Last December (2020), Bluewater Recycling Association (BRA) broke new ground in Ontario residential recycling processing by installing six new Artificial Intelligence (A.I.)-enabled robotic arms in their Material Recovery Facility (MRF). The investment was driven by the need to reduce reliance on hard-to-retain manual sorters and deal with a material mix that is increasingly complex.
Artificial Intelligence uses image matching to sort materials
The A.I. integrated robots sort by identifying distinguishing features in the same way as the human eye. A camera takes a picture of items on the belt, which are then uploaded to a cloud. The image is instantaneously matched with information in the database to identify the items. The arm is then triggered to grab the items it is programmed to pick. Generally, images are matched on the basis of shape and colour. The system continually improves learning from operating experience over time to assure maximum recognition efficiency.
Challenges in manual sorting
BRA, a single stream facility which processes roughly 20,000 tonnes per year, had been operating 10-hour shifts daily with a complement of 12 manual sorters (but needed 16 sorters). While manual sorters can perform up to 30 to 40 picks per minute (ppm), in cases where the material mix is complex, the pick rate is lower, often in the range of 20 to 25 ppm. In terms of how these pick rates affect performance, an efficiency and effectiveness study carried out in the BRA MRF showed that manual sorters in one station were only successful in capturing 48% of the aluminum cans available to be sorted. In another case, where there was a wide range of materials on the belt, the capture rates of the manual sorters fell as low as four to six percent.
In addition to lower capture rates, manual sorters are hard to find and can be harder to retain. The turnover rate is high, meaning there is a need for constant recruiting and training. In recent years, robotics with A.I. capability have been presented as a possible solution to these ongoing challenges.
A.I.-enabled machines reduce staffing needs and increases capture rates
Unable to secure and maintain the full staffing complement needed year over year, BRA bought and installed six Machinex Samurai® recycling sorting robots in late 2020 for $1.9 million. The machines were installed on weekends to minimize disruption and only an occasional shutdown was required.
The A.I.-enabled robots took about three months to reach optimal performance. Now fully operational, the six units replaced 10 sorting positions and are capable of carrying out the work of up to 20 manual sorters. With this new equipment, BRA was able to expand its operations to two shifts with existing staff. The elimination of the overtime yielded substantial savings annually. The new installation also reduced the struggle to find and keep manual sorters from 16 to six per shift. The investment has allowed for the increased capture of high value materials like aluminum, PET, and HDPE from the residue line. It has also allowed for the improvement of material quality which has resulted in increased revenue. With the combined savings and increased revenues, BRA is forecasting a payback on their investment of 2.5 years.
While much faster than humans, image matching has limitations
The image matching software is not always effective. Some containers have the same colour and shape, but their resin types vary by brand. For example, yogurt containers may contain either PP or HDPE. Unable to distinguish one resin type from another, the robots will sort all the yogurt containers into the same bunker. Additionally, the system cannot identify an item if it is sitting atop another (e.g., water bottle on top of a piece of paper), therefore they’ll just leave both items on the belt. The robots do not recognize items unless trained to do so. Robots will not attempt to pick anything it has not been programmed to pick, so unanticipated items, like a bowling ball, diapers, frying pans, and polystyrene insulation will be left on the belts and will move through the entire facility sometimes causing damage or interfering with the picking of targeted materials. For this reason, the robotic A.I. arms are not ideal for the pre-sort process given the complexity of the mix of items presented in the early part of the sortation process.
While the A.I.-enabled robot can reach pick rates almost double that of a human (70 ppm vs. 30 to 40ppm), a person is still more adaptable as they can better interpret what they see (e.g., they easily recognize the water bottle from the piece of paper it is sitting on). This means manual sorters are still needed for the pre-sort.
Integrating optical sorting technology with A.I. to address limitations
In 2021, BRA plans to incorporate the use of Machinex’s MACH Hyspec®, a high-speed, short wave infra-red (SWIR) hyperspectral detection system, that can distinguish material type. The MACH Hyspec® is a standard feature of optical sorters, but its use in the A.I.-enabled robot is in its infancy. The rollout of this feature in BRA’s facility was delayed by months due to challenges faced by the manufacturer in integrating the technologies to work together. Once up and running, the robots will be able to detect the difference between various containers based on their resin type, despite similarities in colour and shape. The MACH Hyspec® is currently in a learning phase and is expected to be fully operational by July 2021.
A.I. shows promise to improve effectiveness and efficiency at MRFs
A.I.-enabled robots allow for consistency and reliability and help address recruiting and staff turnover challenges. A.I.-enabled robots do not require uniforms, PPE or call in sick. Their capabilities are evolving, but they will not stop the conveyor if something is shredding it, or if it has caught on fire. BRA president, Francis Veilleux explains they are not at a point where they’ll replace the need for people entirely. The inbound material mix continues to be plagued by high levels of contamination making it increasingly complex and highly variable. People will be needed for the foreseeable future for many tasks in the MRF.
Watch this space for more learnings in the coming months. For more information on technology-based solutions from a wide range of vendors, check out CIF’s recent Next Gen Tech report.