How AI robots could revolutionise MRF operations

Written by: Peter Raschio | Published:
The MAX-AI machine

Recycling processes must improve, now more than ever.

Recovery and diversion goals are rising, and so are quality requirements. Processors face the challenge of recovering more recyclable commodities from waste while simultaneously producing near-perfect bales to send to market.

It is a challenge that has left many scrambling. However, developments in MRF technology have not kept pace with this escalating demand. At least until the recent rise of artificial intelligence (AI). One company believes it has found a solution.

Green Recycling has announced it will be the first company in the UK to invest in AI robotic sorting to recover recyclable commodities. The kit, known as the Max-AI AQC (for autonomous quality control), comes from US-based Bulk Handling Systems and will be operational later this year at Green Recycling’s commingled dry recyclables MRF in Maldon, Essex.

Introducing Max-AI

Max-AI technology is akin to facial recognition, but for recyclables. Using a process known as deep learning, Max’s neural networks identify recyclables in a similar way to a person. A neural network is a computer program that learns tasks through examples, rather than acting with specific instructions. A camera enables Max to see materials on a conveyor belt, and the neural networks use the visual information provided by the camera to make proper identification decisions.

Whereas a person performing a similar task uses eyes to see, a brain to think and hands to grab, Max employs a camera to see, neural networks to think and a robot to sort. Similar in approach, but superior in ability, Max is able to identify and make prioritisation decisions in near real time and move at a consistently fast pace over multiple shifts.

Neural network

Recent advances in computer hardware and artificial intelligence have made possible the deep learning neural network that is the central nervous system of Max-AI technology. Through deep learning, Max has been trained with millions of images of different materials that have already been identified by humans. Time and time again, Max discovers the best path to take using its artificial brain to reach the correct answer.

Through millions of iterations, the neural networks are built to enable Max to identify new images. Once in place, Max can correctly classify never-before-seen objects. Max has seen hundreds of thousands of similar items, so when a new item is seen, the robot has already worked out the correct thought process to immediately recognise it.

Most recycling equipment has relied on deterministic detection techniques, or an ‘if, then’ approach. If the material fits in parameter X, then it goes in direction A. That includes sorting by size, material density, 2D or 3D and even material type. An optical sorter identifies PET because it can recognise that the NIR light signature from the material is in a certain identifying spectrum range. A programmer has told the device that “if X is greater than Y then eject the object”.

Max-AI technology takes a probabilistic approach – it is not looking at any one thing, but knows what the material is because it has been trained, like a person. When a person looks at a PET bottle, they aren’t sure exactly what features they are looking for, instead they know just the type when they see it. They have seen many others that looks similar, and have learned over time to make the correct decision.

Similar to the way a person isn’t constrained by material type, neither is Max constrained. For example, Max can do more than just pick out PET from other materials; it can pick out PET trays from PET bottles, while at the same time grabbing aluminium, residue and mixed paper in the same location.

Instead of being constrained by only colour analysis, Max can be trained to leave brown paper while removing three-dimensional cardboard. If the operator wanted, Max could only pull Coke and Pepsi bottles. These are just a few examples of how neural network detection can accomplish more than previously available detection technologies.

Max goes Green

According to Green Recycling, the company has long been aware of robotic systems and anticipated the industry eventually moving in this direction. With Max, the AI-based technology seemed different from other robotic sorters, all of which employed expensive sensor-based detection.

The company’s introduction to Max began with a conversation at RWM 2017. Finding no other companies with similar capabilities, Green Recycling decided to visit US-based recycling company Penn Waste to personally see the Max-AI AQC in action. They were impressed with the speed and performance of the equipment, and grew confident after finding the Penn Waste team pleased with their new robotic sorting equipment.

Like all industry leaders, Green Recycling is always on the look-out for the best equipment to meet its operational goals and provide best possible service to its customers. That means finding the highest and best use for the recoverables that are sent to the company’s Essex plant while processing them in the most cost-effective manner.

After the Penn Waste visit, Green Recycling executives were confident that the Max-AI AQC was the next step towards automating their operation. They already have equipment in place to open bags, collect plastic film and OCC, and segregate material using screen and air separation technologies.

The Max-AI AQC boosts the performance of this existing system, working on the recovery line to capture card, newspaper and pamphlets, high-density polyethylene, PET bottles and even wood. The single AQC unit will perform tasks that were previously done by two employees, who will be transitioned to other roles. Green Recycling will also be adding a technology-focused position.

People vs robots

Green Recycling have confirmed that the robotic sorter will not replace any of its employees but instead increase operational consistency. Staffing a MRF is an ever-present challenge. The commodity markets are tight and competition is high, with some operators adding staff and slowing down lines to meet quality. It is crucial that MRF operators maximise throughput and uptime as much as possible to stay ahead.

Manual sorting QC positions can be challenging, monotonous and dirty. For these reasons, operators experience very high employee turnover rates. Significant resources are poured into recruiting, onboarding, training and safety – only to do it all over again, and all too often.

MRF operators also deal with employees showing up late or on occasion not at all, situations that can affect the entire operation. Due to the nature of the job, performance is inconsistent. It can be difficult to stare at a moving conveyor belt all day to begin with, much less having to focus on the items, make smart decisions and sort at consistently high pick rates.

Using people for this type of work is not ideal, but it has been previously necessary to create marketable products. The Max-AI AQC allows Green Recycling to transition employees to other roles, such as maintenance, operations or even a technology position to help the MRF maintain performance.

Future of AI in MRFs

A successful deployment of this technology will lead Green Recycling to further automate its process. The long-term vision is an intelligent, fully autonomous system that can run through multiple shifts. Labour remains one of the largest drivers of operational expenses and prohibits many systems from running multiple shifts – total automation could essentially double a system’s capacity.

Max-AI technology allows MRFs to run longer with lower operational expenses, produce more products with increased purity, capture accurate data for reporting and adapt over time to the changing material mix without major capital expenses. Green Recycling is the first to adopt artificially intelligent recycling equipment in the UK. But from the sounds of it, it surely won’t be the last.

Peter Raschio, marketing manager at Bulk Handling Systems


This material is protected by MA Business Ltd copyright.
See Terms and Conditions.

Comments
Name
 
Email
 
Comments
 

Please view our Terms and Conditions before leaving a comment.