What are the key issues in technological development for textile sorting at the moment?
Dr Karsten Pufahl: Precise sorting is the top priority, because a circular economy is only possible if everything finds its correct recycling channel. Around half of the used clothing in the global North is still wearable. In terms of the waste hierarchy, it is important to ensure that these valuable goods are fished out of the stream. Processes such as fibre-to-fibre recycling are still being developed, because for decades the separation of qualities was done manually. There is now an increasing lack of personnel in Europe for this, also in view of rising labour costs.
Where do you start with your technology?
Dr Karsten Pufahl: The automated process begins with photographing the item of clothing on the conveyor belt. Different AI models then analyse which item of clothing it is. For example, the background is separated from the foreground of the image and the textile is assigned to one of the 120 or so sorting categories that we have set up. In the hierarchical system, the AI recognises step by step whether it is an outer or undergarment, a coat or a jacket, the colour and brand of the item of clothing, etc. The biggest factor is recognising defects such as damage and soiling. There is also a separate quality scale for the sorting channels. The AI recognises the type of damage, for example whether the fibres are worn or torn. The fibres are transported to the respective categories using a conveyor belt and compressed air. Humans no longer need to be involved in this process and can focus on the tasks in advance.
To what extent is the classic analysis of textiles currently already supported by AI?
Dr Karsten Pufahl: There are pilot plants that offer material-specific sorting with neural networks, but these are still the exception. Behind this is a very efficient AI that generates the materials from the spectral data. Artificial intelligence is not yet used on a larger scale, such as for image analysis.
What difference will the technology you and your team develop make?
Dr Karsten Pufahl: Our process is more time and cost efficient. We also offer target group-specific sorting. This allows us to cover criteria that humans alone cannot achieve. The standard process consists of pre-sorting into around 20 categories and fine sorting of the individual streams. Each additional category means that more labour is required. An AI, on the other hand, can implement this sorting for up to 1000 categories alone, for example, and can be quickly reconfigured as required. This is also interesting if you want to sort seasonal goods or pick up on specific trends such as high-quality vintage goods.
Can you say what data the AI is trained with?
Dr Karsten Pufahl: This is an elaborately generated data set. What I can say is that we used individual data sets for the different challenges. The AI is thus given different tasks every day in image classification. We are also in the process of generating and labelling newer data. It's primarily about recognising errors, which is the most time-consuming step, also in terms of the development effort on our side. We are in dialogue with our development partner Remondis on this.
What do you want to change with your work in the textile industry?
Dr Karsten Pufahl: Above all, we want to ensure that textiles find the best possible recycling channel. It is also crucial for us to offer target group-specific sorting that makes second-hand goods more attractive so that they can be sold as ordered in the future. Automation is needed to enable these processes for large quantities of goods.
Can you imagine exhibiting at trade fairs such as Texprocess, Heimtexil or Techtextil?
Dr Karsten Pufahl: As soon as our solution is ready for the market, we will certainly do that.
What are you currently working on?
Dr Karsten Pufahl: We are in the process of perfecting the detection of defects.
Author: Anna Moldenhauer, Stylepark Magazin
www.stylepark.com
