Fast prediction of the combustion properties of biomass pellets using hyperspectral imaging

Jetsada Posom,Kanvisit Maraphum

DOI: https://doi.org/10.1016/j.biombioe.2024.107134

IF: 5.774

2024-03-02

Biomass and Bioenergy

Abstract:Biomass pellets are ideal as a source of energy because of their abundance and high energy density. This study describes the prediction of combustion properties of biomass pellets using hyperspectral near infrared (HSI) spectroscopy. The spectra were extracted from a hyperdata cube from the individual pellets. Spectra wavelength range of 860–1760 nm was used to build the calibration model with either raw spectra or spectra pre-treated by a standard normal variate (SNV), first derivative (D1) and second derivative (D2). The prediction models were established using partial least square (PLS) and deep learning (DL) regression. The calibration model which developed from DL regression has the potential to evaluate the combustion properties, which provided coefficient of determination of prediction (R 2 p) and root mean square error of prediction (RMSEp) of burnout time which were 0.88 and 2.01 min, ignition index were 0.76 and 0.005 %/min 3 , combustion index were 0.91 and 3.14 × 10 −7 % 2 /(°C 3 min 2 ), burnout index were 0.82 and 1.75 × 10 −4 %/min 3 , ignition temperature were 0.70 and 9.89 °C, peak temperature were 0.85 and 6.01 and the final temperature were 0.73 and 32.53 °C, respectively. Most models are suitable for screening. The prediction of combustion parameters using colour distribution mapping can be used to track, manage, and monitor pellet quality and is readily applicable to screening applications.

energy & fuels,biotechnology & applied microbiology,agricultural engineering