Source discrimination by classical characterization methods, FTIR and statistical analysis – a prerequisite for thermochemical conversion of agriculture biomass residues by torrefaction and pyrolysis

The possibility for rapid and facile biomass characterization and classification, including prediction, is strongly beneficial before suitable thermal conversion. Here we explored the hypothesis that classical destructive methods (proximate, ultimate and compositional analysis) and infrared spectroscopy can discriminate, classify and predict the type of biomass resources, focusing on five biological classes of agriculture residues (stalks, hulls, shells, pits, seed cakes). Proximate and ultimate analysis revealed clustering in tridimensional representation and Van Krevelen diagram. Despite showing distinct spider plot patterns for each class, compositional analysis was less useful for direct discrimination. Instead, sample classes were correctly discriminated by principal component analysis and hierarchical clustering. We proposed FTIR as a facile, rapid alternative to classical, destructive methods, which differentiates the biomass residues mainly by the vibration bands of fatty acids and carbohydrates and less by those of lignin. The fine k-nearest neighbors machine learning algorithm applied to FTIR data correctly predicted the class for 96 % of the samples, with 0.94595 validation accuracy, much better than for the destructive methods. Particular features are preserved after torrefaction and pyrolysis, classification of liquid products being the same as the biomass sources. Stepwise pyrolysis can flatten the differences in biomass sources, generating products with more uniform composition.