Real-time data for sorting systems in recycling technology

Sorting systems are now among the most data-intensive areas of the recycling industry. Modern lines capture material flows, purity levels, contamination and throughput figures in ever-higher resolution. At the same time, the complexity of the processes is increasing.

Different input materials, fluctuating quality and changing market requirements mean that operators must constantly adapt to changes. It is precisely at this interface between sorting technology and data analysis that PolyPerception and Tomra are working. The aim is to make real-time operational data directly usable from ongoing sorting processes, so that operators can identify more quickly where performance is being lost or where there is a need for optimisation.

PolyPerception analyses material flows using AI and computer vision

PolyPerception’s systems rely on computer vision, AI and deep learning models to classify material flows within recycling plants in real time. To this end, analysis points are installed at key points along the line. There, cameras and AI models capture the passing materials during ongoing operations. The main challenge lies in the dynamics of such processes. Material compositions change continuously, input qualities fluctuate and, at the same time, enormous amounts of operational data are generated. It is precisely this data that should not only be collected, but also made immediately usable for plant operations. PolyPerception focuses in particular on output streams and residual material fractions. This enables operators to track how cleanly certain materials are actually sorted or how much recoverable material is lost in the residual stream. The key areas of analysis include:

• Monitoring of purity and recovery rates
• Real-time analysis of material streams
• Detection of fluctuations in the sorting process
• Analysis of critical process sections within the line

Ask Poly simplifies data analysis for plant operators

A key component of the new platform is the ‘Ask Poly’ function. This is powered by an AI-based chat agent through which operators can ask questions directly in natural language. Instead of manually exporting data or evaluating various dashboards, users formulate their query directly to the system. The AI then performs the analysis and provides a clear answer. A typical example would be a query regarding the purity of a PET tray stream over the past hour or regarding notable fluctuations during a specific production period. The agent automatically analyses the available data and provides the results immediately. It is striking that PolyPerception has deliberately focused on low barriers to entry. Users should not have to operate complex analysis tools. Instead, the focus is on the rapid interpretation of specific queries. This is particularly relevant in recycling plants, as operational teams are often working at full capacity. Although large amounts of data exist, there is often a lack of time in day-to-day operations for time-consuming evaluations or in-depth data analysis.

Deep learning aims to highlight fluctuations in the sorting process

Analysis becomes particularly difficult where material flows change at short notice. Different input batches, fluctuating moisture levels or varying compositions constantly affect the performance of sorting plants. PolyPerception’s deep learning models are designed to highlight such changes before major quality issues arise. This not only provides operators with feedback on the plant’s current status but also enables them to identify trends at an early stage. This applies, for example, to declining purity levels in individual fractions or increasing losses of recoverable materials in the residual stream. This can be economically crucial, particularly for high-value plastic fractions. Tomra brings its expertise in sensor-based sorting technology to the table. The combination of sorting hardware and AI-supported data analysis demonstrates how traditional sorting plants are increasingly evolving into data-driven process systems.

Recycling plants are moving towards closed-loop control systems

At present, human decisions usually still intervene between analysis and action. The systems provide data and recommendations, but the actual adjustment of the plant is still carried out by operating staff or quality teams. According to PolyPerception, this is likely to change in the coming years. In the long term, data analysis and the sorting process are set to be much more closely integrated. The aim is to create closed-loop control systems in which plants can react automatically to changes. This would significantly expand the role of AI in recycling technology. Systems would no longer merely analyse, but actively intervene in the process – for example, by automatically adjusting sorting parameters or process speeds. This approach is becoming increasingly important, particularly in light of rising demands for purity, recovery rates and efficiency. Operators are under growing pressure to recover high-quality secondary raw materials with as little loss as possible, whilst operating economically.

Data analysis is becoming part of modern recycling technology

This development highlights just how much recycling plants are changing. Whilst mechanical processes used to take centre stage, digital control and analysis systems are now increasingly emerging alongside the actual sorting technology. PolyPerception and Tomra are pursuing an approach designed to integrate complex data analysis directly into day-to-day operations. What matters is not so much the mere collection of data as the ability to quickly derive concrete measures from it. This gives operators a new perspective on the sorting process. Not only do machine performance and throughput become visible, but also the interrelationships between material quality, purity and process stability. Particularly with increasing automation, this combination of AI, real-time data and sensor-based sorting is likely to play an ever-greater role in recycling technology in the future.