Bio-diesel quality characterization towards bio-jet fuel production using infrared spectroscopy and an interpretable...

Area of Science:

• Chemical Engineering
• Spectroscopy
• Machine Learning

Background:

• Efficient use of biomass requires advanced modeling for sustainable resource management.
• Biodiesel-based bio-jet fuel offers a sustainable aviation alternative but faces production challenges due to complex composition and quality parameter characterization.

Purpose of the Study:

• To develop an interpretable machine learning framework using infrared spectral descriptors for predicting and understanding biodiesel quality attributes.
• To enhance the rapid and accurate characterization of key quality parameters in biodiesel production.

Main Methods:

• Evaluated seven machine learning models, focusing on an ensemble model for superior performance.
• Utilized infrared spectral descriptors and Shapley value analysis to identify key spectral features and their relationship to biodiesel properties.
• Developed a user-friendly interface for real-time fuel quality monitoring and process adjustment.

Main Results:

• The ensemble model achieved high accuracy with a mean error of 6.83%, RMSE of 1.43, and R² of 0.95.
• The descriptor-driven approach improved predictive accuracy and model robustness.
• Identified specific infrared spectral drivers, including olefinic C-H and CH3 bending, with varying importance across different biodiesel quality indicators.

Conclusions:

• The developed AI framework effectively combines spectral analysis with machine learning for enhanced bioenergy utilization.
• This approach supports the scalable deployment of biomass-derived aviation fuels and sustainable resource management.
• The interpretable nature of the framework provides mechanistic insights into spectral-property relationships for industrial applications.