Revolutionising waste management: AI improves sorting efficiency and material purity

In the interview with K-Mag, Parshva Mehta explains how he and his co-founders want to revolutionize waste sorting with advanced AI models.

Can you describe PolyPerception's mission and how it's addressing the challenges in waste management and recycling industries?

Parshva Mehta: The motivation behind starting PolyPerception stemmed from a recognition of the growing challenges in the waste management industry, particularly around achieving high material purity and sorting efficiency with increasingly ambitious recycling targets. We saw an opportunity to leverage cutting-edge technologies like AI and deep learning to address these issues head-on.

At its core, PolyPerception utilizes advanced AI models tailored for sorting lightweight packaging materials, which are a significant part of waste streams in Europe. By applying deep learning, we're able to analyze waste streams, optimize sorting processes, and provide plant operators with actionable data for better decision-making.

Could you delve a little deeper into the technology behind PolyPerception and how it enhances waste sorting?

Mehta: We utilize deep learning and vision-based sensors to analyze the waste sorting process at critical points, providing valuable data to operators. Initially, our focus is on accurate measurements of throughput, recovery rates, and output purity. This data lays the foundation for optimizing and eventually automating the sorting process. Our collaboration with TOMRA is crucial here, as it allows us to integrate our analyzers with their sorting machines for a more efficient process.

Our AI models are designed to adapt to different waste streams across Europe, ensuring reliable classification of various materials, such as polyethylene, polypropylene, PET, and more. We tackle challenges like model adaptability, maintaining accuracy over time, and automating the model training process. These efforts aim to provide consistent and accurate sorting across various regions.

Exhibitors & products around digitalisation

• CAD systems
• Simulation software
• Quality assurance software (SPC/SQC)
• Maintenance software
• Equipment & software for manufacturing execution system (MES)

Adaptability seems to be a crucial aspect of your technology. How do you ensure your models remain effective across different regions and waste compositions?

Mehta: Our deep learning models are designed to adapt and generalize across varying waste streams. This means they can accurately classify waste, even as the types of materials in waste streams evolve. It's all about ensuring consistency and reliability in a continouosly changing sorting process.

This adaptability ensures consistent and accurate data measurements, even as new materials enter the market or existing ones undergo changes, thereby maintaining the efficiency and reliability of the sorting process.

How does this work exactly? Do you have an example?

Mehta: Our system targets two primary customer segments: sorting plants and recyclers. A standout example is our collaboration with Indaver in Belgium, leveraging the PMD system for lightweight packaging waste. Indaver utilizes our technology across five sorted streams to meet purity levels set by the producer responsibility organization, Fost Plus. Traditional audits were sporadic and unrepresentative of continuous operations. By integrating our systems, Indaver achieves around-the-clock monitoring of output stream purity, enhancing efficiency and compliance significantly.

Your collaboration with TOMRA has been highlighted as a significant milestone. Can you elaborate on the importance of this partnership?

Mehta: Our collaboration with TOMRA began at the end of 2022, aiming to integrate analyzers into their customers' sorting processes for enhanced data insights. This partnership evolved when TOMRA acquired a 25% stake in PolyPerception, deepening our relationship to improve and monitor sorting lines comprehensively.

This strategic move addresses the industry challenge of disparate machinery communication, aiming for a unified, automated sorting process. Such integration and strategic partnerships are crucial for advancing automation and efficiency in recycling operations, reflecting our long-term vision of close collaboration with TOMRA.

Looking to the future, what are your visions for the waste management industry, and how does Polyperception plan to contribute to that vision?

Mehta: I'm imagining a world where waste sorting isn't just a static, one-size-fits-all process but one that adapts in real-time to the kind of waste it encounters. Right now, the sorting systems are pretty rigid, but imagine if we could analyze the waste flowing in, learn from the sorting results, and then adjust the machinery on the fly to improve efficiency.

This means adding more sophisticated analyzers and leveraging deep learning to not just process waste better but also to handle a diverse range of waste qualities from different regions. It's about making the whole sorting process smarter, more adaptive, and ultimately, more efficient.

And finally: What are the next steps for Polyperception in terms of product development and market expansion?

Mehta: In the short term, we'll continue focusing on European waste sorting and recycling markets. Looking forward, we're exploring expansion outside Europe, particularly in the US. Product-wise, we aim to deepen our integration with TOMRA's sorting machines, leveraging our data for smarter sorting processes.