(154ab) Real-Time Plastic Waste Recognition with Mid-IR Standoff Detection and Advanced Machine Learning | AIChE

(154ab) Real-Time Plastic Waste Recognition with Mid-IR Standoff Detection and Advanced Machine Learning


Zhao, Y. - Presenter, University at Buffalo (SUNY)
Chakraborty, P., University at Buffalo
Meng, Z., University at Buffalo
Thundat, T., University at Buffalo (SUNY)
Plastic waste identification is vital for recycling and mitigating the effects of global warming. Since plastics do not degrade easily with time, the discarded plastics pose a grave environmental problem. Only less than 10% of plastics is recycled in the US. Also, sorting plastic waste in MRF (Materials recovery facility) is often complicated by the presence of additives and contaminants. At present, sorting of plastic waste is carried out manually which is time consuming, costly and labor intensive. Therefore, developing a sorting technique with high selectivity, sensitivity, and minimal false identification is crucial for increasing plastic recycling rates. We have been developing a sensor platform capable of improving the accuracy of automated sorting of various plastics by using molecular identification using mid-infrared (MIR) spectroscopy. This technique of standoff photothermal detection utilizes mid-infrared photo-thermal spectroscopy for characterizing plastic samples. Since MIR is free of overtones, this region is known as the molecular fingerprint region. In this technique, the plastic surface is illuminated with a pulsed laser beam from a tunable quantum cascade laser (QCL). The photothermal detection is accomplished by collecting the scattered light from the plastic surface which contains the molecular vibration spectrum of the sample as well as the contaminants. A silicon bi-material microcantilever, fabricated by depositing a thin metal film on one of its sides is used as a sensitive uncooled infrared detector. Monitoring the bi-material cantilever bending as a function of illumination wavelength provides the IR absorption spectrum of the plastic material. Our method accurately classifies MIR spectra of plastics with a 100% success rate, even in the presence of additives and contaminant residues, using advanced machine learning techniques. This promising approach enables the rapid and efficient characterization of plastic waste in industrial applications where high throughput is essential. It will help in enhancing the quality of plastic sorting process at materials recovery facilities and other industries by making the process inexpensive and time efficient.