A short history of AI
A fully circular economy relies on consistently high-quality recovered materials, which is not yet possible with today’s processes, and many recovered materials are still downcycled. To avoid this, sorting must become more granular. And this is where AI is a game-changer.
Addressing the misconception that AI is a recent trend
The distinction between AI and deep learning is important because there is a common misconception that AI is a recent phenomenon. In fact, AI has been integral to our industry for decades, since it is the broader concept of creating intelligent machines. It refers to any technique that enables computers to ‘mimic’ human intelligence using logic, if-then rules and machine learning.
Machine learning has been a standard feature of our AUTOSORT™ machines for decades. Our early machines, dating back some 30 years, employed basic AI principles. Even back then, our machines were capable of making decisions about which materials to eject and reject. This fundamental ability to mimic human judgment is the essence of AI.
So, while AI has been used in global recycling for years, deep learning is the cutting-edge advancement that is propelling AI to new heights today. Deep learning is a specialized approach within machine learning that focuses on a specific type of algorithm called artificial neural networks. These neural networks are trained by huge amounts of raw data to recognize and store certain intricate patterns and later apply them to new data.
See how AI algorithms work in recycling
To understand the differences between AI, machine learning and deep learning, imagine you're a chef. AI is the entire kitchen, including all the tools, ingredients and the chef's knowledge. It's the overarching concept of creating something intelligent. Machine learning is a specific cooking technique, like a recipe. You follow a set of instructions, adjust based on feedback and learn to cook a specific dish better over time. Deep learning is a more advanced cooking technique that involves learning from experience. Imagine a chef who doesn't follow a specific recipe but learns by observing and tasting different dishes. They adjust their techniques based on the results, becoming more skilled over time.
How TOMRA has embraced the opportunities deep learning presents
TOMRA’s commitment to technological advancement has driven us to explore the potential of deep learning in recent years. Our team of AI experts feed thousands to millions of images into the network as training material until it learns to distinguish certain visual characteristics of material types such as specific bottle caps or packaging shapes.
It can apply this knowledge to new images from the sorting system’s sensors, making it possible to solve some of the most complex sorting tasks which are currently impossible with conventional optical sorting equipment. Furthermore, it enables TOMRA’s sorting systems to continuously improve output purity thanks to the simultaneous integration of real-time data from multiple sensor-based technologies.
Here are just a few of the benefits our customers gain from our deep learning-based solutions:
- Flexibility: With the ever-changing composition of waste, sorting systems need to be agile enough to continuously learn and adapt to new market requirements. Instead of replacing hardware components or even machines, modern deep learning technologies can be retrofitted with software updates as soon as they have been trained by our experts. This allows us to respond more quickly to customer needs.
- Creation of new material streams: AI-powered deep learning enables operators to not only enhance sorting granularity but create new material streams and markets with higher value outputs. I
- Improved sorting: By combining existing optical sorting systems, which are based e.g on near infrared (NIR) and visual information sensors (VIS), with deep learning technologies, we can achieve the highest sorting granularity currently available. This allows us to sort by material type and colour and now, thanks to deep learning, also by shape, size, dimensions or other details.
- Advanced plant automation: The value of deep learning lies in object recognition using full-color cameras. In other words, systems like our GAINnext™ see what the human eye can see. We can automate sorting tasks that previously had to be carried out manually, enabling us to process larger quantities of recyclable materials quickly and efficiently.
- Process optimization: AI-powered sorting systems generate huge amounts of data on material composition, sorting efficiency and equipment performance. By analyzing this data, plant operators can identify optimization opportunities, streamline operations and make informed decisions to improve overall recycling processes. Additionally, the possibilities reach beyond sorting systems. Cameras based on deep learning can be placed at key points in the sorting circuit to keep an eye on the entire process and material flow. AI-based waste flow analysis allows plant operators to continuously monitor the quality of sorted streams, material loss and even ensure compliance with food recycling regulations.
- Solving previously impossible sorting tasks: Deep learning solves previously impossible tasks, as we’ll explore shortly.
A timeline of TOMRA’s pioneering application of deep learning
2019: Introduction of the industry’s first deep learning-based sorting system GAIN (today: GAINnext)
-Removal of PE-silicon cartridges from polyethylene (PE) streams
2022: First deep learning application in the wood market
-Separation of natural wood from processed wood
2023: Further expansion of TOMRA’s deep learning ecosystem
-Deinking/Paper Cleaner application for cleaner paper streams
-PET cleaner application for even higher purity PET bottle streams
-MDF removal
2024: Launch of various groundbreaking new applications, including the industry’s first solution to sort food- vs. non-food plastics packaging; renaming to GAINnextTM
-PET Food vs. Non-Food
-PP Food vs. Non-Food
-HDPE Food vs. Non-Food
-Aluminum Used Beverage Cans (UBC)
2025: More new applications to come!
AI: A catalyst for the green transition
AI is poised to revolutionize resource recovery, making it a crucial tool in the green transition. As regulations tighten and consumer expectations evolve, our industry stands at a critical juncture. Deep learning offers a powerful solution to propel the circular economy forward.
At TOMRA, we also expect advanced AI and cloud technologies to be increasingly utilized for waste analysis, enhancing transparency not only at the sorters but throughout the entire sorting process. This is why we have consistently developed cloud-based monitoring tools like TOMRA Insight and invested in PolyPerception with its AI-based waste analytics platforms.
By unlocking the potential of AI, we can create new markets for higher-value products, further stimulating growth and sustainability.
