Artificial intelligence is already creating waves in the resources and waste sector and will continue to make recycling safer, cheaper and more efficient in the future, writes Andrea Lockerbie.
While the international media are busy debating how artificial intelligence (AI) could result in the end of humanity, the resources and waste sector has been using the technology to help with the many challenges it faces.
At sorting plants, AI computer-vision systems are enabling operators to “see” which materials are going into, though, and out of facilities more clearly.
These systems can make distinctions that were not possible with previous technology: identifying aluminium cans, trays and foil; distinguishing food packaging from non-food packaging; and identifying brands, levels of recyclability, financial value and carbon footprint.
Having this data to hand is a significant change, and can help improve quality, volume, efficiency, reporting and transparency.
AI is bringing a new wave of people to our sector. When Rishi Stocker, CEO of Safi (formerly TrueCircle), and his co-founder looked at the recycling sector, they saw it was “fundamentally being held back by the fact it didn’t have continuous monitoring and real-time data on the quality it produced”.
With their tech backgrounds, they created a trading platform – similar to Amazon – that charges commission per transaction and connects buyers (reprocessors) of recyclable materials with sellers (sorting plants). Integral to the system is an AI unit, which continuously monitors the seller’s output line to prove the quality it produces.
The AI units are given to sellers in return for a commitment to sell a certain percentage of outputs through the platform. As well as that, the Safi system takes care of transport, paperwork and financing – it’s an end-to-end trading system.
Why is it needed? Today, a sorting plant may sell clear PET bottles as a 95/5 grade. This should comprise 95% clear PET but, in reality, this figure can be anything from 83% to 98%, despite the price being the same. Under this way of working, there is no incentive for improving material quality.
“The only way you can facilitate a shift is to show sellers that, when they are transparent and have good quality, they get better pricing,” Stocker says.
The platform, which works with more than 150 facilities across the world, has attracted buyers of plastic, metal and paper, who are willing to pay a premium for data showing the actual quality of the material they are buying.
Prices are based on the past 30 days of collected data. A second version of the Safi AI product sends alerts on dips in quality to the team inspecting outgoing bales, so that substandard batches can be rejected.
Another well-known name in waste-sector AI is Greyparrot, which offers hardware-agnostic AI computer vision units that can be placed strategically on any key conveyor belt in a sorting plant, such as quality-control lines, residue lines, or infeed. They gather data continuously to help a plant understand its waste flows.
A plant can use this data to improve quality and efficiency, work out where further investment is needed, and reduce downtime and material losses.
Greyparrot told Circular that one plastics-recovery facility found it was losing £1.6m of material a year to its residue line. For businesses with more than one plant, this kind of data can help with monitoring and managing portfolios, as well as streamlining compliance and reporting.
Sampling is an obvious area where AI can help. This is typically done manually and is a requirement of the UK Materials Recycling Facility (MRF) Code of Practice. Currently, spot sampling analyses one per cent or less of the material that passes through a plant; AI can do 100%.
With extended producer responsibility (EPR) on the horizon, MRFs will probably need to increase sampling to identify and report what they handle, and generate “evidence points” for packaging material.
Greyparrot calculates that manual samplers need 375 hours to count and record 15 tonnes of material, for £4,300 based on average wages. Automating this with AI would take just six hours and cost £17 for a year-long AI unit licence. When you consider that data reporting could be automated too, the argument for AI seems compelling.
Recycleye’s offering is a combined AI vision and robot picking system that aims to help MRFs sort dry mixed recyclables better.
The actual sorting of the waste is what takes a long time – and effort.
Tom Harrison, the company’s technical sales manager, says sorting is where “Recycleye Robotics” offers the most value: “The actual sorting of the waste is what takes a long time – and effort. At the moment, this is done manually, which comes at a high cost and without much consistency.”
Recycleye’s robotic systems – of which there are 14 installed worldwide – are generally retrofitted into picking cabins as part of a quality clean-up process. Sometimes the systems replace single picking lines, or work upstream of human pickers, taking the pressure off them. “It reduces cost and increases efficiency,” says Harrison.
Robots also offer a solution to current recruitment issues, and humans are reported to be happy to work alongside them, claims Recycleye.
A limitation of AI vision systems is that they can only see what their cameras are pointed at, and robots can only pick what’s accessible. Ideally, waste should be presented in a monolayer on belts so it can all be seen and picked easily.
Bolt-on and separate systems are far less effective, and do not allow AI to be used to its full potential.
This kind of consideration is built into new recycling plants, where there is a chance to fully incorporate AI into the design. A great example is how Sherbourne Recycling has worked with MRF supplier Machinex (see Circular issue 24).
Managing director Richard Dobbs says: “Cameras in key locations are essential, as are belts that are wide enough to enable waste to be presented in a way that allows the technology to work optimally.
“It works much better if it is considered at the outset and included in every aspect of the plant’s design and operation. Bolt-on and separate systems are far less effective, and do not allow AI to be used to its full potential.”
Dobbs believes the key difference with Sherbourne’s new MRF will be “the volume of data generated by the AI system and what we, as an operator, do with that data”. It means more skills focused on data analysis and system control “to get the best out of that data and deliver real-world improvements and future development”.
People are now asking, what can you do for me? How is this going to change my business?
The company is looking to work with local universities and colleges with a focus on AI and machine learning, as part of its academic placement programmes.
Beyond dry mixed recyclables, Recycleye sees potential for its AI sorting system in waste electrical and electronic equipment plants, construction and demolition plants, and residual waste facilities.
Safi, meanwhile, is developing a third product with a camera focused on the material coming out of a seller’s baler, to help certify that bales are within agreed specifications. Stocker sees scrap metal as the next big space for Safi’s deployment.
For Mikela Druckman, Greyparrot’s CEO, AI will enable future MRFs to be much smarter, although humans will still oversee operations. With collaboration, the various sorting technologies can have connectivity and work together, and AI can automatically optimise the plant, similar to the way smart thermostats self-adjust to the temperature.
Plants will even be able to plan: if they pull in weather data, they could automatically adjust for incoming paper being wet on rainy days or for all the seasonal changes in material composition.
Beyond sorting plants, Druckman believes that MRF data can also drive change upstream. Knowing what ends up as waste, and where, can be used to influence producers, consumers and policymakers.
For Dobbs, AI’s other potential applications include “ordering, maintenance, haulage operations, energy optimisation, and much more”.
The recycling sector knows that AI has the potential to transform the way it handles waste, and we are just at the beginning of this exciting journey. As Druckman says: “People are now asking, what can you do for me? How is this going to change my business?”