Machine learning helps researchers separate compostable plastic waste from regular plastic waste with ‘very high’ accuracy

Disposable plastic is everywhere: Food containers, coffee cups, plastic bags. Some of these plastics, known as compostable plastics, can be engineered to biodegrade under controlled conditions. However, they often look exactly like regular plastic, are improperly recycled and as a result contaminate the plastic waste stream and reduce recycling efficiency. Similarly, recyclable plastic is often confused with compostable plastic, resulting in compost contamination.

Researchers at University College London (UCL) published a paper in Frontiers in Sustainability where they used machine learning to automatically sort the different categories decomposable And biodegradable plastic and distinguish them from conventional plastics.

“The accuracy “It is very high and allows the technique to be potentially used in industrial composting and recycling facilities in the future,” said Professor Mark Miodownik, the study’s corresponding author.

Up to perfect accuracy

The researchers worked with different resins ranging in size from 50mm x 50mm and 5mm x 5mm. Normally plastic Samples include PP and PET, commonly used for food box and drinking water bottles, as well as LDPE, are used—among other things—for plastic bag and packing. Compostable plastic samples include PLA and PBAT, used for cup lids, tea bags and magazine wrappers; as well as palm leaves and sugarcane, both biomass-derived materials used for packaging. The samples are divided into a training set, which is used to build classification models; and a test suite, used to check accuracy.

The results show a high success rate: The model achieves perfect accuracy for all materials when the samples are measured above 10 mm x 10 mm. However, for materials made from sugar cane or palm leaves with dimensions of 10 mm x 10 mm or less, the misclassification rates were 20% and 40%, respectively.

Looking at the 5mm x 5mm pieces, some materials were identified more reliably than others: For LDPE and PBAT pieces, the misclassification rate was 20%; and both biomass-derived materials were misidentified at 60% (cane) and 80% (palm leaves). However, this model can identify PLA, PP and PET flakes without error, regardless of sample measurements.

Out of sight

“Currently, most compostable plastics are considered as contaminants in the conventional plastic recycling process, reducing their value. Trommel sorting and density are applied to fecal screening. mix and reduce the presence of other materials.However, the level of contamination from the current screening process is unacceptably high,” explains Miodownik. “The advantages of compostable packaging are realized only when they are industrially composted and do not enter the environment or contaminate other waste streams or soil.”

To improve accuracy, a team of scientists including Nutcha Teneepanichskul, Professors Helen Hailes and Miodownik from UCL’s Plastics Waste Innovation Center tested conventional, compostable and plastic resins. various biodegradation, by using hyperspectral image (HSI) to develop the classification model. HSI is an imaging technique that detects invisible chemical signatures of various materials while scanning them, generating a pixel-by-pixel chemical description of a sample. AI models were used to interpret these descriptions and perform material recognition.

Poor plastic management in high levels of industrial composting and recycling makes a reliable sorting mechanism necessary. “Currently, the recognition rate is too low for industrial-scale deployment,” pointed out Miodonik. However, “we can and will improve it because automatic sorting is the key technology to make biodegradable products. plastic a sustainable alternative to recycling.”