Turning data into action

As material recovery facilities (MRFs) increasingly use technology such as optical and robotic sorters and artificial intelligence, more data on their day-to-day operations has become available.

But, with so much data coming in, it can be difficult to determine which is the most actionable.

According to a panel of recycling industry veterans assembled at the Recycling Today Media Group’s MRF Operations Forum in October, sales pitches from equipment manufacturers have included bold performance claims, but the need for useful data remains.

“[Data collection] is something we’re more and more involved with,” said Jeff Nella, area recycling director – U.S., for Vaughn, Ontario-based GFL Environmental Inc. and the moderator for the Advanced Opticals and MRF Data session. “We’re constantly collecting more and more MRF data, and we’re inundated with all types of manufacturers who are making all different types of claims about what information is available, what data they can provide to us and what we should be doing with it.”

Volker Rehrmann, executive vice president and head of Germany-based equipment manufacturer Tomra Recycling, a division of Norway-based Tomra Systems ASA, added that determining what data is valuable can be difficult for manufacturers and operators alike.

“What is the value of just collecting data and you can’t make any use out of it,” he said. “You cannot improve your operations by using it. And that’s the challenge of where we’re at right now. … Collecting a lot of data is easy. Making business use out of that is a tough road.”

The right place

The panel, which also included Riel Johnson, vice president of resource recovery for City of Industry, California-based Athens Services, and Ron Grinold, Great Lakes area manager for Phoenix-based Republic Services, agreed the placement of sorting technology within a MRF is important for data collection.

Nella, whose company runs more than 30 MRFs in North America, said optical units found a home on the container line, then migrated to fiber lines.

Grinold said there is room for improvement in terms of where opticals typically are located and suggested positioning an additional optical behind one to serve as a quality control measure.

“I think there’s still room for advancement,” he said. “Layering the typical optical with artificial intelligence (AI) on top of that, that’s coming, and that’s going to be the next big push.”

Johnson said he’s seen benefits to placing optical units on the last-chance line. “We always want to know what’s going into our trash,” he said. “If we’re losing good material, we want to be able to capture that data and, over time, [have] historical data. Right now, without the data collection, we have to do hand sorts and open up the trash compactors. I think that’s the best opportunity.”

Once data have been collected from a MRF’s sorting technology, the panelists agreed it is important to control and own that information rather than what Nella described as “leasing back that information from a vendor.”

“I think it’s important we have our own data,” Grinold said.

“Also, I’m not a big fan of subscription fees. I think if you buy a machine, it should come with the AI already in it, the neural network already loaded up. I don’t think there’re that many changes in the containers … . Maybe you pay for an upgrade, and you pay that one fee at one time. … But I’m not a fan of a continual needle in my arm to get my neural network to me.”

Johnson compared managing sorting equipment to smartphones. “Every year there’s a new one coming out with a new technology,” he said of optical sorters, adding “Does that require a subscription fee to be a part of that, or is that [an upgrade] you can choose to purchase?” 

Some of the data collected in MRFs has been used to improve operations. Grinold, who oversees multiple MRFs across Michigan, Ohio and Indiana, said that while upgrading the opticals at one facility, AI was used to provide feedback on picks that Republic could then take to its equipment provider to adjust.

“[We could see] the number of picks that were going by, like, for example, how many PET [polyethylene terephthalate] objects were going by,” he said. “You didn’t have to stand there with a clicker; you didn’t have to stand there and take samples. The AI was doing it live, and it was a great system.…  What came out of that, too, though, is it set a bar for acceptability as well.”

Nella added that lots of high-tech information is moving back and forth, but many MRFs still have older, standard optical sorters that measure square inches or square centimeters on the belt and provide feedback on what is on the belt. With newer systems, Nella asked, “Is that data that’s now just packaged in a little bit sexier format than what we’ve always had? Because I can look at two optical units that are following each other and I can see what’s on the belt on one, and see the square inches covered by whatever I was looking at on belt No. 2, so I can see what I missed or didn’t get.”

