Inline classification of polymer films using Machine learning methods.

Area of Science:

• Materials Science
• Computer Science
• Environmental Science

Background:

• Plastic packaging film waste (PPFW) poses a significant recycling challenge in Austria, with 150,000 tons annually being thermally recovered.
• Current methods cannot distinguish between mechanically recyclable monomaterial and non-recyclable multimaterial films, hindering effective waste management.

Purpose of the Study:

• To develop and validate machine learning models for the inline classification of PPFW into monolayer and multilayer categories.
• To enhance the sortability of plastic waste, thereby increasing recycling rates and reducing reliance on thermal recovery.

Main Methods:

• Utilized spectral fingerprinting in transflection for film analysis.
• Applied machine learning models for classification.
• Employed feature selection techniques like Principal Component Analysis (PCA) and Minimum Redundancy Maximum Relevance (MRMR) F-Tests to identify optimal spectral ranges.

Main Results:

• Achieved a prediction accuracy of 85% on unseen plastic packaging film waste specimens.
• Feature selection reduced model complexity and prediction time without compromising accuracy.
• Demonstrated minimal prediction latency, confirming inline applicability.

Conclusions:

• Machine learning models based on spectral data are effective for inline sorting of plastic packaging film waste.
• The developed models can accurately differentiate between monolayer and multilayer films, paving the way for improved recycling processes.
• This approach offers a viable solution to increase the recycling rate of PPFW.