The diversity of plastic material chemical composition and structure complicates the ability to identify and separate waste types to the degree necessary for effective recycling. Because plastic typically must be at least 96% pure by polymer type to be recycled by current industrial standards, available methods are not always sufficiently accurate to keep pace with the immense amount of plastic waste that must be sorted and processed. A new solution based on hyperspectral imaging and machine learning offers scope for the rapid, precise identification and separation of 12 different varieties of plastic on recycling facility conveyor belts.

The rate of recycling will increase with a system capable of distinguishing 12 types of plastic. Source: VestforbrændingThe rate of recycling will increase with a system capable of distinguishing 12 types of plastic. Source: VestforbrændingCurrent separation methods involve near-infrared technology or density tests such as flotation separation, which can separate some plastic fractions such as polypropylene and polyethylene terephthalate but are not as accurate or versatile as the hyperspectral system. The new system includes a hyperspectral camera and spectrograph with a spectral resolution of 8.3 nm in the range of 955 nm to 1700 nm and installed inline on a conveyor belt system through which the plastic samples are transported. An unsupervised machine learning model was developed by the researchers in Denmark to distinguish between different plastic types based on the spectral information recorded and transferred to a computer as the plastics are scanned on the conveyor belt.

The technology has been demonstrated at pilot scale and will be implemented at commercial facilities in spring 2022. Researchers from Aarhus University and plastic recycling companies Vestforbrænding, Dansk Affaldsminimering Aps and PLASTIX contributed to this development, which is described in Vibrational Spectroscopy.

To contact the author of this article, email