Investing in the latest sorting technology pays off in the long run.
By Rick Zettler

Sorting automation technology available to the MRF is rapidly evolving. If your recovery facility is using optical sorting technology that is 10 years old, there is a possibility that your yields are dropping, landfilling costs for lost materials are increasing, and you are relying on more manual labor to pull desired material from the waste stream. Even machines fewer than 10 years old may have updated technologies or add-ons from the supplier that could help improve the sort.

“I commonly see MRFs holding on to inefficient and old equipment and not replacing it with new and cutting-edge technology that could substantially increase their yield and profitability,” offers Sebastian Ward, Key Accounts Manager for TOMRA Recycling Sorting. “Leveraging today’s latest sorting technologies can help reallocate manual labor from the line, and labor is one of the biggest challenges MRFs in North America are facing today.”

However, it is tempting to hold onto and maintain the equipment for as long as possible. It is paid for, so it must offer the lowest cost per pound of material recovered, right? Parker Bynum, Area Sales Manager for TOMRA Recycling says this often is not the case. “Think about when you first start using a new computer or buy the latest smart phone. They’re fast. They do everything you want them to do and more. They allow you to be much more efficient than you were,” he says. “Now imagine you still using that same computer and phone 10 years later and how much technology has evolved. It’s that way with sorting automation.”


Cloud-based technology delivers remote monitoring of near-real-time sorting data like throughput, material and size distribution, and acceptance and rejection rates. Photos courtesy of TOMRA.


The workarounds that MRFs must do to compensate for outdated sorting technology will, in the long run, cost the operation more in lost yield, equipment maintenance and labor. With advanced sorting automation technologies available, it is the right time to work with a technology supplier to consider an upgrade.


Leveraging the latest sorting technologies can help to improve material yield and purity while also helping to reallocate manual labor from the line.

Reallocating Resources
Line sorting automation has a convergence of technologies today. Optical sorters leveraging traditional artificial intelligence (AI) have been available and improving for decades. The emerging technology in the market leverages large datasets of material stored images and deep learning AI algorithms to further improve the sort.

“The basic understanding we need is that both are different sorting technologies, which have their own unique values,” explains Indrajeet Prasad, Product Manager, Deep Learning for TOMRA Recycling Sorting. “Near infrared (NIR) and other sensors in optical sorters do the sorting based on material types, distinguishing between polyethylene terephthalate (PET), polyethylene (PE), and polypropylene (PP) in the case of plastics. Deep learning, on the other hand, sorts based on visual information—what the eye can see.”

It is this visual-based sorting where deep learning AI made its first mark on the recycling industry. The camera, deep learning algorithms and robotic “arms” to sort materials were integrated at the back end of the sorting line to perform a final quality control check and recover material that would have been lost to the landfill. This helped to reallocate workers from the pick line, which also aided in increased spacing between workers required by social distancing during the pandemic.

Reallocating workers from the line to positions of more value for the MRF is what Ward and Bynum see as the key to why it is important to upgrade to the latest technology. Sorters are designed to automate the process while increasing yield, purity, and profitability. “Labor is tight in the United States, and due to the difficult working environment on the sorting line, MRFs face a lot of turnover with line workers,” says Ward. “Additionally, minimum wages are increasing. Add to wages the benefits for these workers, and having a person on the line is quite expensive. Looking at this year over year and augmenting the line with AI-based technology to reallocate one individual worker can quickly help to cover automation costs.”
Bynum sees the labor issue also impacting equipment maintenance, which can make line sorting performance deteriorate. “Labor shortages lead to equipment maintenance suffering. Preventative maintenance falls off, and combined with the machine’s age, this can significantly impact sorting performance.”


PET Cleaner deep learning application identifies and removes 92 percent of opaque with titanium dioxide protection to enhance sorting of transparent and color PET.


Pairing Deep Learning with Optical Sorters
For the last few years, applications have been developed for deep learning AI to enhance the sorting performance for MRFs. It combines the best of both worlds—the material composition recognition of the optical sorter and the visual recognition of the AI camera. Combining an NIR and other sensors with deep learning algorithms is the peanut butter cup of the recycling world that solves difficult-to-solve sorting tasks.

The combination addresses several limitations of the camera-based AI solution alone when sorting key materials at the MRF. For one, the pair enables material to be sorted using valve-block technology that is much faster than the robotic arm. “There is an inherent lack of efficiency, as the arms are designed to mimic the picking action of a human, so they offer around 70 or 80 picks per minute,” says Ward. Bynum agrees and adds, “Valve blocks deliver a much higher capacity, and we are seeing purity rates in excess of 95 percent at throughput rates reaching 6 tons per hour.”

Beyond speed, Prasad mentions another challenge associated with the arms used to sort material using AI. “Speed is definitely one factor, but the suction cups on the robotic arms require maintenance virtually every day,” he says. “This increased maintenance can pose an issue, especially with a tight labor market.”

