Production of plastics is growing at an alarming rate, projected to double in demand by 2050 and already responsible for 4.5% of global greenhouse gas emissions.1 This is triggering governmental actions and research on faster and more efficient ways to recycle plastics.
Because effective recycling is only possible if the mixed plastic waste can be separated into different polymers, factors such as transparency and color might not be the most accurate way to differentiate materials. Hyperspectral imaging is being used as a more accurate part of the sorting process.
Base polymers include PE (Polyethylene), PET (Polyethylene Terephthalate), PS (Polystyrene), PP (Polypropylene), PA
(Polyamide or Nylon), and PVC (Polyvinyl Chloride). These are used to create everyday items such as drinking cups, food trays, bottles, building materials, and textiles.
Material can be sorted before or after flaking. The material
is washed and travels over a conveyor system, passing under hyperspectral-imaging systems that analyze the reflected light and apply chemometric classification models pixel-by-pixel or by object. Results are transferred to a PLC (Programmable Logic Controller) for sorting action.
If precise color sorting is required, hyperspectral sensor
in the VNIR (400 – 1000nm) wavelength range can be utilized. These are far more accurate than conventional RGB color cameras. However, the NIR (900 – 1700 nm) wavelength range is especially well suited to distinguishing the spectral “fingerprints” of polymers. Statistical and machine-learning algorithms are used to recognize and subsequently sort material in real time.
This feasibility study utilized an inno-spec GmbH (now a
Headwall Group company) hyperspectral imaging system consisting of a RedEye 1.7 (950 – 1700 nm) push-broom camera, compact scanning stage, tungsten halogen lighting, and a host computer running perClass Mira Software. Plastic samples of known composition were scanned and a model created to show which samples were made of which polymer. Once properly trained, mixed batches of plastic were passed under the RedEye sensor, detected and false-colored using perClass Mira in real time! In a factory scenario, the positions of each plastic type is sent downstream.
But what about black plastic? Black plastic presents a challenge because of the lack of reflected light across the VNIR and NIR parts of the spectrum. In this case, the BlackEye hyperspectral sensor with a wavelength range of 2900 – 4200 nm (2.9 µm – 4.2 µm) can be used.
___________________________________________________________
1Plastic futures and their CO2 emissions, Nature 612(7939):272-276, https://doi.org/10.1038/s41586-022-05422-5
