Monday, May 2, 2011
Grand Ballroom C-D, 2nd fl (Sheraton Seattle)
Compositional analysis of the feedstocks used in the production of second-generation bioethanol could be performed with infrared spectroscopy, which provides a valid alternative to commonly used wet chemistry methods. In this work, near-infrared (NIR) acquired in diffuse reflectance mode and mid-infrared (MIR) spectroscopy in attenuated total reflectance mode, coupled with multivariate calibration methods were used to estimate the composition of different biomass residues. Prediction models were constructed for ammonia-pretreated sugarcane bagasse. Other models included cassava leaves and cassava stems as well. Moisture, cellulose, hemicellulose, lignin, extractives and ash contents were measured. For each component, several hundreds of models were built based on the calibration samples, as various parameters were used (type of data preprocessing, method for wavelength selection, regression method). All the models were subjected to cross-validation, which allowed to retaining the ones that showed small root mean square error of cross-validation. Based on these models, the composition of the prediction samples was estimated. The accuracy of these predictions was also assessed. The bagasse models yielded relatively good results for hemicellulose, extractive and ash concentrations, while their predictive power for cellulose and lignin were lower. The results obtained with the bagasse-cassava models are weaker, which is most probably due to the strong heterogeneity of the samples under study. If NIR-MIR can be used to estimate the hemicellulose content of an ammonia-treated bagasse, then the sugar yield by enzymatic hydrolysis can be predicted.
Keywords: NIR, MIR, lignocellulosic materials.