Finally, pairing sensor-based traditional AI with deep learning offers a more granular sort of plastics. “The camera with deep learning will see color, shape, and texture of the material on the belt. So, it can recognize a green soda bottle or a clear water bottle or a milk jug,” adds Prasad. “When combined with an NIR sensor of the optical sorter, we can distinguish that it’s a PET bottle with either a PET or PVC label.”


AI-powered deep learning plus NIR technology enables deeper material recognition at high throughput speeds to offer higher sorting purity and quality.


Plastics Applications
Plastics sorting tasks are where this powerful combination plays a significant role. One of the first applications originated five years ago in Europe with selectively sorting silicone cartridges. More recently, the European market has seen the ability to sort PET, HDPE (high-density polyethylene), and PP (polypropylene) food-grade from non-food-grade plastic packaging. Bynum says previously the industry’s goal was just material recovery. “Now, it’s replacing virgin material with post-consumer recycled (PCR) content,” and this is a critical time, says Ward, for the industry struggling with PCR supply.

“Recycled plastic feedstock is currently insufficient to keep pace with industry demand, and evolving environmental policies like Extended Producer Recycling (EPR) programs will continue to place demands on the strained supply,” claims Ward. “Current and future deep learning applications can allow MRFs to increase plastics material recovery, improve bale purity, and MRF profitability, and offer the flexibility to adapt to future market opportunities.”

Another new application, soon to be launched in North America, is an NIR-deep learning application developed for high accuracy sorting of opaque white packaging, textiles and foils from PET streams. Leveraging deep learning, it identifies and removes opaque with titanium dioxide protection, and it significantly enhances the sorting performance of transparent PET and color PET by removing polyester textile waste.

Prasad mentions that deep learning combined with sensor-based sorting targets hard to eject materials in the stream to enhance the performance. “In the case of opaque materials, adding the visual information on top of sensor recognition, it is now capable to determine if the object is opaque compared to other material types.”


The optical sorter sensor sorts based on material types like PET, PE and PP plastics, while deep learning AI sorts based on visual data.

Emerging Opportunities
This is not to say no sorting application exists for the camera plus deep learning AI alone. It does well distinguishing the visual differences between bags. Additionally, metal recycling and specifically used beverage container (UBC) classification can benefit from deep learning AI using visual information alone.

Whereas optical sorters alone accurately identify and sort aluminum from the material stream, trained deep learning AI technology takes the next step to sort food-grade UBC from the rest of the aluminum. “Virtually every MRF in North America currently has a cleaner line where a person is hand-picking UBC from the stream to get final output of the clean UBC fraction,” comments Ward.

Only visual information is required to quickly distinguish between the desired UBC fractions and cat food cans and other types of light metals. The MRF gets the best benefit from the camera and deep learning algorithms alone, which also allows for the reallocation of at least one picker from the line. Plus, North American MRFs have the option of choosing between the robotic arm or high-speed valve block ejection for the line.

Apart from upgrading to the latest sorting technologies available, MRFs now have at their disposal cloud-based solutions that provide more sorting data, allowing for fact-based decision-making. By providing near-real-time monitoring and insight into digital metrics such as throughput, material and size distribution, acceptance and rejection rates, among other data, these monitoring solutions give MRFs the critical information necessary to optimize sorting performance. The actionable data helps reduce machine downtime, optimize machine settings, maximize throughput, sort to target quality, improve the operation’s efficiency, and reduce costs. “I’m excited about the technology combinations offered to customers in addition to the bread-and-butter optical sorter solutions. This can help increase customers’ sorting efficiency even more so than in the past,” concludes Bynum. Ward adds, “If you are unsure about deep learning AI and implementing it properly, talk to the supplier. We are here to help MRFs be as effective and profitable as possible with the innovative technology.”
With the recent advancements in sorting technologies and expansion of deep-learning-based AI applications, now could be the right time to make system upgrades to increase flexibility to take advantage of market opportunities. | WA

President of Z-Comm LLC, Rick Zettler is a writer, photographer and award-winning PR and Marketing consultant specializing in the mining, recycling, construction and road building industries.

TOMRA Recycling designs and manufactures sensor-based sorting technologies for the global recycling and waste management industry to transform resource recovery and create value in waste. The company was the first to develop advanced waste and metals sorting applications using high capacity near infrared (NIR) technology to extract the most value from resources and keep materials in a loop of use and reuse. To date, more than 9,000 systems have been installed in 100 countries worldwide. TOMRA Recycling is a division of TOMRA Group. TOMRA was founded on an innovation in 1972 that began with the design, manufacturing, and sale of reverse vending machines (RVMs) for automated collection of used beverage containers. Today, TOMRA is leading the resource revolution to transform how the planet’s resources are obtained, used and reused to enable a world without waste. For more information, visit