Johnson, whose company operates several MRFs in the Los Angeles area, said it comes down to an operator’s trust in the data being received. Along with Nella, he suggested equipment manufacturers help manage operational controls for MRF systems to better manage the stream of data.

“Quite frankly, none of us are scientists,” Nella said. “It’s just tip floor’s full, tip floor’s empty, good day. Is this stuff that we need to get off our plate and push back to some of our system providers, who are doing our operational controls for our systems?”

According to Grinold, “By the time you’re getting your data at the end of the day, it’s too late. The day’s gone. If it’s live and it’s changing your screen angles, for example, or it’s changing the speed of your infeeds, that’s real stuff right there. And it’s immediate. As we know, each day’s a little different, and each load’s a little different, and to be able to not rely on an operator to crawl around the railing and say, ‘Hey, speed up this and slow down that,’ and let the artificial intelligence do that, that makes a lot of sense.”

Reality for robots

When it comes to the use of robotic sorters, Nella recalled that about five years ago the technology was considered a fad, but the units have since found a place on container and residual lines and continue to evolve, though their pick rates remain slower than optical sorters. “It’s something we use on the very last belt before the compactor, but upstream we have to go with a faster recovery technology,” he said.

“Realistically, the robotic arms, when it comes to picks per minute, you can’t compete with an optic,” Grinold said. “The struggle, though, is volume is king. That’s where you make your money. The last thing we want to do is slow down the system so the robotic arm can perform better. How do we get the robotic arm to meet our current standards?”

“We feel that, if anything, robotic arms are decreasing,” Nella said. “In just the last three years, if we look at a MRF design, it may have five, six, seven robot stations in it. I can tell you, some of the newest MRFs that are performing as well as anything I’ve seen, they have zero robotic arms. We are looking at removing robotic arms and adding optical sorters, doing more with the flexibility of optical sorters with two and three tracks.”

In the future, the panelists agreed the AI linked to robotics could be used more for understanding what is on the belts and making decisions based on that data than relying on them for item capture. Still, Grinold said robotics could perform well on last-chance lines in place of human pickers as hiring and retaining employees has become increasingly difficult.

“I look at this evolution of the technology as rapidly changing,” Nella said. “I know it’s not going fast enough for all of us. But I look in five years, and I say [AI is] going to be directly linked into our controls, and our supervisors are going to be doing less or just confirming recommendations we’re getting from this information that’s coming through.”

Unpacking the black box

As an equipment developer, Rehrmann compared MRF operations to a black box that Tomra and other companies are trying to solve through the data their sorters collect.

In the future, he expressed hope that evolving sorting technology could allow MRFs to function at a maximum level at all times. “This is what drives us,” he said. “This is the vision we have. I can’t promise you that it’s already there now. But I can promise you, if you ask yourself what’s going to happen in five years, I would go as far as to say that we’ll be able to do that.

“You are producing the material,” he continued. “You’re a production company. Compared to other industries, we’re far behind when it comes to those types of systems where you have full control of your production process, making sure when there’s the smallest deviation somewhere in this black box, you notice it and it would ideally correct itself automatically and, if not, you would immediately be notified and aware of that and be sure your plant always runs as close as possible to the optimum level. I think that’s the highest level we as a supplier can give you.”

Rehrmann said data was the answer to improving the technology and that higher numbers of inexpensive sensors installed around a MRF could lead to breakthroughs. He highlighted Tomra Insight, a cloud-based digital solution that provides operators with daily reports on their sorters’ performance as well as timely alerts when something goes wrong.

“In the future, I think we’ll get closer to using this data, making the data more intelligent,” he said. “Just data alone has no value. Someone needs to interpret this data into action for you. And this is what we’re working on.”

The author is associate editor of Recycling Today and can be reached at cvoloschuk@gie.net.