Johanna Reimers, CEO and co-founder of REFIND
It was only 10 days ago when I watched a presentation that mentioned the company REFIND, a Swedish company that provides intelligent sorting and grading solutions using machine vision and learning, with very impressive advances especially regarding batteries and the so-called e-waste. I was particularly impressed by their solutions for batteries, and the simplicity of their desktop-grader sorting solution that needs only a smartphone and a light tunnel to create consistent and smooth illumination for scanning and recognition of objects. I got in contact with Johanna Reimers, CEO and co-founder of REFIND (more about her at the end of her interview), and she was kind enough to answer to my questions. I am really thankful for her swift and thoughtful response, and I am pretty sure that all my readers will enjoy her interview. For my understanding, REFIND sets a new role modeller the waste sector, you will understand why reading below.
How Refind is using the advances of the 4th Industrial Revolution to improve waste management? Which are your specific innovations in software and/or hardware?
We are teaching systems and machines to recognize objects based on their looks, by using cameras and computers with artificial intelligence software. Using machine vision in itself is not an innovation, it has been used for a long time within manufacturing, mainly in order to detect irregularities. For this you need a predictable stream of objects, either in terms of sizes or shapes or the specific look of it. You can use template based matching to understand if the object is within or outside of the specifications. However, in the waste and recycling business, the object stream is very unpredictable – both in terms of what kind of objects that will turn up and also in terms of the condition of the objects.
Because of the unpredictability of the stream, manpower is needed to recognise, sort and register. Manual sorting is expensive and often most of the used electronics will be sorted on a very basic level, shredded and sorted by material sensor equipment into ferrous and non-ferrous fractions. For material recycling this can make sense as long as it is possible to get a high material purity in the streams. However, if the intention is to keep the product value as long as possible, shredding is not an option. Many of the used electronics are disposed of before they have reached end of life. They reach “end-of-first-use” when their owner buys a new device, and could be reused, repurposed or refurbished, but often ends up in a drawer of the consumer. For reuse sorting, manual sorting is necessary in order to check if the products are still functional.
We believe our machines can enable a more cost-efficient and advanced sorting, more similar to human sorting, leading to more reuse being possible.
To link this with the 4th industrial revolution, all of our machines are accessible remotely. The classifier, the “brain” of the machine, used is usually built up on a central location, and then distributed across the installed base of machines around the world. This depends of course of the application, so if a customer needs exclusivity in the classifier knowledge, we can arrange that by limiting the data access.
We believe our machines can enable a more cost-efficient and advanced sorting, more similar to human sorting, leading to more reuse being possible.
We have not invented any new technology, but we have used the existing technology to further develop it for new types of applications: batteries, used electronics and fish species to mention the most important areas.
In terms of your products, which one is your most commercial one? How many time did it take you to develop this product and commercialize it and what are the lessons learnt?
We started with recognizing batteries, which is a niche market, to prove that it works. The Optical Battery Sorter, OBS, is the most commercialized product we have, and it has now reached its fourth version. The first version was built and installed in production in 2012, it took about 2 years to develop. However, by then the company was more of a student project in terms of funding and pace. We are still developing and improving the battery sorter, but not so much in terms of the software or the sorting, but the feeding of objects.
We have learned that recognizing and sorting batteries is not very difficult. As long as you are prepared to collect thousands and thousands of batteries, scan them (take photos of them) and train a classifier to recognise them. Which we are the only ones who have done! The physical sorting is done by using compressed air, shooting batteries into their sorting bins in an exact and reliable way. The most difficult part is to line up batteries, of different sizes, as fast as possible. We have tried vibrating feeders, gravity and ended up at step feeders as the last and best way, when considering manufacturing, shipping and maintenance. We can now line up and classify 15 batteries per second.
The other applications, sorting of mobile phones and used electronics, is still on a research or development level. We know it works, but it is not a commercialized product yet.
The fish species recognition unit is the most exciting product that we are developing right now. It builds on exactly the same foundation as the other applications, the only thing that differs is the outer conditions. This device must withstand saltwater and sunlight and be possible to hose off without damaging any of the electronic components. Again, it is the mechanics and material handling that causes the biggest challenge. The software is easy. But, this is because we have chosen this path. We could have put less limitations to the mechanical handling, the illumination and the vision, and let the software solve it instead. However, that would have required a much higher number of images for training. And since that part is cumbersome, we have decided on this strategy.
Few months ago Refind introduced its latest product, the first reverse vending machine for batteries. What is the problem it resolves? What were the main challenges you faced?
The reverse vending machine creates an incentive for consumers to return their batteries by offering a refund to the consumer. The machine will print a discount coupon based on the number of returned batteries. So, the machine is supposed to increase battery collection, raise awareness of the need for battery collection and also act as marketing for the companies behind the machine. In our case, it was Energizer who branded the machine as well as funded the discount, in a cooperation with Coop Norway, a retail chain.
I believe increase in automation, information and legislation is the key to a more circular economy. Our technology can generate more information, so it is one of the enablers but not the only one.
The main challenges from our perspective was to make the machine user friendly and reliable. Being the first of its kind, the machine needs to be intuitive and welcoming, still without losing the safety measurements of taking care of hazardous material. Another challenge was also the time frame. We had a very hard deadline and it coincided with another delivery, which always is challenging for a small company.
I was impressed by your desktop grader and its lean design. Can you describe how it works and what are its advantages?
It is a light tunnel and a smartphone camera that works as an identification and registration aid. It has no moving parts, hence the desktop name, and could work as bar code scanner – a device that helps you recognize and register different types of items. The items need to be possible to recognize based on their looks – the camera is our only sensor – but we can also add OCR reader or bar code reader functionality to the software if needed.
The desktop grader can be put on top of conveyor belt, if the material handling is to be integrated to the overall solution. In that sense, it is a modular equipment. The smartphone runs an app, Regrab, that allows the user to collect and label images of different objects. When enough images are collected, we train a classifier and connect it to the app. It will run the classification program to recognize and register the different objects.
The advantage with an automatic recognition and registration is that you could link any kind of data to the sorting criteria. For instance, you could integrate it to your warehouse levels of those objects or the Ebay prices and demands for the objects. This means that the sorting criteria can be both static (model name, color, other types of fixed data) or dynamic (supply/demand levels, prices, age). This surpasses the human capacity when it comes to handling information.
Let us know your vision on the future of machine learning and its potential applications for circular economy and waste management
We believe in using technology to minimize and reduce waste. Used electronics and the fish industry are areas where there are enormous amounts of waste and where a reduce would have a great impact. We don’t claim to know all possible applications for the technology, so we aim to make it as general as possible and to focus on the software and the identification units to be as flexible as possible. I believe increase in automation, information and legislation is the key to a more circular economy. Our technology can generate more information, so it is one of the enablers but not the only one.
Johanna Reimers holds a Master of Science in Industrial engineering and management from Chalmers University of Technology. She worked for 10 years as an application consultant and project manager at EVRY, an IT company specialized in business management systems. In 2013, she joined Optisort as project manager, a company later on transformed into Refind Technologies AB, which she co-founded. In 2015, she took over the CEO role in the company.
