Infrared spectroscopy and machine learning for post-consumer plastics recycling.

Area of Science:

• Materials Science
• Environmental Science
• Data Science

Background:

• Plastic pollution is a major environmental issue, necessitating effective recycling strategies.
• Recycled plastics must meet quality standards comparable to virgin materials, requiring accurate classification.
• Infrared (IR) spectroscopy is a key technique for polymer identification, but spectral similarities pose classification challenges.

Purpose of the Study:

• To develop and validate an automated Machine Learning (ML) classifier for identifying post-consumer plastic types using IR spectroscopy.
• To overcome spectral similarities that hinder accurate plastic classification.
• To enhance the reliability of recycled polymer identification for industrial applications.

Main Methods:

• Integration of Infrared (IR) spectroscopy with advanced Machine Learning (ML) algorithms.
• Development of an automated classifier trained on a dedicated IR spectral database.
• Comparative analysis of traditional ML algorithms (Random Forest) and Convolutional Neural Network (CNN) architectures using the Quasar platform.

Main Results:

• Both CNN and Random Forest (RF) algorithms achieved high accuracy in plastic classification (1.000 and 0.998, respectively).
• The integration of deep learning with IR data significantly improved the accuracy and reliability of plastic recognition.
• The developed ML-based classifier demonstrated robust performance in identifying four prevalent plastic types: HDPE-B, HDPE-P, LDPE, and PP.

Conclusions:

• The combination of IR spectroscopy and ML offers a powerful solution for accurate plastic material classification.
• This approach supports the industrial transition towards high-quality recycled polymers by ensuring material consistency.
• Advanced ML, particularly deep learning, enhances the effectiveness of spectroscopic methods for environmental applications.