Machine learning-based prediction of biomass pyrolysis kinetics: integrating mechanistic modeling and compositional...

Area of Science:

• Biomass Pyrolysis
• Thermochemical Conversion
• Renewable Energy

Background:

• Accurate kinetic and thermodynamic parameters are crucial for understanding biomass pyrolysis.
• Sapodilla leaves were selected as a representative lignocellulosic feedstock.
• Both experimental and machine learning (ML) approaches were employed.

Purpose of the Study:

• To determine kinetic and thermodynamic parameters of sapodilla leaf pyrolysis.
• To optimize renewable thermochemical conversion processes.
• To compare mechanistic fitting with ML-based prediction of kinetic parameters.

Main Methods:

• Thermogravimetric experiments were conducted at multiple heating rates.
• The Coats-Redfern method was used for mechanistic interpretation.
• A machine learning framework was developed to predict kinetic parameters using compositional data.

Main Results:

• Pyrolysis kinetics varied with temperature regime, showing diffusion/reaction-order at low temperatures and nucleation control at high temperatures.
• Thermodynamic analysis indicated an endothermic, non-spontaneous process with negative entropy change.
• ML models demonstrated moderate predictive capability for kinetic parameters, identifying key compositional descriptors.

Conclusions:

• Combined mechanistic and ML approaches offer a proof-of-concept for understanding biomass pyrolysis.
• The study highlights limitations in current ML predictive robustness and generalizability for kinetic parameters.
• Further research is needed to enhance predictive accuracy and broaden applicability.