Automating ingenuity

Consumers of secondary commodities, including plastic, paper, metal and wood scrap, all maintain specifications related to the purity and the quality that these materials must possess, as they are key ingredients in the goods they produce.

Many of these consumers advocate for collection that is as source-separated as possible. That is, if a plant is using clear PET (polyethylene terephthalate) scrap as feedstock, the owner of the plant wants only that specific feedstock shipped to the facility without contamination.

Managers of waste and recycling companies often reply, however, that to obtain sufficient volumes of a given material—including clear PET bottles—using only pinpoint targeted collection methods will come up short.

This disparity between what is able to be easily collected and what materials are desired by a given consumer has helped lead to one of the priciest research and development (R&D) efforts in the waste and recycling sector this century: the ongoing quest to deploy automation to separate commingled recyclables from one another.

The effort has a long history of industry stakeholders bridging mechanical, magnetic and laser-optical techniques to achieve this end goal. Some of the latest technology deployed fits into the artificial intelligence (AI) or machine learning categories with a healthy side order of robotics included.

Defining the tactics

Operators of material recovery facilities (MRFs) and other recycling plants face the risk of being overwhelmed with technical terms as they sift through pitches and proposals from technology and machinery vendors.

To understand the basics of these technologies, distinctions need to be made regarding the differences between AI and machine learning. AI is considered a broader category of computer systems developed to perform tasks normally requiring human intelligence, including visual perception and follow-up decision-making.

Within AI, machine learning is defined by ExpertSystems.com as “an application of AI that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it to learn for themselves.”

When it comes to identifying and separating materials in a commingled recycling application, machine learning is a segment of AI that is playing a starring role.

Companies including Europe-based TOMRA Sorting Recycling; Eugene, Oregon-based Bulk Handling Systems (BHS); Canada-based Machinex; and Finland-based ZenRobotics are among those focusing on tying machine learning into devices and systems to offer thorough automated sorting options for recyclers.

Learning to sort

Companies such as BHS with its MaxAI systems, Machinex with its SamurAI line of sorting machinery and ZenRobotics with its Recycler have all gained attention and investments from recycling plant operators who have calculated that such advanced technology will yield a healthy return on investment (ROI).

Even before machine learning became an integral part of commingled recycling sorting systems, investments in automation focused on two priorities: improving the quality of shipped secondary commodities and decreasing labor costs on the sorting line.

BHS marketing plays heavily into both. “Max,” the spokes-robot for MaxAI, states on BHS’s website, “I don’t get sick. I don’t need breaks, lunches or days off. I work harder, longer and better than anyone else.” Regarding quality, Max adds, “I’m more accurate and more efficient than anyone could be.” And when it comes to machine learning capabilities, Max states, “Thanks to my intelligent neural network, I’m capable of learning on the job so I can adapt to changing conditions and variables.”

Jonathan Ménard, an executive vice president with Machinex, describes the company’s SamurAI product line as a “self-aware sorting robot [that] answers a worldwide need for increased automation.”

Ménard says the SamurAI was unveiled at the April WasteExpo event in Las Vegas, and then internationally a month later at the IFAT trade fair in Germany. He says the MRF market has responded positively to it. “Machinex has officially sold eight SamurAI units, including three of them that will be running before the end of 2018,” he says.

Buyers in the MRF segment are attempting to garner their ROI on the commingled container front, and they’ve turned to AI-enabled technologies to fast-track improvements in lowering contamination rates.

“The majority of our applications are currently for the recovery of different types of recyclables on the reject quality control line, which mainly allows the recovery of natural and colored HDPE [high-density polyethylene], PET [polyethylene terephthalate], metals, Tetra Pak and other types of plastics otherwise missed by the previous sorting equipment,” Ménard says.

Advances in robotics

Learning-enabled robots are also gaining a presence in the sorting of mixed construction and demolition materials (C&D), where objects in a commingled stream can be picked in either a negative or positive sort. Recently, robots programmed by ZenRobotics have made an impact on how operators in this space are able to sort incoming materials.

Operators of mixed C&D recycling facilities face labor cost and quality control issues similar to those encountered by MRF operators. As have many MRF operators, C&D recycler Walter Biel of Austin, Texas-based Recon Services has invested in machine learning and robotics to address both of those issues.

In 2017, Biel and his staff worked with ZenRobotics and its U.S. distributor Plexus Recycling Technologies, Denver, to become the first C&D recycler in the country to deploy ZenRobotics robot sorting arms. (A profile of Recon’s overall operations can be found on the Construction & Demolition Recycling magazine website at www.CDRecyler.com.)

At the Recon plant in Austin, two robot arms with “smart grippers” have been programmed and deployed to pick 12 different kinds of materials, and they can separate plastics based on polymer, color, shape and size.

Recon Services says the robots can make roughly 2,000 picks per hour, selecting objects with market value in a positive sort process. By comparison, according to Recon and ZenRobotics, humans can make approximately 800 such picks per hour.

“The robots have added a positive piece to our overall concept,” Biel told Construction & Demolition Recycling magazine earlier in 2018. “Being the first to implement something always has its good and bad, but it never affected any decision we made. It was something we saw value in and decided to add into our operation.”

Whether to be a pioneer of a new technology or wait and see if early adopters benefit will be an ongoing decision-making process for recyclers of all commingled materials as AI and machine learning continue to be configured to work along with sorting devices.

One thing is for sure: These technologies aren’t going away. Sorting technology companies have committed to AI, machine learning and robotics as an integral part of their future. The question is how advanced these machines can become.

According to Ménard, this is an ongoing process.

“We are still working with our partner [Colorado-based] AMP Robotics to further develop enhanced identification technologies and capture rates for plastics applications,” he comments.

When it comes to robotic arm sorting, Ménard states, “In reality, the robot is a tool powered by the AI. Once the neural network of the AI is well-developed, the technology can be inserted in many existing sorting technologies to enhance their performances (recognition, purity, maintenance requirements, auto adjustments, etc.).”

Technology providers such as Machinex are not standing still with their current AI-related product lines, says Ménard. Beyond the sorters themselves, he says a “focal point of development” at the company right now involves “technologies needed to design a connected smart facility,” referred to as “Industry 4.0.”

“We are currently establishing the foundation needed to collect and analyze essential data that will be available and useful to track for the MRF operator,” he says. “This data would help any operator in his or her decision-making process and would be supported by clear indicators of what is currently going on in the plant. Ultimately, this information would also lead to automatic adjustments of certain equipment, assuring complete system performance optimization.”

When it comes to investing in machine learning and AI sorting technology, recyclers clearly will continue to have plenty of vendors and options to choose from. By examining variables such as the type of incoming feedstocks, desired purity rates, speed, accuracy, manpower and automation, operators can help narrow their search and find the sorting solution that best fits their needs.

The author is a senior editor with the Recycling Today Media Group who can be contacted at btaylor@gie.net.