Tomra Sorting Recycling has launched Gain, a deep learning-based sorting technology that enhances the performance of its sensor-based sorting machines.
Rather than adding a new machine or line at the plant, the new Gain technology is an add-on option for the company’s Autosort machines, lowering the investment to increase sorting accuracy without sacrificing throughput speed.
By classifying objects based on sensor data, Gain enables the sorting of materials that previously could not be separated with high levels of purity, without compromising the throughput speed of Autosort.
“Deep learning increases the sorting sophistication, effectiveness and flexibility of our market-leading Autosort machines,” says Carlos Manchado Atienza, regional director Americas for Tomra Sorting Recycling.
“Gain technology will allow our sorting machines to adapt to new waste streams, which gives our customers more flexibility to adapt to changing market conditions.”
Deep learning, as a method of artificial intelligence (AI), allows computers to imitate human learning. Humans make associations with what they have seen before and what they are seeing now to identify various materials. Machines are taught to do the same, but much faster.
Tomra machines have deployed AI since the early days of sorting, and the company continually innovates and evolves this technology. New Tomra Gain Technology takes sorting AI to a new level with algorithms rooted in deep learning.
Traditional machine learning requires features engineered by a domain expert, whereas deep learning, a subset of machine learning, does not. It learns from thousands of captured images of specific material types, which should be separated in the sorting task.
Deep learning mimics the activity of large numbers of layers of neurons in the human brain to learn complex tasks. This way, during machine training, Gain learns how to connect the artificial neurons to classify objects.
The first version of the Gain technology released is specifically developed for ejecting PE-silicon cartridges from a polyethylene (PE) stream by using camera information. Silicon remaining in the cartridges means separating those cartridges from the wanted PE material is necessary in order to increase sorted material purity.
In addition to detecting common forms of silicon cartridges, Gain can also detect smaller double-cartridges, mostly used for two-component adhesives, as well as deformed or partly destroyed cartridges. Since Tomra machines employ air jets to separate materials, even clustered cartridges can be sorted out, a task that even today’s fastest picking robot arms struggle with.
The new technology was trained for this task with thousands of images and achieves an overall ejection rate of 99 percent of the cartridges using two systems in a sequence.