This Is AuburnElectronic Theses and Dissertations

Prediction of Heating and Ignition Properties of Biomass Dusts Using Near Infrared Spectroscopy

Date

2014-07-11

Author

Dhiman, Jaskaran

Type of Degree

thesis

Department

Biosystems Engineering

Abstract

Dusts (i.e. particles of size less than 500 μm) are generated during handling and processing of biomass feedstock. Similar to damages that have been reported from ignition of dusts obtained from industries, ignited biomass dust may potentially cause fire and explosion in biorefinery plants that can result in human fatalities, serious injuries and substantial monetary loss. Control measures to prevent the heating and ignition of biomass dusts will play a critical role in development of safety guidelines and standards for bio-based industries. The research aims at quantifying and predicting (using NIR spectroscopy) the heating and ignition properties of dust from ten biomass feedstocks. Three different types of coals were also used for comparison purposes. The range of values obtained for these properties were 240°C-335°C (minimum hot surface ignition temperature, MIT), 266°C-448°C (temperature of onset of rapid volatilization, TORV), 304°C-485°C (temperature of maximum rate of mass loss, TMML), 242°C-423°C (oxidation temperature, TOXY), 206°C-249°C (temperature of onset of rapid exothermic reaction, TRE) and 354°C-429°C (maximum temperature reached during exothermic reaction, TME). Coefficient of determination (R2) values for internal validation of prediction models developed using PCA on raw NIR spectral data for MIT, TORV, TMML, TOXY, TRE and TME were 0.994, 0.984, 0.963, 0.737, 0.931 and 0.901 respectively, whereas, first derivative NIR spectral data yielded R2 (calibration) for these properties as 0.976, 0.964, 0.943, 0.798, 0.923 and 0.895 respectively. Four different biomass dusts (eucalyptus, pine, sweetgum and switchgrass) were used to validate the prediction models externally. Coefficient of determination (R2) values for all the models was obtained less than 0.28. Poor performance of models under external validation was attributed to small sample sizes of the feedstocks that were used during building of prediction models.