Convolutional neural networks enhance pyrolysis gas chromatography mass spectrometry identification of coated papers

In the evolving paper industry, the accurate identification of coated paper components is increasingly essential for sustainability and recycling efforts. This study employed pyrolysis gas chromatography-mass spectrometry (Py-GC-MS) to examine eight different coated paper types. A key finding was the minimal interference of the paper substrate with the pyrolysis products of the coatings, ensuring reliable analysis. The study initially utilized machine learning techniques, specifically random forests and support vector machines, to analyze the pyrolysis data, achieving notable accuracy. However, these traditional methods necessitated complex preprocessing steps. In response, 1D convolutional neural network (1D-CNN) were explored as a more streamlined approach. The 1D-CNN directly processed the extracted ion chromatograms, bypassing conventional intermediate steps, and achieved a predictive accuracy of 95.2% in identifying various coated paper compositions. Additionally, the study highlighted the importance of selecting an optimal pyrolysis temperature for effective feature extraction in machine learning models. Specific markers for coated papers, including polyethylene (PE), polypropylene (PP), polyethylene terephthalate (PET), polybutylene succinate (PBS), polylactic acid (PLA), polybutylene adipate terephthalate (PBAT), polyhydroxy alkanoate (PHA), and waterborne polyacrylates (WP), were identified. This research presents an innovative approach in coated paper identification, blending Py-GC-MS with advanced machine learning, setting a foundation for further research in product integrity and environmental impact.