I sit down to talk with Stuart Bashford, Head of R&D at Bühler, to discuss the changing landscape of optical sorting, and how Artificial Intelligence is paving the way in the future for improved yield and improved profits in food processing.
Stuart: Thanks. I’ve worked for Bühler for now around 11 years, I’ve been the Digital Office for over 7 years, working on setting the vision and strategy for Bühler Group’s digital transformation. Now, I look after the Research & Development activities for the Optical Sorting business. I’m a technology guy, so it is a dream come true to work with this technology every day.
Stuart: In optical sorting, we’re talking about a very specific area of AI, we’re not talking about generative AI or ChatGPT style AI, but we are talking about AI in the application of image analysis. This area involves deep learning and Convolutional Neural Networks, where an algorithm is trained to identify accept and reject at a level that humans cannot comprehend, creating a far higher level of accuracy and yield.
Stuart: The simple answer to that is with this new technology, we can make you more productive, and we can make you more money. Every organization has a challenging business environment: high raw material prices, high energy costs and supply chain resiliency issues. At times like this, it is important to make existing systems as productive as they can be. Tests of our new upcoming AI sorter have shown a yield increase of 4-5%. That means processors will be making more good product out of the same amount of incoming product. This helps hit sustainability targets, and ultimately, helps to generate more profit.
Stuart: The key thing for our customers is to maximize yield, which includes maximizing uptime. It does not matter to our customers what the buzzword is, whether AI, CNN or digitalization, so long as they are getting a tangible benefit. The benefits to processors from this particular technology include reduced energy, increased yield, and increased quality.
Stuart: Convolutional Neural Networks present a massive step forward in the way a sorter identifies an object. In simple terms, we capture data, and we go through a process of labelling the data and training the algorithm of the sorter with that data. This involves a vast number of labelled images. During the sorting process, the machine scans every single grain, or individual product, and identifies whether it is a defect using the reference library of labelled images it has been trained on. This translates to the most accurate sort ever, and the highest yield. All this technology is already embedded in the machine, so the customer does not have to be aware of it.
Let’s use a practical example. The first application we will be looking at with this machine is gluten-free oats, which are often mixed at harvest with gluten products like barley and rye. Our new AI sorter will look at the grains individually, refer to its training (the labelled grains), and look for commonalities of what makes up an oat vs what makes up a barley. It may come up with 2000 commonalities of what makes an oat vs what makes a barley. This is done in a fraction of a second, and provides a higher level of defect identification, and in this case, the highest purity of gluten-free oats.
Stuart: Yes. Our new sorter, the SORTEX AI700, will be coming soon, complete with all the benefits we’ve discussed. Our first application will be oats, and afterwards, we will train it on other applications throughout 2025. This is really a step forward in optical sorting technology, and why we are saying it’s the ‘new age of optical sorting’.
Stuart: It’s a pleasure.
