Interpretable Machine Learning Methods for Monitoring Polymer Degradation in Extrusion of Polylactic Acid.

Area of Science:

• Polymer Processing and Materials Science
• Real-time Monitoring and Quality Control
• Machine Learning in Industrial Applications

Background:

• Extrusion-induced degradation significantly impacts poly L-lactic acid (PLA) properties.
• Current soft sensor approaches often lack interpretability, limiting industrial adoption.
• Accurate real-time monitoring is crucial for quality control in polymer processing.

Purpose of the Study:

• To develop an interpretable soft sensor for real-time monitoring of PLA degradation during extrusion.
• To compare the performance of Recursive Feature Elimination (RFE) with other regression methods.
• To identify key process parameters influencing PLA molecular weight and mechanical properties.

Main Methods:

• Utilized in-process sensor data (temperature, pressure, NIR spectra) and machine settings.
• Implemented and compared Recursive Feature Elimination (RFE) with Partial Least Squares (PLS), Principal Component Regression (PCR), and Random Forest (RF).
• Developed an RFE-RF algorithm for predicting molecular weight and yield stress.

Main Results:

• The RFE-RF algorithm achieved high accuracy in predicting molecular weight and yield stress for medical-grade PLA under controlled conditions.
• RFE-RF outperformed other methods in simplicity, interpretability, and accuracy across various process conditions and machine setups.
• Extruder exit temperature was identified as the primary predictor of polymer molecular weight degradation.

Conclusions:

• RFE-based soft sensors offer a promising, interpretable, and computationally efficient solution for quality control in processing thermally sensitive polymers like PLA.
• This study demonstrates the first real-time monitoring of molecular weight degradation during PLA processing across different machine settings.
• Process insights revealed that pressure and temperature at later extrusion stages, not molecular weight change, primarily affect mechanical properties.