Sorting it out

The decision is made. A recycling facility plans to install equipment leveraging the latest artificial intelligence (AI) technology to improve sorting performance. That’s the easy step. Now what?

A broad range of AI software technologies and suppliers exist, and choosing the right solution can help offer a more granular sort. Most often, the first thought is a robotic arm sorter using deep-learning AI algorithms to pick multiple materials from the conveyor belt for a final quality check.

More recently, however, optical sorter manufacturers also have started leveraging the latest AI technology based on deep learning to purify anything from plastics to used beverage cans (UBCs). Combining optical sorters with deep learning-based AI holds the potential to recover more material at a higher purity than with traditional technology.

Implementing AI correctly at the right stage of the circuit to meet purity, throughput and yield goals can be a complicated process. The barrier to entry for AI software providers to the recycling industry is relatively low, resulting in an influx of new suppliers.

All that is required is a camera and the latest AI algorithms for identifying specific material. The result has been startup companies offering AI solutions, most often in the form of robotic arm sorters. These startups face a steep learning curve to understand the economics of the recycling industry.

Optical sorter suppliers offer the advantage of well-established relationships with partners that have been working in the market for decades. Inherent optical sorter machine sensors can be combined with the latest AI algorithms based on explicitly trained artificial neural networks to offer a more granular sort beyond the capabilities of a camera and software alone. Plus, these suppliers have a deep understanding of the multiple sorting steps at the facility, allowing them to provide guidance on the best technologies to employ at the proper stage.

Understanding specific needs

It may sound fundamental, but recyclers first must develop a good strategy of how to leverage the latest AI at the facility, if necessary. This can include considering the types of material that need to be sorted—both today and in the future—and whether the material stream is consistent or variable. Recyclers also must consider their desired purity, yield and throughput as well as level of contamination in the material stream.

Robotic arm sorters from AI technology providers might be able to sort the material using only a camera and algorithms, but will this meet the recycler’s targeted purity and yield levels? Whether the facility needs 70 percent or 95 percent purity is a critical factor.

Throughput also will determine the type of equipment installed for leveraging AI solutions. Using cameras and AI algorithms, robotic arm sorters are trained to see what the human eye can see, differentiating materials by color, texture and shape. While they may offer more consistent product recognition at pick rates like a human, this equates to roughly 0.5 tons per hour of throughput per sorter. Relatively quick and easy to install, these sorters can be effective to replace the manual labor used for the final quality check.

But what options exist for the rest of the sorting circuit?

Optical sorters and AI

It should be no surprise that optical sorter manufacturers are offering the latest AI solutions as part of their technology portfolios. Advancements in cameras, sensor technology and traditional AI programs employed by optical sorters can, in many cases, meet the recycler’s purity, yield and throughput goals.

For more complex sorting tasks beyond the capabilities of conventional sorters, suppliers offer more sophisticated deep learning-based AI add-on technologies, such as Tomra Recycling’s GainNext, to sort objects they previously could not. These technologies can be integrated into existing lines to improve sorting at targeted locations, and this intelligent AI technology enables recyclers to target a specific value stream to increase purity.

Combining an optical sorter sensor, such as near infrared, with a color camera and the latest AI sorting algorithms delivers a more granular sort beyond the AI solutions using only a camera.

In the case of plastics, an optical sorter plus the latest AI combination delivers sorting beyond the shape, color and texture of plastics in the stream. While optical sorter sensors dive into material composition to distinguish between polyethylene terephthalate (PET), polypropylene, polyethylene, high-density polyethylene, low-density polyethylene and more, the intelligent AI algorithms offer the ability to detect food-grade packaging from nonfood-grade containers, white and opaque PET bottles and silicone cartridges, for example. Because artificial neural networks have been trained using thousands of images with deep-learning AI, there is more flexibility to target specific fractions.

UBCs are another area where recyclers can benefit from installing the latest AI. 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 UBCs from the rest of the aluminum.

The latest AI technology can be integrated with optical sorters at various stages in the circuit to extract valuable material from the waste stream, making it possible to achieve purity levels exceeding 95 percent. This is accomplished at much higher throughput rates than robots that use AI, reaching 8 tons per hour.

Technology optimization

While many AI technologies and options currently exist, one constant must be adhered to with system integration—optimization. Equally as important as location and type of AI sorter, technology optimization is a key to success.

The optimization process starts as early as testing the feed material received at the facility. Next, the facility should work with the supplier to select the right technology, installed at the right location, to meet desired targets. The process continues with the recycler and supplier working together to dial in sorting performance after installation. Finally, the optimization process extends to after the sale with support, so the equipment/technology delivers desired results well into the future.

Recyclers must do their homework, not only pertaining to the type of technology required, but also the company supplying the equipment. Consider the company’s longevity in the recycling industry and experience with sorting the type of infeed material. Discuss how the equipment will perform, not only today, but throughout its life cycle.

Ty Rhoad is vice president of sales — Americas for Germany-based Tomra Recycling, which designs and manufactures sensor-based sorting technologies for the recycling and waste management industry. He is based in Tomra Recycling’s Charlotte, North Carolina, office and can be reached at ty.rhoad@tomra.com.