Compostable plastics and recyclable plastics often look the same, leading to contamination of the waste streams and lowered efficiency in recycling. University College London (UCL) researchers recently published a paper in Frontiers in Sustainability detailing their use of machine learning to find a solution: creating an automated system that can accurately sort compostable and biodegradable plastics from conventional plastics. The researchers used different types of plastic for their training and testing sets, ranging from 50mm by 50mm to 5mm by 5mm in size. They found that their model accurately identified PLA, PP, and PET pieces regardless of size and had a misclassification rate of up to 40% or less on smaller samples.
The UCL team was led by Professor Mark Miodownik, who stated that the accuracy of the model could potentially be used in industrial facilities in the future. Currently, compostable plastics have to be mixed with conventional plastics, leading to contamination and reducing the potential value of recyclable products. In order to overcome this, the researchers used hyperspectral imaging (HSI) to produce a pixel-by-pixel chemical description of the samples. Artificial intelligence (AI) models were then used to interpret these descriptions and make a material identification.
UCL’s Plastic Waste Innovation Hub has been active in making the transition to sustainable packaging more efficient. The Hub is a research program designed to lead the way towards reducing plastic waste with innovative policy and technological solutions. Prof Helen Hailes and Nutcha Teneepanichskul are key researchers in the Waste Innovation Hub and worked alongside Professor Miodownik on the study. Demand for plastic packages is ever increasing and the Hub is working towards recycling and composting solutions that could be implemented immediately in a cost-effective manner. The Hub is set to revolutionize the need to reduce the amount of plastic waste created through the use of the machine learning system.
