Prediction of thermal degradation of biopolymers in biomass under pyrolysis atmosphere by means of machine learning

Biomass is the most widespread among renewable energy sources and offers many advantages. However, the heterogeneous structure of biomass brings many disadvantages. Therefore, characterization of thermal degradation of biopolymeric structures in biomass such as hemicellulose (HC), cellulose (CL), and lignin (LN) is very important for the efficiency of any biomass-based thermal process. On the other hand, the characterization of these biopolymers requires various experimental procedures that consume resources and time. Artificial neural networks (ANN) as a machine learning approach provide a remarkable opportunity to identify patterns in the complex structure of biomass fuels and their thermochemical degradation processes. In this study, a new model was developed for the first time to generate differential thermogravimetric analysis (DTG) curves for HC, CL and LN in biomass using proximate analysis results of raw biomass. DTG curves were evaluated using a ANN model developed with the open-source TensorFlow library in Python software. ANN model performed excellently with R2 values above 0.998. The results show that the newly developed model can estimate the thermal degradation for any temperature, so that biopolymer fractions in the degraded biomass can be calculated immediately, which has not been reported before.

Automated summary

Biomass, as the most widespread renewable energy source, has many advantages but also faces the disadvantage of its heterogeneous structure. Therefore, understanding the thermal degradation of biopolymers in biomass is crucial for the efficiency of biomass-based thermal processes. Artificial neural networks (ANN) offer a promising approach for characterizing the complex structure of biomass. In this study, an ANN model was developed to generate differential thermogravimetric analysis (DTG) curves for biopolymers in biomass, based on proximate analysis results. The ANN model performed excellently, with R2 values above 0.998, and can estimate thermal degradation for any temperature.