry's response to Trimclean winning the Texprocess Innovation Award 2024 in the "Digitalisation and AI" category shows that there is a demand for pioneering solutions in the field of automatic textile sorting: "Since the awards ceremony in April 2024, we have noticed a significant increase in interest in Trimclean", says Gryspeert. "Almost all the enquiries we receive mention the Innovation Award".

Cover picture: The SOEX system automatically sorts textiles by material and colour and will also be able to differentiate between product types such as T-shirts, trousers and dresses in the future.